RingGuard Phase 3 - initial HF Spaces deploy

Graph-based fraud ring detection for Razorpay merchants.
FastAPI dashboard serving precomputed detection results.

Stack: Python 3.11, FastAPI, LightGBM, LangGraph, NetworkX, Groq, Razorpay SDK

.env.example

@@ -0,0 +1,4 @@
+# Not required for Phase 1. Needed starting Phase 2/3.
+GROQ_API_KEY=
+RAZORPAY_KEY_ID=
+RAZORPAY_KEY_SECRET=

.gitignore

@@ -0,0 +1,16 @@
+venv/
+__pycache__/
+*.pyc
+.env
+.DS_Store
+data/synthetic/graph.gpickle
+data/synthetic_train/graph.gpickle
+
+# Phase execution plans and summaries - keep local only, not in public repo
+phase-1-execution-plan.md
+phase-2-execution-plan.md
+phase-3-execution-plan.md
+phase1_summary.md
+phase2_summary.md
+phase3_summary.md
+data/synthetic_train/

Dockerfile

@@ -0,0 +1,29 @@
+FROM python:3.11-slim
+
+# Phase 3 deployment — RingGuard FastAPI dashboard
+# Razorpay actions are precomputed and baked into cases_demo.json / audit_log.jsonl.
+# The deployed container only DISPLAYS these precomputed results; it does NOT
+# re-execute Razorpay calls or LLM calls on page load (per Phase 3 Section 6.2).
+
+WORKDIR /app
+
+# Install dependencies
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Copy source code and precomputed results
+COPY agent/       agent/
+COPY api/         api/
+COPY audit/       audit/
+COPY data/        data/
+COPY graph/       graph/
+COPY integration/ integration/
+COPY scoring/     scoring/
+COPY eval/        eval/
+
+# HuggingFace Spaces expects port 7860
+ENV PORT=7860
+EXPOSE 7860
+
+# Start the FastAPI dashboard
+CMD ["uvicorn", "api.main:app", "--host", "0.0.0.0", "--port", "7860"]

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Rohit Patil
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README.md

