diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t5/task.toml b/harbor/tasks/v3_depth5_delayed_net_revenue_t5/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..15ce2242f64efd1e26d575d03dd654c918e82ffc --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t5/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_delayed_net_revenue_t5" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t5/tests/check_output_shape.py b/harbor/tasks/v3_depth5_delayed_net_revenue_t5/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t5/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t5/tests/test.sh b/harbor/tasks/v3_depth5_delayed_net_revenue_t5/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t5/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t6/environment/Dockerfile b/harbor/tasks/v3_depth5_delayed_net_revenue_t6/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t6/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t6/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_delayed_net_revenue_t6/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t6/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t6/environment/task.json b/harbor/tasks/v3_depth5_delayed_net_revenue_t6/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..b40a4b2118ecdc6f5471c50fa10a2738a18b2988 --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t6/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_delayed_net_revenue_t6", + "depth": 5, + "family": "delayed_net_revenue", + "params": { + "region": "LATAM", + "category": "B", + "month_start": 202601, + "month_end": 202604 + }, + "question": "For region LATAM, category B, months 202601-202604: get the paid/refunded orders, keep only orders whose ship_days exceeded promised_days, compute net USD revenue after loyalty discounts and refunds for those delayed orders, and report the total.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t6/instruction.md b/harbor/tasks/v3_depth5_delayed_net_revenue_t6/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..7b69c3c5acf587b80f4b0e05abdbdab40733cf51 --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t6/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region LATAM, category B, months 202601-202604: get the paid/refunded orders, keep only orders whose ship_days exceeded promised_days, compute net USD revenue after loyalty discounts and refunds for those delayed orders, and report the total. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t6/task.toml b/harbor/tasks/v3_depth5_delayed_net_revenue_t6/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..e8540895b78369a76d1c0615c19640e42df5c217 --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t6/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_delayed_net_revenue_t6" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t6/tests/check_output_shape.py b/harbor/tasks/v3_depth5_delayed_net_revenue_t6/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t6/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t6/tests/test.sh b/harbor/tasks/v3_depth5_delayed_net_revenue_t6/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t6/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t7/environment/Dockerfile b/harbor/tasks/v3_depth5_delayed_net_revenue_t7/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t7/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t7/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_delayed_net_revenue_t7/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t7/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t7/environment/task.json b/harbor/tasks/v3_depth5_delayed_net_revenue_t7/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..57f85caf2ad3b5552c6fbc2e251b1d0b8d8a19a9 --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t7/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_delayed_net_revenue_t7", + "depth": 5, + "family": "delayed_net_revenue", + "params": { + "region": "LATAM", + "category": "C", + "month_start": 202603, + "month_end": 202606 + }, + "question": "For region LATAM, category C, months 202603-202606: get the paid/refunded orders, keep only orders whose ship_days exceeded promised_days, compute net USD revenue after loyalty discounts and refunds for those delayed orders, and report the total.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t7/instruction.md b/harbor/tasks/v3_depth5_delayed_net_revenue_t7/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..3cc3f41bfa447dc4c739bc5d3f60b0102b403ba9 --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t7/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region LATAM, category C, months 202603-202606: get the paid/refunded orders, keep only orders whose ship_days exceeded promised_days, compute net USD revenue after loyalty discounts and refunds for those delayed orders, and report the total. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t7/task.toml b/harbor/tasks/v3_depth5_delayed_net_revenue_t7/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..f99b09d45a2b329a27ae63b9e420186c6c444eac --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t7/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_delayed_net_revenue_t7" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t7/tests/check_output_shape.py b/harbor/tasks/v3_depth5_delayed_net_revenue_t7/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t7/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t7/tests/test.sh b/harbor/tasks/v3_depth5_delayed_net_revenue_t7/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t7/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t8/environment/Dockerfile b/harbor/tasks/v3_depth5_delayed_net_revenue_t8/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t8/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t8/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_delayed_net_revenue_t8/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t8/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t8/environment/task.json b/harbor/tasks/v3_depth5_delayed_net_revenue_t8/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..4d054662a641b5a4d369f11af932de03839340fa --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t8/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_delayed_net_revenue_t8", + "depth": 5, + "family": "delayed_net_revenue", + "params": { + "region": "NA", + "category": "D", + "month_start": 202602, + "month_end": 202606 + }, + "question": "For region NA, category D, months 202602-202606: get the paid/refunded orders, keep only orders whose ship_days exceeded promised_days, compute net USD revenue after loyalty discounts and refunds for those delayed orders, and report the total.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t8/instruction.md b/harbor/tasks/v3_depth5_delayed_net_revenue_t8/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..616b69f52437e5ca85faca54ca14edabdadcf2a7 --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t8/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region NA, category D, months 202602-202606: get the paid/refunded orders, keep only orders whose ship_days exceeded promised_days, compute net USD revenue after loyalty discounts and refunds for those delayed orders, and report the total. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t8/task.toml b/harbor/tasks/v3_depth5_delayed_net_revenue_t8/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..3fe768a5df5cafa46ebae99d647c054426c9c94c --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t8/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_delayed_net_revenue_t8" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t8/tests/check_output_shape.py b/harbor/tasks/v3_depth5_delayed_net_revenue_t8/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t8/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t8/tests/test.sh b/harbor/tasks/v3_depth5_delayed_net_revenue_t8/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t8/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t9/environment/Dockerfile b/harbor/tasks/v3_depth5_delayed_net_revenue_t9/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t9/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t9/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_delayed_net_revenue_t9/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t9/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t9/environment/task.json b/harbor/tasks/v3_depth5_delayed_net_revenue_t9/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..921a1c720263ee1c2782b22d999150587972f841 --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t9/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_delayed_net_revenue_t9", + "depth": 5, + "family": "delayed_net_revenue", + "params": { + "region": "EU", + "category": "A", + "month_start": 202602, + "month_end": 202605 + }, + "question": "For region EU, category A, months 202602-202605: get the paid/refunded orders, keep only orders whose ship_days exceeded promised_days, compute net USD revenue after loyalty discounts and refunds for those delayed orders, and report the total.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t9/instruction.md b/harbor/tasks/v3_depth5_delayed_net_revenue_t9/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..66fd4f8811f044511a7803a8759f030e7c2b4781 --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t9/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region EU, category A, months 202602-202605: get the paid/refunded orders, keep only orders whose ship_days exceeded promised_days, compute net USD revenue after loyalty discounts and refunds for those delayed orders, and report the total. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t9/task.toml b/harbor/tasks/v3_depth5_delayed_net_revenue_t9/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..c77d191527c3aecf8b3301c94abdeb0717f1c08a --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t9/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_delayed_net_revenue_t9" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t9/tests/check_output_shape.py b/harbor/tasks/v3_depth5_delayed_net_revenue_t9/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t9/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_delayed_net_revenue_t9/tests/test.sh b/harbor/tasks/v3_depth5_delayed_net_revenue_t9/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_delayed_net_revenue_t9/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t0/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t0/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t0/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t0/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t0/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t0/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t0/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t0/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..6c4b4e0cf79093c2ab584f17e054500c3ef9a318 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t0/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t0", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "NA", + "category": "A", + "month_start": 202601, + "month_end": 202603 + }, + "question": "Find the top-selling product id for region NA, category A, months 202601-202603. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t0/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t0/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..ef65c748eb7e463b606f6e11c28b5e09be339c8c --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t0/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region NA, category A, months 202601-202603. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t0/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t0/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..8fa8ad4d3c42b0a2496d4091ceecb4f8d0666f5c --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t0/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t0" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t0/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t0/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t0/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t0/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t0/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t0/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t1/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t1/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t1/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t1/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t1/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t1/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t1/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t1/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..abe836e4a9b55d8aca86a72ff9edc1ed9bf313bc --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t1/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t1", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "NA", + "category": "B", + "month_start": 202604, + "month_end": 202606 + }, + "question": "Find the top-selling product id for region NA, category B, months 202604-202606. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t1/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t1/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..8f32b3ae5edee55e3ced45625a664ba41a77027d --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t1/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region NA, category B, months 202604-202606. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t1/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t1/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..4c94e953532e3fba09209395705bb042a817a3d2 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t1/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t1" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t1/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t1/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t1/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t1/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t1/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t1/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t10/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t10/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t10/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t10/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t10/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t10/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t10/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t10/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..1739fca69f0457328f545b13b638e9c329ac5210 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t10/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t10", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "APAC", + "category": "B", + "month_start": 202604, + "month_end": 202606 + }, + "question": "Find the top-selling product id for region APAC, category B, months 202604-202606. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t10/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t10/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..e3c342b7af559b15239dce2e1d95c324d8a10e1f --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t10/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region APAC, category B, months 202604-202606. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t10/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t10/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..64b700a04c72244caa2d303c22e2ab46e0d7558c --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t10/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t10" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t10/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t10/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t10/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t10/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t10/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t10/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t11/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t11/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t11/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t11/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t11/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t11/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t11/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t11/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..a4b04c4bdbb6d5c660fdb5e05d3917c49f5d989e --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t11/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t11", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "LATAM", + "category": "A", + "month_start": 202601, + "month_end": 202603 + }, + "question": "Find the top-selling product id for region LATAM, category A, months 202601-202603. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t11/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t11/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..150efbf53323a3eca9a839c7275656389ef87421 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t11/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region LATAM, category A, months 202601-202603. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t11/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t11/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..310cf37c106eb4fbcd7d9b24093b30667e137242 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t11/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t11" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t11/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t11/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t11/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t11/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t11/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t11/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t12/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t12/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t12/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t12/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t12/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t12/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t12/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t12/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..02024bad632edaa5c05c2517b4aa57b6c93490fb --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t12/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t12", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "NA", + "category": "C", + "month_start": 202601, + "month_end": 202606 + }, + "question": "Find the top-selling product id for region NA, category C, months 202601-202606. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t12/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t12/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..b21bfa7bf2e3f38b63326a93bccdaaa1f0da7136 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t12/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region NA, category C, months 202601-202606. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t12/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t12/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..12c9b9f3bc071703ab10f9da0433465c801272b0 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t12/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t12" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t12/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t12/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t12/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t12/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t12/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t12/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t13/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t13/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t13/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t13/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t13/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t13/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t13/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t13/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..29819b78a02d2ce993d9432e74ed5b264829e555 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t13/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t13", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "EU", + "category": "B", + "month_start": 202601, + "month_end": 202603 + }, + "question": "Find the top-selling product id for region EU, category B, months 202601-202603. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t13/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t13/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..3b4f06bf8846810c3d14bc203bc6c4d39473a1b7 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t13/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region EU, category B, months 202601-202603. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t13/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t13/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..4fdda93046b50e048f390fca1a90fc6f6f95a0b3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t13/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t13" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t13/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t13/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t13/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t13/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t13/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t13/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t14/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t14/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t14/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t14/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t14/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t14/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t14/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t14/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..2fd15ab2872ba453301408ad81f218065ffc751c --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t14/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t14", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "APAC", + "category": "C", + "month_start": 202602, + "month_end": 202606 + }, + "question": "Find the top-selling product id for region APAC, category C, months 202602-202606. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t14/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t14/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..55e6add1908f56ae566b6f4c1e8d47d939b04738 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t14/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region APAC, category C, months 202602-202606. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t14/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t14/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..ad9ae3ef53d55832afd02f8fb4ba9b417dc0f9a9 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t14/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t14" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t14/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t14/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t14/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t14/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t14/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t14/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t15/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t15/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t15/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t15/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t15/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t15/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t15/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t15/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..10ba11236fae3d2ff0e32c8ebd9e529bc24b0cec --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t15/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t15", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "LATAM", + "category": "D", + "month_start": 202601, + "month_end": 202605 + }, + "question": "Find the top-selling product id for region LATAM, category D, months 202601-202605. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t15/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t15/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..5edd37645e3ae461c29d32a697474d0f96baa282 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t15/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region LATAM, category D, months 202601-202605. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t15/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t15/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..d7822803d77fc797698345f9f147d05f55e4cf6f --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t15/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t15" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t15/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t15/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t15/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t15/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t15/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t15/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t16/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t16/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t16/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t16/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t16/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t16/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t16/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t16/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..a0dcdd50b95f98fd5798a3ec121a73131d0cd37b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t16/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t16", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "NA", + "category": "A", + "month_start": 202604, + "month_end": 202606 + }, + "question": "Find the top-selling product id for region NA, category A, months 202604-202606. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t16/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t16/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..2f218268a184942ea3e99a4e1f2d92376a34ee68 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t16/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region NA, category A, months 202604-202606. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t16/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t16/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..daed4e81dc60bb8721272a9ddb2a2956c31b6b4e --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t16/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t16" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t16/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t16/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t16/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t16/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t16/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t16/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t17/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t17/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t17/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t17/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t17/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t17/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t17/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t17/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..428b80ea1c7293449dd2c4b1ecb625d6dee62e61 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t17/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t17", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "EU", + "category": "D", + "month_start": 202601, + "month_end": 202602 + }, + "question": "Find the top-selling product id for region EU, category D, months 202601-202602. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t17/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t17/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..2ee8b918f4280927bc8980d5ab7fc67c039b198d --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t17/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region EU, category D, months 202601-202602. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t17/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t17/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..101d0c41cb1144050a62149d121a68664b80309e --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t17/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t17" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t17/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t17/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t17/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t17/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t17/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t17/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t18/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t18/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t18/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t18/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t18/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t18/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t18/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t18/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..0f4097e886ed81cd0cab61072a270bba8918679f --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t18/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t18", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "APAC", + "category": "A", + "month_start": 202603, + "month_end": 202604 + }, + "question": "Find the top-selling product id for region APAC, category A, months 202603-202604. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t18/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t18/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..a7120abd00341cd7d593b8b6515e21d20b39c54d --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t18/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region APAC, category A, months 202603-202604. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t18/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t18/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..a608de20cec9c6d25cdb68945a6d3a8075377ec5 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t18/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t18" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t18/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t18/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t18/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t18/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t18/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t18/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t19/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t19/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t19/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t19/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t19/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t19/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t19/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t19/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..018a97b2e8a6467d953739f1dc2bcff84b4672ac --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t19/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t19", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "LATAM", + "category": "B", + "month_start": 202605, + "month_end": 202606 + }, + "question": "Find the top-selling product id for region LATAM, category B, months 202605-202606. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t19/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t19/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..cecc72e9783d848749d831f3d11233c09b0db2c0 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t19/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region LATAM, category B, months 202605-202606. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t19/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t19/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..d9608a968c575670e233b95aabe0d8b3514140ef --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t19/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t19" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t19/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t19/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t19/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t19/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t19/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t19/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t2/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t2/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t2/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t2/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t2/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t2/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t2/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t2/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..1f8d98a7d759b083740916dcbbc3c786bf50f368 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t2/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t2", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "EU", + "category": "C", + "month_start": 202601, + "month_end": 202604 + }, + "question": "Find the top-selling product id for region EU, category C, months 202601-202604. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t2/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t2/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..d64c967fd7d352d409a13988f9a279f68763011e --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t2/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region EU, category C, months 202601-202604. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t2/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t2/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..185fefceebfce65d3932b0d6d895bdbe5a5116c4 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t2/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t2" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t2/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t2/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t2/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t2/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t2/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t2/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t20/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t20/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t20/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t20/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t20/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t20/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t20/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t20/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..67789d47d77459e621ad3148407eeb9f1e3cb580 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t20/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t20", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "NA", + "category": "C", + "month_start": 202602, + "month_end": 202604 + }, + "question": "Find the top-selling product id for region NA, category C, months 202602-202604. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t20/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t20/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..295e06e3c149cebd6ff9403b821944ab8cbdd0d6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t20/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region NA, category C, months 202602-202604. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t20/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t20/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..8c7b5d2b516416d8956bc585516ba6918c8ccc94 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t20/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t20" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t20/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t20/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t20/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t20/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t20/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t20/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t21/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t21/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t21/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t21/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t21/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t21/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t21/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t21/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..56e2589ec66703405a1b53458880352ed0fa3c3a --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t21/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t21", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "EU", + "category": "B", + "month_start": 202604, + "month_end": 202606 + }, + "question": "Find the top-selling product id for region EU, category B, months 202604-202606. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t21/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t21/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..1675c390a04b49cd838a1757af16d2f1775ecfba --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t21/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region EU, category B, months 202604-202606. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t21/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t21/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..c08ae68ab16ef2ae2ec382606bdba7ef98080bcb --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t21/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t21" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t21/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t21/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t21/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t21/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t21/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t21/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t22/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t22/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t22/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t22/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t22/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t22/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t22/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t22/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..3cd8044ff1adbfc85007c4be5218297af486bd68 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t22/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t22", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "APAC", + "category": "D", + "month_start": 202603, + "month_end": 202605 + }, + "question": "Find the top-selling product id for region APAC, category D, months 202603-202605. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t22/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t22/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..9a072daf050ed1fda42189022bfc16f91e128eac --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t22/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region APAC, category D, months 202603-202605. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t22/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t22/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..a4bea6e195522ba3e7fdab3e2babae4626e8eab2 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t22/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t22" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t22/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t22/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t22/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t22/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t22/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t22/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t23/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t23/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t23/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t23/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t23/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t23/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t23/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t23/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..cece7ac2d312fcc76e2d06a8f9bf00b118e30709 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t23/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t23", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "LATAM", + "category": "C", + "month_start": 202601, + "month_end": 202602 + }, + "question": "Find the top-selling product id for region LATAM, category C, months 202601-202602. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t23/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t23/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..9dfa634abb2c8db0688c05b5892e93cc46157420 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t23/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region LATAM, category C, months 202601-202602. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t23/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t23/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..70f63c7aee68c99eed87eb58d88207fefd6026a1 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t23/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t23" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t23/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t23/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t23/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t23/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t23/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t23/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t24/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t24/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t24/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t24/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t24/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t24/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t24/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t24/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..7951a48fba68a5f5cbcb28c54217f3653d748907 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t24/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t24", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "NA", + "category": "D", + "month_start": 202601, + "month_end": 202603 + }, + "question": "Find the top-selling product id for region NA, category D, months 202601-202603. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t24/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t24/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..08d9c597109b0075221c1dac7d55027bbfbb8599 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t24/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region NA, category D, months 202601-202603. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t24/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t24/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..f7b1c6adf4f7e65877a1fc7624b01c905e6886eb --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t24/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t24" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t24/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t24/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t24/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t24/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t24/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t24/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t3/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t3/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t3/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t3/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t3/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t3/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t3/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t3/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..522562b3451e926353e9276d7b194661f551782f --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t3/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t3", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "EU", + "category": "D", + "month_start": 202603, + "month_end": 202606 + }, + "question": "Find the top-selling product id for region EU, category D, months 202603-202606. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t3/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t3/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..f801307343e04903f52ed2fc10e3d98a453cefb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t3/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region EU, category D, months 202603-202606. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t3/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t3/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..e26255b9290abfa10cb115810852e07c7fe5c995 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t3/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t3" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t3/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t3/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t3/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t3/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t3/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t3/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t4/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t4/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t4/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t4/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t4/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t4/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t4/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t4/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..000376302ad36f6574b1e3a70c1e961c2113eb19 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t4/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t4", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "APAC", + "category": "A", + "month_start": 202602, + "month_end": 202605 + }, + "question": "Find the top-selling product id for region APAC, category A, months 202602-202605. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t4/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t4/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..ff943e32443d82a0b98da081beaf517c4ea85140 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t4/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region APAC, category A, months 202602-202605. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t4/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t4/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..25993b99458cab7d32a320f3864ff35aaf1f3c0e --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t4/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t4" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t4/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t4/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t4/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t4/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t4/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t4/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t5/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t5/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t5/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t5/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t5/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t5/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t5/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t5/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..54f93da60713899df3211946c3a2913d420b042a --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t5/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t5", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "APAC", + "category": "D", + "month_start": 202601, + "month_end": 202606 + }, + "question": "Find the top-selling product id for region APAC, category D, months 202601-202606. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t5/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t5/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..65d0f9c657814b3c92c4cdbbd65b4ab2c32d7eaf --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t5/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region APAC, category D, months 202601-202606. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t5/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t5/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..d6d0b36076606ec6f34c84040fde7f4d6bdd3b39 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t5/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t5" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t5/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t5/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t5/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t5/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t5/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t5/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t6/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t6/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t6/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t6/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t6/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t6/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t6/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t6/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..4443b823d813066aab30fd30924f13b78d4a1e0f --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t6/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t6", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "LATAM", + "category": "B", + "month_start": 202601, + "month_end": 202604 + }, + "question": "Find the top-selling product id for region LATAM, category B, months 202601-202604. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t6/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t6/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..120d57c45041cecac023f557bc9831b891668908 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t6/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region LATAM, category B, months 202601-202604. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t6/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t6/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..c056e9cecd87d005f2802e482ba4a3fd8efa2235 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t6/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t6" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t6/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t6/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t6/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t6/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t6/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t6/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t7/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t7/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t7/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t7/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t7/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t7/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t7/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t7/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..56dcbd6b1ee5502f1b9b2471748594afec92c902 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t7/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t7", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "LATAM", + "category": "C", + "month_start": 202603, + "month_end": 202606 + }, + "question": "Find the top-selling product id for region LATAM, category C, months 202603-202606. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t7/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t7/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..77e552b5407aac23adf5aca64736e63c47d2c8c2 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t7/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region LATAM, category C, months 202603-202606. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t7/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t7/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..f468135360b64a5369599186b314f65f0242750e --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t7/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t7" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t7/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t7/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t7/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t7/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t7/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t7/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t8/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t8/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t8/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t8/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t8/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t8/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t8/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t8/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..0624c806ca23d0ac16ad50b8582c7bb75034cc17 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t8/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t8", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "NA", + "category": "D", + "month_start": 202602, + "month_end": 202606 + }, + "question": "Find the top-selling product id for region NA, category D, months 202602-202606. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t8/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t8/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..cdaea3b9c6f8a2b3a135a2848b6bbd65e41ac1f2 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t8/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region NA, category D, months 202602-202606. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t8/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t8/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..d93cac72c3cd0c5b32eec2735575b6918f5cea67 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t8/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t8" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t8/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t8/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t8/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t8/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t8/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t8/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t9/environment/Dockerfile b/harbor/tasks/v3_depth5_reorder_shortfall_t9/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t9/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t9/environment/flowbench_tools.py b/harbor/tasks/v3_depth5_reorder_shortfall_t9/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t9/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t9/environment/task.json b/harbor/tasks/v3_depth5_reorder_shortfall_t9/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..ce14eaacf2d9298d47906fb6895a7cf94bee608b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t9/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth5_reorder_shortfall_t9", + "depth": 5, + "family": "reorder_shortfall", + "params": { + "region": "EU", + "category": "A", + "month_start": 202602, + "month_end": 202605 + }, + "question": "Find the top-selling product id for region EU, category A, months 202602-202605. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position).", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t9/instruction.md b/harbor/tasks/v3_depth5_reorder_shortfall_t9/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..3fcb48f52c0e310ff7c9892c3df2cf0b3d92ef9a --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t9/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +Find the top-selling product id for region EU, category A, months 202602-202605. For that product, compute inventory position (on_hand - reserved + inbound), estimate daily demand as floor(units sold in the window / 4) with a minimum of 1, compute planning demand as (lead_time_days + 21 safety-stock days) * daily demand, and report the reorder shortfall max(0, planning demand - inventory position). + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t9/task.toml b/harbor/tasks/v3_depth5_reorder_shortfall_t9/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..7fe180fd0be63b74a0a1a6937590a5651478ba53 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t9/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth5_reorder_shortfall_t9" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 5 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t9/tests/check_output_shape.py b/harbor/tasks/v3_depth5_reorder_shortfall_t9/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t9/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth5_reorder_shortfall_t9/tests/test.sh b/harbor/tasks/v3_depth5_reorder_shortfall_t9/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth5_reorder_shortfall_t9/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t0/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t0/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t0/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t0/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t0/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t0/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t0/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t0/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..2670b231e22edc2fb1c6faaffe2999c1427f1785 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t0/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t0", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "NA", + "category": "A", + "month_start": 202601, + "month_end": 202603 + }, + "question": "For region NA, category A, months 202601-202603: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t0/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t0/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..6b484099b20a25ff92e0fb27413be1f9855252ce --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t0/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region NA, category A, months 202601-202603: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t0/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t0/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..6448dfaa006cbcfef0702779e209d5936c9bf8ed --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t0/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t0" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t0/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t0/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t0/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t0/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t0/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t0/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t1/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t1/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t1/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t1/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t1/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t1/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t1/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t1/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..87a5f9789f25d4432ab64349428cf988a07d97e2 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t1/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t1", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "NA", + "category": "B", + "month_start": 202604, + "month_end": 202606 + }, + "question": "For region NA, category B, months 202604-202606: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t1/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t1/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..4c12f8f8f6b30ec987c1a49a0ec0541517812208 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t1/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region NA, category B, months 202604-202606: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t1/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t1/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..b53ce686394d60f22c6321b065cc9f90d37efa77 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t1/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t1" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t1/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t1/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t1/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t1/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t1/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t1/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t10/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t10/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t10/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t10/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t10/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t10/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t10/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t10/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..1f74fcc9f7775782b45cff5a758f5106dc94e4b7 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t10/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t10", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "APAC", + "category": "B", + "month_start": 202604, + "month_end": 202606 + }, + "question": "For region APAC, category B, months 202604-202606: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t10/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t10/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..745f957f53727653fb2f99ddfbb1a4934e6d662b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t10/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region APAC, category B, months 202604-202606: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t10/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t10/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..f8a9a2541599806049c585157e8f0b3063547b62 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t10/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t10" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t10/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t10/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t10/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t10/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t10/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t10/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t11/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t11/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t11/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t11/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t11/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t11/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t11/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t11/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..ed046d9f72d69e7e13586c25f1581a489d03bac5 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t11/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t11", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "LATAM", + "category": "A", + "month_start": 202601, + "month_end": 202603 + }, + "question": "For region LATAM, category A, months 202601-202603: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t11/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t11/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..7fc9c127234fecd35399f88a939106e2b59b4fc9 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t11/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region LATAM, category A, months 202601-202603: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t11/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t11/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..15889942f6b847e30a9a5a6946b3b499f5d2a58f --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t11/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t11" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t11/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t11/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t11/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t11/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t11/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t11/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t12/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t12/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t12/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t12/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t12/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t12/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t12/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t12/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..b8b5328db0862a4b1538c1644b0f929d11dad4db --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t12/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t12", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "NA", + "category": "C", + "month_start": 202601, + "month_end": 202606 + }, + "question": "For region NA, category C, months 202601-202606: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t12/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t12/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..1e4c0368b04b658dc23f8d556c828b78ce292a30 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t12/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region NA, category C, months 202601-202606: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t12/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t12/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..c0959bab4fb9a91864e1ba98d7e682b0c3cf22b9 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t12/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t12" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t12/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t12/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t12/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t12/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t12/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t12/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t13/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t13/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t13/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t13/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t13/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t13/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t13/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t13/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..28d4d6a81ee9391928dfef1b133b1805b0429220 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t13/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t13", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "EU", + "category": "B", + "month_start": 202601, + "month_end": 202603 + }, + "question": "For region EU, category B, months 202601-202603: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t13/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t13/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..51e4b02be281f2f3995ae913b6288aa02a371fe0 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t13/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region EU, category B, months 202601-202603: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t13/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t13/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..49706f99704c026aa8786c0c98cc4269d89524cf --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t13/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t13" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t13/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t13/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t13/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t13/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t13/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t13/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t14/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t14/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t14/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t14/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t14/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t14/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t14/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t14/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..7de0e5deb01bd832a5c1e44ce82f0cd8a41b4e7a --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t14/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t14", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "APAC", + "category": "C", + "month_start": 202602, + "month_end": 202606 + }, + "question": "For region APAC, category C, months 202602-202606: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t14/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t14/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..31d756339e6428c0704b86e63d19fd874771f64f --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t14/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region APAC, category C, months 202602-202606: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t14/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t14/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..8c3604f97207cfa54f6cc65a8e6f5f7471092b53 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t14/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t14" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t14/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t14/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t14/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t14/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t14/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t14/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t15/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t15/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t15/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t15/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t15/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t15/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t15/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t15/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..a84f95bd2fc5580b926dd4734c5e9018d498a1a8 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t15/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t15", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "LATAM", + "category": "D", + "month_start": 202601, + "month_end": 202605 + }, + "question": "For region LATAM, category D, months 202601-202605: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t15/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t15/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..ec5956ef3bcac2f0f3e258317a44b290304a413d --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t15/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region LATAM, category D, months 202601-202605: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t15/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t15/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..d0474f6982da1c3d98707eab341bcf81e950d9ed --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t15/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t15" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t15/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t15/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t15/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t15/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t15/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t15/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t16/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t16/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t16/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t16/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t16/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t16/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t16/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t16/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..84753eec7b81f60b5007d2d4f8809677d60449fd --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t16/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t16", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "NA", + "category": "A", + "month_start": 202604, + "month_end": 202606 + }, + "question": "For region NA, category A, months 202604-202606: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t16/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t16/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..946051ad813a70fc8a332f746aa875ece89c006d --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t16/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region NA, category A, months 202604-202606: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t16/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t16/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..7a99b04b6dd8b92d044a7c229eeefd03898ab425 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t16/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t16" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t16/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t16/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t16/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t16/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t16/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t16/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t17/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t17/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t17/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t17/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t17/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t17/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t17/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t17/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..ac78d49a033ed3b0ea712221b41dad1d1a205e31 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t17/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t17", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "EU", + "category": "D", + "month_start": 202601, + "month_end": 202602 + }, + "question": "For region EU, category D, months 202601-202602: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t17/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t17/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..3bbc108ae969b34f64c1632a8e78bb1dfa6d2225 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t17/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region EU, category D, months 202601-202602: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t17/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t17/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..ec07ee9248183f5e08a3815832cc95b45c8a718c --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t17/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t17" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t17/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t17/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t17/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t17/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t17/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t17/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t18/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t18/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t18/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t18/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t18/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t18/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t18/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t18/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..e7c2a5ad00d90d06f8674c2e223587810b11d0d9 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t18/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t18", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "APAC", + "category": "A", + "month_start": 202603, + "month_end": 202604 + }, + "question": "For region APAC, category A, months 202603-202604: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t18/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t18/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..96e6d56e7fcb787c2765f8f89a68ebc039272a87 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t18/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region APAC, category A, months 202603-202604: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t18/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t18/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..6983e6ad266617c0a0aee2c9e3d2f741c2064649 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t18/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t18" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t18/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t18/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t18/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t18/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t18/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t18/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t19/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t19/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t19/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t19/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t19/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t19/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t19/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t19/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..503444d6dbf02cc865956c22bfd0d57a5dd19ade --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t19/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t19", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "LATAM", + "category": "B", + "month_start": 202605, + "month_end": 202606 + }, + "question": "For region LATAM, category B, months 202605-202606: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t19/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t19/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..aaddb468b4e9ebce807169a3f963a28dabd4bdfb --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t19/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region LATAM, category B, months 202605-202606: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t19/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t19/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..d0d77337d5e1afb9da6457ac305b49ceb4d9f4a5 