@@ -0,0 +1,320 @@
+---
+title: RingGuard
+colorFrom: red
+colorTo: orange
+sdk: docker
+pinned: true
+license: mit
+short_description: Graph-based fraud ring detection for Razorpay merchants
+---
+# RingGuard
+
+![Track](https://img.shields.io/badge/Track-AI%20Risk%20Manager-blue) ![Event](https://img.shields.io/badge/Event-Razorpay%20AI%20Buildathon%202026-orange) ![Status](https://img.shields.io/badge/Status-Phase%203%20Complete-brightgreen) ![Python](https://img.shields.io/badge/Python-3.11-blue) ![License](https://img.shields.io/badge/License-MIT-green)
+
+**Coordinated fraud detection for Razorpay merchants — finding the rings per-order scoring can't see.**
+
+RingGuard analyzes a merchant's own order and account data to detect coordinated abuse — accounts secretly linked by shared devices, addresses, or payment instruments, working together to exploit returns, promo codes, or chargebacks. Instead of scoring one order at a time, it builds a graph of the merchant's entire transaction history, finds the structure a per-order model is blind to, and hands a human reviewer an explainable case file instead of a black-box flag. Built for Track 02 — AI Risk Manager.
+
+---
+
+## Table of Contents
+
+1. [The Problem](#the-problem)
+2. [The Solution](#the-solution)
+3. [Why This, Not Thirdwatch](#why-this-not-thirdwatch)
+4. [How It Works](#how-it-works)
+5. [AI Judgment — Where Each Tool Is Used](#ai-judgment--where-each-tool-is-used)
+6. [System Architecture](#system-architecture)
+7. [Tech Stack](#tech-stack)
+8. [Repository Structure](#repository-structure)
+9. [Getting Started](#getting-started)
+10. [Usage](#usage)
+11. [Evaluation & Metrics](#evaluation--metrics)
+12. [Known Limitations](#known-limitations)
+13. [Policy & Safety Boundaries](#policy--safety-boundaries)
+14. [Audit Trail](#audit-trail)
+15. [Roadmap](#roadmap)
+16. [Demo](#demo)
+17. [Acknowledgments](#acknowledgments)
+18. [License](#license)
+
+---
+
+## The Problem
+
+Cash-on-delivery return-to-origin rates in Indian e-commerce run as high as 30–40%, and coordinated abuse — not isolated bad actors — is a meaningful share of that loss. Fraud rings deliberately spread activity across multiple "different" accounts specifically so that any system scoring one order at a time sees N separate low-risk transactions instead of one high-risk network. Per-order risk engines, however sophisticated, are structurally blind to this by design — they were never built to look across accounts.
+
+## The Solution
+
+RingGuard is a defensive, AI-assisted investigation agent that:
+
+- **Ingests** a merchant's order, customer, device, address, and payment data (Razorpay test-mode-shaped).
+- **Builds a graph** connecting accounts through shared devices, payment instruments, addresses, and contact details.
+- **Detects communities** in that graph — clusters of accounts more connected to each other than chance would suggest.
+- **Scores each cluster** with a trained model, not hand-tuned rules, using graph structure and behavioral features.
+- **Explains the finding** in plain language via an LLM that reads only the structured evidence it's given — never free-reasoning, never inventing facts.
+- **Recommends a bounded action** — flag, hold, or escalate — gated by deterministic policy rules, with a human always in the loop for anything above the auto-hold threshold.
+
+It never bans an account or seizes funds. It flags, it explains, and it waits for a human on anything that matters.
+
+## Why This, Not Thirdwatch
+
+Razorpay has owned Thirdwatch since 2019 — a mature fraud/RTO engine scoring individual orders on roughly 300 device and behavioral signals. Razorpay's Agent Studio also ships a Dispute Responder Agent (chargeback evidence) and a Subscription Recovery Agent. None of these look *across* orders and accounts for structure — they operate at the level of one order or one case at a time.
+
+RingGuard is not a better per-order scorer. It's a different problem shape: network-level detection of coordinated behavior that is invisible when you only ever look at one order at a time. That's the entire differentiation argument, and it's intentional that it's stated here, first, rather than left for a judge to have to ask about.
+
+## How It Works
+
+```
+Synthetic order/account stream (Razorpay Orders/Payments-shaped)
+        │
+        â–¼
+Entity resolution — normalize phone/email/address, hash device + payment fields
+        │
+        â–¼
+Graph construction — Customer / Device / Address / PaymentInstrument nodes
+        │
+        â–¼
+Community detection — connected components + Louvain (deterministic, seed=42)
+        │
+        â–¼
+Feature extraction — cluster size, strong/weak link ratio, order velocity,
+                      return rate, promo usage, ₹ exposure (13 features)
+        │
+        â–¼
+LightGBM risk model — single learned probability, no hand-blended scoring
+        │
+        â–¼
+Policy gate (deterministic, v3.0)
+   ├── risk ≥ 0.85 and exposure ≤ ₹25,000 →  bounded auto-hold (test mode, Razorpay API)
+   ├── risk ≥ 0.85 and exposure > ₹25,000 →  escalate, no auto-action
+   ├── 0.5 ≤ risk < 0.85                   →  escalate to human review queue
+   └── risk < 0.5                          →  log only
+        │
+        â–¼