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t19/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t19" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t19/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t19/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t19/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t19/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t19/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t19/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t2/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t2/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t2/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t2/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t2/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t2/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t2/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t2/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..99a473421aaaa46965df4a6cbb9be6b626cd4ec9 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t2/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t2", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "EU", + "category": "C", + "month_start": 202601, + "month_end": 202604 + }, + "question": "For region EU, category C, months 202601-202604: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t2/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t2/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..b0dd190de47cbd6aa73502e5752cd48530b7b313 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t2/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region EU, category C, months 202601-202604: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t2/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t2/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..478668eb833e2da523714b4b106f3ba2cd29a1ba --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t2/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t2" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t2/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t2/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t2/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t2/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t2/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t2/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t20/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t20/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t20/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t20/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t20/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t20/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t20/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t20/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..39eb40b29a723d3c2c45a6a12fb83f06fbe9e97a --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t20/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t20", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "NA", + "category": "C", + "month_start": 202602, + "month_end": 202604 + }, + "question": "For region NA, category C, months 202602-202604: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t20/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t20/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..b9c28fe2fde3a85cfa2a26ee8a333708f7e3f6f1 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t20/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region NA, category C, months 202602-202604: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t20/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t20/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..b183c6e3b9727e7fc3f2ba904e8d8e426fb329b2 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t20/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t20" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t20/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t20/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t20/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t20/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t20/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t20/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t21/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t21/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t21/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t21/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t21/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t21/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t21/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t21/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..0cda1284c9474e161126fa12d11953e2c5d2973c --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t21/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t21", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "EU", + "category": "B", + "month_start": 202604, + "month_end": 202606 + }, + "question": "For region EU, category B, months 202604-202606: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t21/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t21/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..adf718a3e3b05d897a03df39d662fc1985b2177b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t21/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region EU, category B, months 202604-202606: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t21/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t21/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..3fd3941ee052fa5f4ff7d60af7971dbc1778b7dc --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t21/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t21" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t21/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t21/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t21/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t21/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t21/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t21/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t22/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t22/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t22/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t22/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t22/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t22/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t22/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t22/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..324a793defae9fe5fe4712a6335896c12a8d6c7d --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t22/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t22", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "APAC", + "category": "D", + "month_start": 202603, + "month_end": 202605 + }, + "question": "For region APAC, category D, months 202603-202605: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t22/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t22/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..bbde85ed13cdb679efaf4ac4022e63f56c6858ec --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t22/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region APAC, category D, months 202603-202605: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t22/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t22/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..cc052e829a51c4bbc7e4f9a4456a756f8c822291 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t22/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t22" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t22/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t22/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t22/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t22/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t22/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t22/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t23/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t23/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t23/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t23/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t23/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t23/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t23/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t23/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..473e67605d834b0dcedf2f379c7e3d9ae20e4701 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t23/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t23", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "LATAM", + "category": "C", + "month_start": 202601, + "month_end": 202602 + }, + "question": "For region LATAM, category C, months 202601-202602: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t23/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t23/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..3e311a9c8348dc86698c43f92485dfe41c70c99f --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t23/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region LATAM, category C, months 202601-202602: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t23/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t23/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..6c2dbae04f60c5d94595fa562d0858c39efd4470 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t23/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t23" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t23/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t23/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t23/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t23/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t23/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t23/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t24/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t24/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t24/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t24/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t24/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t24/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t24/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t24/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..3f0018e073bceb8ddc43a5adfbec4b1168150d72 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t24/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t24", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "NA", + "category": "D", + "month_start": 202601, + "month_end": 202603 + }, + "question": "For region NA, category D, months 202601-202603: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t24/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t24/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..4a7415ad428d3827aff3ca34f4e20b40ae8ffa9c --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t24/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region NA, category D, months 202601-202603: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t24/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t24/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..95d527e6480b65e15372fc01da43b94146a0b8d5 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t24/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t24" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t24/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t24/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t24/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t24/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t24/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t24/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t3/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t3/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t3/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t3/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t3/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t3/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t3/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t3/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..0e5bd6edbba8964172b94eea82a920765677d9e6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t3/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t3", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "EU", + "category": "D", + "month_start": 202603, + "month_end": 202606 + }, + "question": "For region EU, category D, months 202603-202606: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t3/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t3/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..59b09066d89285bdcb52095ce2c092423e27b0e9 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t3/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region EU, category D, months 202603-202606: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t3/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t3/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..50ba53519dd8fdb24d969abd63d1e3d9af87d8e9 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t3/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t3" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t3/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t3/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t3/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t3/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t3/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t3/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t4/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t4/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t4/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t4/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t4/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t4/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t4/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t4/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..56c08ebc16b957152d0a5d521e475c60a4438a80 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t4/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t4", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "APAC", + "category": "A", + "month_start": 202602, + "month_end": 202605 + }, + "question": "For region APAC, category A, months 202602-202605: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t4/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t4/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..17d8bd2bdf5179fd250b9c7b854f42c5525308a9 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t4/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region APAC, category A, months 202602-202605: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t4/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t4/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..39822f5cd82bca80f4c7cebe8af1fd4ab94c8b24 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t4/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t4" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t4/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t4/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t4/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t4/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t4/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t4/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t5/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t5/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t5/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t5/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t5/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t5/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t5/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t5/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..1ca3241918d99f38535296caa105ce03c7ae61eb --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t5/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t5", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "APAC", + "category": "D", + "month_start": 202601, + "month_end": 202606 + }, + "question": "For region APAC, category D, months 202601-202606: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t5/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t5/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..d8aa6dc21c1848d7470ac38768c13d60b6960b50 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t5/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region APAC, category D, months 