+Investigator agent (LangGraph + Groq LLM) — evidence bundle in, case narrative out
+        │
+        â–¼
+Razorpay test-mode action (order.create + order.edit notes)
+        │
+        â–¼
+Audit logger — every stage, every decision, timestamped (audit_log.jsonl)
+```
+
+A cluster built entirely from **weak links** (e.g., three accounts sharing only an address — plausible for a family household) is designed to score low. Distinguishing that from a **strong-link** cluster (shared device fingerprint or payment instrument) is what keeps false positives down, and it's built into the feature set from day one, not patched in afterward.
+
+## AI Judgment — Where Each Tool Is Used
+
+| Stage | Tool | Why this tool, not another |
+|---|---|---|
+| Graph construction, edge rules, policy gate | Deterministic code | Money-adjacent decisions need to be boring and auditable — no model judgment here |
+| Community detection, clustering | Classical graph algorithms (Louvain, connected components) | Structural pattern detection, not a language problem |
+| Ring risk scoring | LightGBM (supervised, tabular) | Faster, cheaper, and more auditable than an LLM for a numeric classification task |
+| Case narrative + human-readable explanation | LLM (Groq — `openai/gpt-oss-120b`) | The one place synthesizing evidence into something a reviewer can act on quickly genuinely benefits from language generation |
+| Final action | Deterministic policy gate, human-approved above threshold | The LLM never has authority to move money or take action on its own |
+
+## System Architecture
+
+```mermaid
+flowchart TD
+    A[Synthetic Order Stream] --> B[Entity Resolution]
+    B --> C[Graph Construction - NetworkX]
+    C --> D[Community Detection - Louvain]
+    D --> E[Feature Extraction]
+    E --> F[LightGBM Risk Model]
+    F --> G{Policy Gate}
+    G -->|risk >= 0.85, exposure <= 25k| H[Auto-Hold via Razorpay API]
+    G -->|risk >= 0.85, exposure > 25k| I[Escalate - Human Review]
+    G -->|0.5 <= risk < 0.85| I
+    G -->|risk < 0.5| J[Log Only]
+    H --> K[Investigator Agent - Groq LLM]
+    I --> K
+    K --> L[Case Narrative + Confidence]
+    L --> M[Audit Log + Dashboard]
+    J --> M
+```
+
+## Tech Stack
+
+| Layer | Choice |
+|---|---|
+| Language / API | Python 3.11, FastAPI |
+| Graph | NetworkX + python-louvain |
+| Scoring | LightGBM |
+| Agent orchestration | LangGraph |
+| LLM | Groq (`openai/gpt-oss-120b`, plain chat completion — no tools) |
+| Razorpay integration | Official Razorpay Python SDK v2.0.1 (test-mode API calls verified) |
+| Synthetic data | Faker |
+| Dashboard | HTML served by FastAPI |
+| Deployment | HuggingFace Spaces (Docker) |
+
+No paid dependencies anywhere in this stack.
+
+## Repository Structure
+
+```
+ringguard/
+├── README.md
+├── requirements.txt
+├── .env.example
+├── Dockerfile
+├── data/
+│   ├── generate_synthetic_data.py
+│   ├── augment_demo_dataset.py     ← Phase 3: adds RING11 + BOUNDARY01
+│   └── synthetic/
+│       ├── cases_demo.json         ← 28 cases with Razorpay action_taken
+│       └── ...
+├── graph/
+│   ├── build_graph.py
+│   └── detect_communities.py
+├── scoring/
+│   ├── features.py
+│   ├── train_model.py
+│   ├── score_demo.py
+│   └── model.lgb
+├── agent/
+│   ├── graph.py
+│   ├── investigator.py
+│   ├── policy.py
+│   └── run_demo.py
+├── integration/
+│   └── razorpay_actions.py         ← Phase 3: real test-mode Razorpay calls
+├── api/
+│   └── main.py                     ← Phase 3: FastAPI dashboard
+├── eval/
+│   └── evaluate.py
+└── audit/
+    ├── logger.py                   ← Phase 3: append-only audit log
+    ├── generate_audit_log.py
+    └── audit_log.jsonl
+```
+
+## Getting Started
+
+### Prerequisites
+
+- Python 3.11+
+- A free [Groq API key](https://console.groq.com)
+- Razorpay test-mode API keys (`rzp_test_...`) from the [Razorpay Dashboard](https://dashboard.razorpay.com)
+
+### Installation
+
+```bash
+git clone https://github.com/rohitdecodes/ringguard.git
+cd ringguard
+python -m venv venv
+source venv/bin/activate    # Windows: venv\Scripts\activate
+pip install -r requirements.txt
+cp .env.example .env        # then fill in GROQ_API_KEY, RAZORPAY_KEY_ID, RAZORPAY_KEY_SECRET
+```
+
+## Usage
+
+```bash
+# 1. (One-time) Augment the demo dataset with RING11 + BOUNDARY01
+python data/augment_demo_dataset.py
+
+# 2. Build the graph and detect candidate clusters
+python graph/build_graph.py --data-dir data/synthetic
+python graph/detect_communities.py --data-dir data/synthetic
+
+# 3. Score clusters and run the LangGraph agent
+python scoring/score_demo.py
+python agent/run_demo.py
+
+# 4. Execute Razorpay test-mode actions
+python integration/razorpay_actions.py
+
+# 5. Generate the audit log
+python audit/generate_audit_log.py
+
+# 6. Launch the dashboard
+uvicorn api.main:app --reload --port 7860
+# open http://localhost:7860
+```
+
+## Evaluation & Metrics
+
+Measured on a held-out split at the **cluster level** (no ring's accounts appear in both train and test). Fixed seed=42, stratified 80/20 split.
+
+| Metric | Value |
+|---|---|
+| **Precision** (ring detection) | **1.0000** |
+| **Recall** (ring detection) | **1.0000** |
+| **F1** | **1.0000** |
+| False-positive cost (₹, held orders wrongly flagged) | **₹0.00** |
+| End-to-end latency (feature extraction + inference) | **0.043 ms/cluster** |
+