202601-202606: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t5/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t5/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..88fe9949da4d8e2be0e6847644297ecb4552e6b9 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t5/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t5" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t5/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t5/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t5/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t5/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t5/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t5/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t6/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t6/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t6/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t6/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t6/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t6/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t6/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t6/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..7c58714a8f89e2a8a922e79d4e16ef1c52896cb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t6/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t6", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "LATAM", + "category": "B", + "month_start": 202601, + "month_end": 202604 + }, + "question": "For region LATAM, category B, months 202601-202604: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t6/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t6/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..9799989fd1d48d33cdd8e6ce5ca6915adce62839 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t6/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region LATAM, category B, months 202601-202604: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t6/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t6/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..263d0dfddaa1e0c9e4d8f4ba9727181fc07a7533 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t6/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t6" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t6/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t6/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t6/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t6/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t6/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t6/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t7/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t7/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t7/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t7/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t7/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t7/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t7/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t7/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..70e9c1c33c3f9d31b969dbf649ba8996646dae0f --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t7/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t7", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "LATAM", + "category": "C", + "month_start": 202603, + "month_end": 202606 + }, + "question": "For region LATAM, category C, months 202603-202606: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t7/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t7/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..09fa5bbe0d1811f462cd4782182b08df9bb0de76 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t7/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region LATAM, category C, months 202603-202606: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t7/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t7/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..dc21cb2e30bcb1b21bc9ef4a578cd7f85f641060 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t7/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t7" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t7/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t7/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t7/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t7/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t7/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t7/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t8/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t8/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t8/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t8/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t8/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t8/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t8/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t8/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..15ec703bf41def795afd481a764c1d2a4ed70b56 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t8/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t8", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "NA", + "category": "D", + "month_start": 202602, + "month_end": 202606 + }, + "question": "For region NA, category D, months 202602-202606: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t8/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t8/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..7a7180ff81b2cc3e52b3658a77b9dbb686947f1f --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t8/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region NA, category D, months 202602-202606: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t8/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t8/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..53e90d98db9fb4d2deb87185af633b5ddc446e11 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t8/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t8" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t8/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t8/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t8/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t8/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t8/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t8/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t9/environment/Dockerfile b/harbor/tasks/v3_depth6_breached_ticket_revenue_t9/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t9/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t9/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t9/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t9/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t9/environment/task.json b/harbor/tasks/v3_depth6_breached_ticket_revenue_t9/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..fd4ed09951e79936328d14c8bda738e685e1252d --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t9/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_breached_ticket_revenue_t9", + "depth": 6, + "family": "breached_ticket_revenue", + "params": { + "region": "EU", + "category": "A", + "month_start": 202602, + "month_end": 202605 + }, + "question": "For region EU, category A, months 202602-202605: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t9/instruction.md b/harbor/tasks/v3_depth6_breached_ticket_revenue_t9/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..c8722d31172a1cef2896dbf66e457f9a811d6f40 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t9/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region EU, category A, months 202602-202605: get the paid/refunded orders, find high and critical tickets for those orders, keep tickets that breach a 24-hour first-response or 120-hour resolution SLA, map those breached tickets back to their unique order ids, compute net USD revenue after discounts and refunds for those orders, and report the total at-risk revenue. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t9/task.toml b/harbor/tasks/v3_depth6_breached_ticket_revenue_t9/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..82ac7e143e0941a07d7c8825d121e555755e8e94 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t9/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_breached_ticket_revenue_t9" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t9/tests/check_output_shape.py b/harbor/tasks/v3_depth6_breached_ticket_revenue_t9/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t9/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_breached_ticket_revenue_t9/tests/test.sh b/harbor/tasks/v3_depth6_breached_ticket_revenue_t9/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_breached_ticket_revenue_t9/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_delay_sla_burden_t0/environment/Dockerfile b/harbor/tasks/v3_depth6_delay_sla_burden_t0/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_delay_sla_burden_t0/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_delay_sla_burden_t0/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_delay_sla_burden_t0/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_delay_sla_burden_t0/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_delay_sla_burden_t0/environment/task.json b/harbor/tasks/v3_depth6_delay_sla_burden_t0/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..a779e2f5a6055af20b92b59b9507a53a90783571 --- /dev/null +++ b/harbor/tasks/v3_depth6_delay_sla_burden_t0/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_delay_sla_burden_t0", + "depth": 6, + "family": "delay_sla_burden", + "params": { + "region": "NA", + "category": "A", + "month_start": 202601, + "month_end": 202603 + }, + "question": "For region NA, category A, months 202601-202603: get the orders, count delayed orders where ship_days exceeded promised_days, find high and critical support tickets for those orders, count tickets that breach a 24-hour first-response or 120-hour resolution SLA, and report delayed_order_count + breached_ticket_count.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_delay_sla_burden_t0/instruction.md b/harbor/tasks/v3_depth6_delay_sla_burden_t0/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..7ff25a98c52f795f89deedbe3d74a5efe1342d75 --- /dev/null +++ b/harbor/tasks/v3_depth6_delay_sla_burden_t0/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region NA, category A, months 202601-202603: get the orders, count delayed orders where ship_days exceeded promised_days, find high and critical support tickets for those orders, count tickets that breach a 24-hour first-response or 120-hour resolution SLA, and report delayed_order_count + breached_ticket_count. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_delay_sla_burden_t0/task.toml b/harbor/tasks/v3_depth6_delay_sla_burden_t0/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..74c6903e3d41ef9ada85e4d0c534bf9e5e3b25e3 --- /dev/null +++ b/harbor/tasks/v3_depth6_delay_sla_burden_t0/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_delay_sla_burden_t0" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_delay_sla_burden_t0/tests/check_output_shape.py b/harbor/tasks/v3_depth6_delay_sla_burden_t0/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_delay_sla_burden_t0/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_delay_sla_burden_t0/tests/test.sh b/harbor/tasks/v3_depth6_delay_sla_burden_t0/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_delay_sla_burden_t0/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_delay_sla_burden_t1/environment/Dockerfile b/harbor/tasks/v3_depth6_delay_sla_burden_t1/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_delay_sla_burden_t1/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_delay_sla_burden_t1/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_delay_sla_burden_t1/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_delay_sla_burden_t1/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_delay_sla_burden_t1/environment/task.json b/harbor/tasks/v3_depth6_delay_sla_burden_t1/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..15fae0617704dfb3eb719527215e57153c6d0ac3 --- /dev/null +++ b/harbor/tasks/v3_depth6_delay_sla_burden_t1/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_delay_sla_burden_t1", + "depth": 6, + "family": "delay_sla_burden", + "params": { + "region": "NA", + "category": "B", + "month_start": 202604, + "month_end": 202606 + }, + "question": "For region NA, category B, months 202604-202606: get the orders, count delayed orders where ship_days exceeded promised_days, find high and critical support tickets for those orders, count tickets that breach a 24-hour first-response or 120-hour resolution SLA, and report delayed_order_count + breached_ticket_count.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_delay_sla_burden_t1/instruction.md b/harbor/tasks/v3_depth6_delay_sla_burden_t1/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..eceb0454c837d25c080ea3f19c3435d0180e2063 --- /dev/null +++ b/harbor/tasks/v3_depth6_delay_sla_burden_t1/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region NA, category B, months 202604-202606: get the orders, count delayed orders where ship_days exceeded promised_days, find high and critical support tickets for those orders, count tickets that breach a 24-hour first-response or 120-hour resolution SLA, and report delayed_order_count + breached_ticket_count. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_delay_sla_burden_t1/task.toml b/harbor/tasks/v3_depth6_delay_sla_burden_t1/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..31bab0784dfa246997cb82ec15e02fb15cac0c63 --- /dev/null +++ b/harbor/tasks/v3_depth6_delay_sla_burden_t1/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_delay_sla_burden_t1" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0 diff --git a/harbor/tasks/v3_depth6_delay_sla_burden_t1/tests/check_output_shape.py b/harbor/tasks/v3_depth6_delay_sla_burden_t1/tests/check_output_shape.py new file mode 100644 index 0000000000000000000000000000000000000000..19511d8e36ceaadad8043c93afbca73e685e53b6 --- /dev/null +++ b/harbor/tasks/v3_depth6_delay_sla_burden_t1/tests/check_output_shape.py @@ -0,0 +1,42 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import os +import re +from pathlib import Path + + +def write_reward(score: float, diagnostic: str) -> None: + log_dir = Path(os.environ.get("LOG_DIR", "/logs/verifier")) + log_dir.mkdir(parents=True, exist_ok=True) + (log_dir / "reward.txt").write_text(f"{score:.6g}\n") + (log_dir / "reward.json").write_text( + json.dumps({"reward": score, "smoke_only": True}, indent=2) + "\n" + ) + (log_dir / "diagnostic.txt").write_text(diagnostic + "\n") + + +def main() -> int: + app_dir = Path(os.environ.get("APP_DIR", "/app")) + task = json.loads((app_dir / "task.json").read_text()) + answer_path = app_dir / "answer.txt" + if not answer_path.exists(): + write_reward(0.0, "missing /app/answer.txt") + return 1 + + got = answer_path.read_text().strip() + if task["answer_format"] == "exact integer": + ok = bool(re.fullmatch(r"-?\d+", got)) + diagnostic = "valid integer smoke output" if ok else f"not an exact integer: {got!r}" + else: + ok = bool(re.fullmatch(r"[A-Za-z]+", got)) + diagnostic = "valid code smoke output" if ok else f"not an exact code: {got!r}" + + # This is intentionally not official scoring; it only verifies I/O shape. + write_reward(1.0 if ok else 0.0, diagnostic) + return 0 if ok else 1 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/harbor/tasks/v3_depth6_delay_sla_burden_t1/tests/test.sh b/harbor/tasks/v3_depth6_delay_sla_burden_t1/tests/test.sh new file mode 100644 index 0000000000000000000000000000000000000000..8943b5d6a288c8fd56844aec54a60479395af94b --- /dev/null +++ b/harbor/tasks/v3_depth6_delay_sla_burden_t1/tests/test.sh @@ -0,0 +1,3 @@ +#!/usr/bin/env bash +set -u +python /tests/check_output_shape.py diff --git a/harbor/tasks/v3_depth6_delay_sla_burden_t10/environment/Dockerfile b/harbor/tasks/v3_depth6_delay_sla_burden_t10/environment/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..97377a0aa30bcf17ddd3345fff9e77e27b28ddb3 --- /dev/null +++ b/harbor/tasks/v3_depth6_delay_sla_burden_t10/environment/Dockerfile @@ -0,0 +1,4 @@ +FROM python:3.12-slim +WORKDIR /app +COPY flowbench_tools.py /app/flowbench_tools.py +COPY task.json /app/task.json diff --git a/harbor/tasks/v3_depth6_delay_sla_burden_t10/environment/flowbench_tools.py b/harbor/tasks/v3_depth6_delay_sla_burden_t10/environment/flowbench_tools.py new file mode 100644 index 0000000000000000000000000000000000000000..96661210d32ed1d8089959097124c723844255a3 --- /dev/null +++ b/harbor/tasks/v3_depth6_delay_sla_burden_t10/environment/flowbench_tools.py @@ -0,0 +1,311 @@ +"""Deterministic FlowBench v3 tools. This file contains no task answers.""" + + +REGIONS = ["NA", "EU", "APAC", "LATAM"] +CATEGORIES = ["A", "B", "C", "D"] +CURRENCIES = {"NA": "USD", "EU": "EUR", "APAC": "JPY", "LATAM": "BRL"} +FX_TO_USD_BP = {"USD": 10000, "EUR": 10900, "JPY": 67, "BRL": 1850} +TIERS = ["std", "gold", "plat"] +DISCOUNT_PCT = {"std": 0, "gold": 10, "plat": 20} +CHANNELS = ["web", "store", "partner"] +MONTHS = [202601, 202602, 202603, 202604, 202605, 202606] + +N_CUSTOMERS = 72 +N_PRODUCTS = 48 +N_ORDERS = 720 +N_TICKETS = 260 +N_RETURNS = 180 + + +def _h(*xs): + import hashlib + s = "|".join(str(x) for x in xs).encode() + return int.from_bytes(hashlib.sha256(s).digest()[:8], "big") & 0x7FFFFFFF + + +def build_data(): + customers = [] + for cid in range(N_CUSTOMERS): + region = REGIONS[_h("region", cid) % len(REGIONS)] + customers.append({ + "customer_id": cid, + "region": region, + "tier": TIERS[_h("tier", cid) % len(TIERS)], + "segment": ["consumer", "smb", "enterprise"][_h("segment", cid) % 3], + }) + + products = [] + for pid in range(N_PRODUCTS): + cat = CATEGORIES[_h("cat", pid) % len(CATEGORIES)] + unit_price = 12 + _h("price", pid) % 240 + cogs = unit_price * (45 + _h("margin", pid) % 35) // 100 + products.append({ + "product_id": pid, + "category": cat, + "unit_price_usd": unit_price, + "unit_cogs_usd": cogs, + "lead_time_days": 3 + _h("lead", pid) % 18, + }) + + inventory = [] + for pid in range(N_PRODUCTS): + for region in REGIONS: + inventory.append({ + "product_id": pid, + "region": region, + "on_hand": 10 + _h("onhand", pid, region) % 140, + "reserved": _h("reserved", pid, region) % 28, + "inbound": _h("inbound", pid, region) % 85, + }) + + orders = [] + for oid in range(N_ORDERS): + customer_id = _h("ocust", oid) % N_CUSTOMERS + product_id = _h("oprod", oid) % N_PRODUCTS + month = MONTHS[_h("month", oid) % len(MONTHS)] + qty = 1 + _h("qty", oid) % 9 + ship_days = 1 + _h("shipdays", oid) % 16 + promised_days = 3 + _h("promise", oid) % 9 + status = ["paid", "paid", "paid", "paid", "cancelled", "refunded"][ + _h("status", oid) % 6] + orders.append({ + "order_id": oid, + "customer_id": customer_id, + "product_id": product_id, + "month": month, + "qty": qty, + "status": status, + "channel": CHANNELS[_h("channel", oid) % len(CHANNELS)], + "ship_days": ship_days, + "promised_days": promised_days, + }) + + returns = [] + for rid in range(N_RETURNS): + oid = _h("roid", rid) % N_ORDERS + order = orders[oid] + if order["status"] == "cancelled": + continue + returned_qty = 1 + _h("rqty", rid) % max(1, order["qty"]) + reason = ["defect", "late", "changed_mind", "wrong_item"][_h("reason", rid) % 4] + returns.append({ + "return_id": rid, + "order_id": oid, + "month": MONTHS[_h("rmonth", rid) % len(MONTHS)], + "returned_qty": returned_qty, + "reason": reason, + }) + + tickets = [] + for tid in range(N_TICKETS): + oid = _h("toid", tid) % N_ORDERS + severity = ["low", "medium", "high", "critical"][_h("sev", tid) % 4] + opened = MONTHS[_h("tmonth", tid) % len(MONTHS)] + first_response_hours = 1 + _h("resp", tid) % 96 + resolution_hours = first_response_hours + _h("res", tid) % 240 + tickets.append({ + "ticket_id": tid, + "order_id": oid, + "opened_month": opened, + "severity": severity, + "first_response_hours": first_response_hours, + "resolution_hours": resolution_hours, + }) + + return { + "customers": customers, + "products": products, + "inventory": inventory, + "orders": orders, + "returns": returns, + "tickets": tickets, + "fx_to_usd_bp": FX_TO_USD_BP, + } + + +DATA = build_data() +CUSTOMERS = {c["customer_id"]: c for c in DATA["customers"]} +PRODUCTS = {p["product_id"]: p for p in DATA["products"]} +ORDERS = {o["order_id"]: o for o in DATA["orders"]} + + +def region_currency(region: str) -> str: + return CURRENCIES[region] + + +def get_orders(region: str, category: str, month_start: int, month_end: int) -> list: + """Paid/refunded order ids in a region/category/month window.""" + out = [] + for o in DATA["orders"]: + if o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + p = PRODUCTS[o["product_id"]] + if c["region"] == region and p["category"] == category and month_start <= o["month"] <= month_end: + out.append(o["order_id"]) + return sorted(out) + + +def order_gross_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * p["unit_price_usd"] + + +def order_margin_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + return o["qty"] * (p["unit_price_usd"] - p["unit_cogs_usd"]) + + +def refund_usd(order_id: int) -> int: + o = ORDERS[order_id] + p = PRODUCTS[o["product_id"]] + qty = sum(r["returned_qty"] for r in DATA["returns"] if r["order_id"] == order_id) + qty = min(qty, o["qty"]) + return qty * p["unit_price_usd"] + + +def customer_tier(order_id: int) -> str: + o = ORDERS[order_id] + return CUSTOMERS[o["customer_id"]]["tier"] + + +def apply_discount(amount_usd: int, tier: str) -> int: + return amount_usd * (100 - DISCOUNT_PCT[tier]) // 100 + + +def net_revenue_usd(order_id: int) -> int: + gross = apply_discount(order_gross_usd(order_id), customer_tier(order_id)) + return max(0, gross - refund_usd(order_id)) + + +def to_local(amount_usd: int, currency: str) -> int: + return amount_usd * 10000 // FX_TO_USD_BP[currency] + + +def inventory_position(product_id: int, region: str) -> int: + row = next(x for x in DATA["inventory"] + if x["product_id"] == product_id and x["region"] == region) + return row["on_hand"] - row["reserved"] + row["inbound"] + + +def product_lead_time(product_id: int) -> int: + return PRODUCTS[product_id]["lead_time_days"] + + +def product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int: + totals = {} + for oid in get_orders(region, category, month_start, month_end): + o = ORDERS[oid] + totals[o["product_id"]] = totals.get(o["product_id"], 0) + o["qty"] + if not totals: + return -1 + return min((-qty, pid) for pid, qty in totals.items())[1] + + +def units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int: + total = 0 + for o in DATA["orders"]: + if o["product_id"] != product_id or o["status"] not in ("paid", "refunded"): + continue + c = CUSTOMERS[o["customer_id"]] + if c["region"] == region and month_start <= o["month"] <= month_end: + total += o["qty"] + return total + + +def tickets_for_orders(order_ids: list, severity: str) -> list: + s = set(order_ids) + return sorted(t["ticket_id"] for t in DATA["tickets"] + if t["order_id"] in s and t["severity"] == severity) + + +def ticket_order_id(ticket_id: int) -> int: + return next(t["order_id"] for t in DATA["tickets"] if t["ticket_id"] == ticket_id) + + +def sla_breached(ticket_id: int, first_response_limit_hours: int, + resolution_limit_hours: int) -> bool: + t = next(x for x in DATA["tickets"] if x["ticket_id"] == ticket_id) + return (t["first_response_hours"] > first_response_limit_hours or + t["resolution_hours"] > resolution_limit_hours) + + +def delayed_orders(order_ids: list) -> list: + s = set(order_ids) + return sorted(o["order_id"] for o in DATA["orders"] + if o["order_id"] in s and o["ship_days"] > o["promised_days"]) + + +def unique_customers(order_ids: list) -> list: + return sorted({ORDERS[oid]["customer_id"] for oid in order_ids}) + + +def count_items(values: list) -> int: + return len(values) + + +def sum_values(values: list) -> int: + return sum(values) + + +def count_true(values: list) -> int: + return sum(1 for v in values if bool(v)) + + +def count_below(values: list, threshold: int) -> int: + return sum(1 for v in values if v < threshold) + + +TOOLS = { + "region_currency": ("region_currency(region: str) -> str", + "ISO currency code for a region.", region_currency), + "get_orders": ("get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]", + "Paid/refunded order ids for a region, product category, and inclusive month window.", + get_orders), + "order_gross_usd": ("order_gross_usd(order_id: int) -> int", + "Gross order revenue in USD before discounts/refunds.", order_gross_usd), + "order_margin_usd": ("order_margin_usd(order_id: int) -> int", + "Gross order margin in USD before discounts/refunds.", order_margin_usd), + "refund_usd": ("refund_usd(order_id: int) -> int", + "Refund amount in USD for returned quantity on an order.", refund_usd), + "customer_tier": ("customer_tier(order_id: int) -> str", + "Customer loyalty tier for an order.", customer_tier), + "apply_discount": ("apply_discount(amount_usd: int, tier: str) -> int", + "Apply the order customer's loyalty discount to a USD amount.", apply_discount), + "net_revenue_usd": ("net_revenue_usd(order_id: int) -> int", + "Net USD revenue after loyalty discount and refunds.", net_revenue_usd), + "to_local": ("to_local(amount_usd: int, currency: str) -> int", + "Convert a USD integer amount to a local-currency integer.", to_local), + "inventory_position": ("inventory_position(product_id: int, region: str) -> int", + "Available inventory position: on_hand - reserved + inbound.", + inventory_position), + "product_lead_time": ("product_lead_time(product_id: int) -> int", + "Supplier lead time in days for a product.", product_lead_time), + "product_id_for_top_seller": ( + "product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int", + "Product id with the highest units sold in a region/category/month window.", + product_id_for_top_seller), + "units_sold": ("units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int", + "Units sold for one product in a region and month window.", units_sold), + "tickets_for_orders": ("tickets_for_orders(order_ids: list[int], severity: str) -> list[int]", + "Support ticket ids for given orders and severity.", tickets_for_orders), + "ticket_order_id": ("ticket_order_id(ticket_id: int) -> int", + "Order id associated with a support ticket.", ticket_order_id), + "sla_breached": ( + "sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool", + "Whether a support ticket breaches first-response or resolution SLA.", + sla_breached), + "delayed_orders": ("delayed_orders(order_ids: list[int]) -> list[int]", + "Subset of order ids whose ship_days exceeded promised_days.", delayed_orders), + "unique_customers": ("unique_customers(order_ids: list[int]) -> list[int]", + "Unique customer ids among the given orders.", unique_customers), + "count_items": ("count_items(values: list) -> int", "Length of a list.", count_items), + "sum_values": ("sum_values(values: list[int]) -> int", "Sum a list of integers.", sum_values), + "count_true": ("count_true(values: list[bool]) -> int", "Count true values.", count_true), + "count_below": ("count_below(values: list[int], threshold: int) -> int", + "Count values below a threshold.", count_below), +} + +__all__ = [name for name in TOOLS.keys()] + ['TOOLS'] diff --git a/harbor/tasks/v3_depth6_delay_sla_burden_t10/environment/task.json b/harbor/tasks/v3_depth6_delay_sla_burden_t10/environment/task.json new file mode 100644 index 0000000000000000000000000000000000000000..2a55fbe3cd33d381d4b9259b232f9ede3185e03b --- /dev/null +++ b/harbor/tasks/v3_depth6_delay_sla_burden_t10/environment/task.json @@ -0,0 +1,13 @@ +{ + "task_id": "v3_depth6_delay_sla_burden_t10", + "depth": 6, + "family": "delay_sla_burden", + "params": { + "region": "APAC", + "category": "B", + "month_start": 202604, + "month_end": 202606 + }, + "question": "For region APAC, category B, months 202604-202606: get the orders, count delayed orders where ship_days exceeded promised_days, find high and critical support tickets for those orders, count tickets that breach a 24-hour first-response or 120-hour resolution SLA, and report delayed_order_count + breached_ticket_count.", + "answer_format": "exact integer" +} diff --git a/harbor/tasks/v3_depth6_delay_sla_burden_t10/instruction.md b/harbor/tasks/v3_depth6_delay_sla_burden_t10/instruction.md new file mode 100644 index 0000000000000000000000000000000000000000..636a6166bc782b9f1181e9fae7d8e67063a6fdcd --- /dev/null +++ b/harbor/tasks/v3_depth6_delay_sla_burden_t10/instruction.md @@ -0,0 +1,38 @@ +# FlowBench v3 Public Task + +You are in `/app`. The file `/app/flowbench_tools.py` contains deterministic +Python tools over a synthetic business-operations dataset. Use those tools +to compute the requested value, then write the final output to +`/app/answer.txt`. + +Required output: +- `/app/answer.txt` must contain only the exact integer. +- Use integer arithmetic for currency and counts. Do not use floating point + arithmetic. + +Task: +For region APAC, category B, months 202604-202606: get the orders, count delayed orders where ship_days exceeded promised_days, find high and critical support tickets for those orders, count tickets that breach a 24-hour first-response or 120-hour resolution SLA, and report delayed_order_count + breached_ticket_count. + +Available tools: +- `region_currency(region: str) -> str`: ISO currency code for a region. +- `get_orders(region: str, category: str, month_start: int, month_end: int) -> list[int]`: Paid/refunded order ids for a region, product category, and inclusive month window. +- `order_gross_usd(order_id: int) -> int`: Gross order revenue in USD before discounts/refunds. +- `order_margin_usd(order_id: int) -> int`: Gross order margin in USD before discounts/refunds. +- `refund_usd(order_id: int) -> int`: Refund amount in USD for returned quantity on an order. +- `customer_tier(order_id: int) -> str`: Customer loyalty tier for an order. +- `apply_discount(amount_usd: int, tier: str) -> int`: Apply the order customer's loyalty discount to a USD amount. +- `net_revenue_usd(order_id: int) -> int`: Net USD revenue after loyalty discount and refunds. +- `to_local(amount_usd: int, currency: str) -> int`: Convert a USD integer amount to a local-currency integer. +- `inventory_position(product_id: int, region: str) -> int`: Available inventory position: on_hand - reserved + inbound. +- `product_lead_time(product_id: int) -> int`: Supplier lead time in days for a product. +- `product_id_for_top_seller(region: str, category: str, month_start: int, month_end: int) -> int`: Product id with the highest units sold in a region/category/month window. +- `units_sold(product_id: int, region: str, month_start: int, month_end: int) -> int`: Units sold for one product in a region and month window. +- `tickets_for_orders(order_ids: list[int], severity: str) -> list[int]`: Support ticket ids for given orders and severity. +- `ticket_order_id(ticket_id: int) -> int`: Order id associated with a support ticket. +- `sla_breached(ticket_id: int, first_response_limit_hours: int, resolution_limit_hours: int) -> bool`: Whether a support ticket breaches first-response or resolution SLA. +- `delayed_orders(order_ids: list[int]) -> list[int]`: Subset of order ids whose ship_days exceeded promised_days. +- `unique_customers(order_ids: list[int]) -> list[int]`: Unique customer ids among the given orders. +- `count_items(values: list) -> int`: Length of a list. +- `sum_values(values: list[int]) -> int`: Sum a list of integers. +- `count_true(values: list[bool]) -> int`: Count true values. +- `count_below(values: list[int], threshold: int) -> int`: Count values below a threshold. diff --git a/harbor/tasks/v3_depth6_delay_sla_burden_t10/task.toml b/harbor/tasks/v3_depth6_delay_sla_burden_t10/task.toml new file mode 100644 index 0000000000000000000000000000000000000000..edd706fc4bbda29debe31d3dce6bfe6babe5f10b --- /dev/null +++ b/harbor/tasks/v3_depth6_delay_sla_burden_t10/task.toml @@ -0,0 +1,28 @@ +schema_version = "1.3" + +[task] +name = "mindlab/v3_depth6_delay_sla_burden_t10" +description = "FlowBench v3 public task metadata. Public verifier is smoke-only." +authors = [{ name = "Mind Lab", email = "contact@mindlab.ltd" }] +keywords = ["flowbench", "tool-use", "data-orchestration", "operations"] + +[metadata] +benchmark = "FlowBench-v3" +split = "public-task-only" +category = "data-processing" +depth = 6 + +[verifier] +timeout_sec = 120.0 + +[agent] +timeout_sec = 900.0 + +[environment] +network_mode = "no-network" +build_timeout_sec = 600.0 +os = "linux" +cpus = 1 +memory_mb = 2048 +storage_mb = 2048 +gpus = 0