+Holdout: 22 clusters (12 positive fraud rings, 10 negative benign clusters).
+
+### Policy action breakdown on held-out set
+| Action | Count |
+|---|---|
+| auto_hold | 0 |
+| escalate | 12 |
+| log_only | 10 |
+
+### Demo dataset (28 clusters, post Phase 3 augmentation)
+| Action | Count |
+|---|---|
+| auto_hold | 1 (RING11) |
+| escalate | 10 |
+| log_only | 17 (includes BOUNDARY01) |
+
+## Known Limitations
+
+**Read this before evaluating the precision/recall numbers above.**
+
+The current P=R=F1=1.0 figures reflect that the synthetic training data does not yet contain genuinely ambiguous or partially-ring-like examples. The model has not been tested on a hard case — one where real human experts would disagree. Specifically:
+
+1. **Training distribution boundary:** The Phase 2 model uses `avg_order_amount` as its strongest feature (importance=145 vs. 120 for chargeback_rate). Positive training examples have `avg_order_amount ≥ ₹3,249`; all negatives are below ₹3,171. This creates a near-hard threshold. Real rings could fall on either side of it.
+
+2. **Auto-hold threshold calibration (Phase 3 change):** The original ₹5,000 auto-hold exposure cap is mathematically incompatible with the training distribution — a 3-member ring with just enough orders to form strong links (2 per customer × ₹3,300 each) totals ≈₹19,800. The threshold was updated to ₹25,000 in Phase 3 (v3.0). This is a legitimate calibration, but it means the original threshold was never validated against real data either.
+
+3. **Razorpay MCP server not available:** The `@razorpay/mcp-server` npm package returns 404 on the npm registry. Razorpay actions were executed via the official Python SDK (`razorpay==2.0.1`), which calls the same REST API. All calls are real, test-mode, and verified — but the MCP-specific tooling described in the original plan is not available.
+
+4. **Synthetic data only:** All evaluation is against Faker-generated data. Real merchants have messier entity resolution challenges, more diverse ring structures, and ambiguous cases the current training data cannot represent. Real-world P/R/F1 should be expected to be significantly lower.
+
+5. **`BOUNDARY01` detection behavior:** This deliberately ambiguous cluster (3 accounts sharing one address, no chargebacks, low return rate) scored `risk=0.0035` and was correctly routed to `log_only` — not `auto_hold`. The model correctly declined to flag it. However, this isn't a hard-won result from a genuinely difficult case; it's the model doing nothing to an input that looks entirely unlike its training positives.
+
+Real-world performance validation requires labeled production data. The current metrics are a correctness proof-of-concept, not a deployment readiness certificate.
+
+## Policy & Safety Boundaries
+
+RingGuard is **defense-only**, per Track 02's rules:
+
+- It can flag, hold, and escalate. It cannot ban, seize funds, or take any irreversible action.
+- Auto-hold only fires below ₹25,000 total cluster exposure (Phase 3 v3.0 calibration) and only at high confidence (risk ≥ 0.85).
+- Anything above that threshold, or in the mid-confidence band, goes to a human review queue — never auto-actioned.
+- The LLM has no authority to trigger an action. It explains a decision the policy gate already made.
+- The LLM (`openai/gpt-oss-120b`) is called as a plain chat completion with no `tools` parameter — it has no web-search or code-execution access. It only knows what's in the evidence bundle.
+
+## Audit Trail
+
+Every case logs: `case_id, timestamp, cluster_members, evidence_bundle, graph_score_inputs, lightgbm_probability, llm_narrative, llm_confidence_statement, policy_decision, action_taken, human_reviewer (if escalated), outcome, reason_if_not_auto_actioned`.
+
+Available as `audit/audit_log.jsonl` (one JSON object per line, append-only). Every case in `cases_demo.json` has a matching audit entry, including log_only cases. The dashboard at `GET /cases/{case_id}/detail` shows the full audit chain for each case.
+
+## Roadmap
+
+Building in three phases. Detailed, task-level execution plans for each phase — written for agentic AI coding execution — are maintained separately.
+
+- ✅ **Phase 1 — Detection Core:** synthetic data generation, entity resolution, graph construction, community detection. Output: a working, deterministic pipeline that turns raw order data into flagged candidate clusters.
+- ✅ **Phase 2 — Intelligence Layer:** feature extraction, LightGBM risk model, held-out evaluation, LangGraph agent orchestration, Investigator LLM node, policy gate. Output: scored, explained, policy-gated cases.
+- ✅ **Phase 3 — Integration & Submission:** dataset augmentation (RING11 + BOUNDARY01), model swap (`openai/gpt-oss-120b`), Razorpay test-mode integration, audit trail, FastAPI dashboard, deployment, final evaluation, demo, submission.
+
+## Demo
+
+*Live deployment: HuggingFace Spaces — see phase3_summary.md for URL.*
+
+*Pitch video: to be recorded.*
+
+## Acknowledgments
+
+Built with [LangGraph](https://langchain-ai.github.io/langgraph/), [Groq](https://groq.com) (`openai/gpt-oss-120b`), [LightGBM](https://lightgbm.readthedocs.io/), [NetworkX](https://networkx.org/), and the [Razorpay Python SDK](https://github.com/razorpay/razorpay-python).
+
+## License
+
+MIT — see [LICENSE](LICENSE).
+
+

agent/.gitkeep

@@ -0,0 +1 @@
+

agent/graph.py

@@ -0,0 +1,265 @@
+"""
+graph.py
+--------
+LangGraph orchestration for RingGuard fraud investigation.
+
+State: RingGuardState
+  cluster           — the candidate cluster record
+  features          — extracted feature dict
+  risk_probability  — model score
+  policy_decision   — "auto_hold" | "escalate" | "log_only"
+  llm_narrative     — LLM explanation (None if log_only or if GROQ unavailable)
+  llm_confidence    — LLM confidence statement (None if log_only)
+
+Flow:
+  score_node → policy_node → [conditional] investigator_node → END
+  (investigator_node is SKIPPED for log_only decisions)
+
+Usage:
+    from agent.graph import build_graph, run_cluster
+    results = run_cluster(cluster, features, risk_probability)
+"""
+
+import os
+import sys
+from typing import Optional, TypedDict
+
+from langgraph.graph import StateGraph, END
+
+BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+sys.path.insert(0, BASE_DIR)
+
+from agent.policy import decide as policy_decide
+from agent.investigator import investigate
+
+# Audit logger — imported lazily to keep graph.py usable without audit module
+_AUDIT_ENABLED = True
+try:
+    from audit.logger import log_case as _log_case
+except ImportError:
+    _AUDIT_ENABLED = False
+    _log_case = None
+
+
+# ---------------------------------------------------------------------------
+# State definition
+# ---------------------------------------------------------------------------
+
+class RingGuardState(TypedDict):
+    cluster: dict
+    features: dict
+    risk_probability: float
+    policy_decision: Optional[str]
+    llm_narrative: Optional[str]
+    llm_confidence: Optional[str]
+    grounding_ok: Optional[bool]
+
+
+# ---------------------------------------------------------------------------
+# Nodes
+# ---------------------------------------------------------------------------
+
+def score_node(state: RingGuardState) -> RingGuardState:
+    """
+    Score node: wraps the already-computed risk_probability from the model.
+    The probability was computed externally (by scoring/score_demo.py or live
+    inference) and passed in via state — this node simply forwards it.
+    """
+    # risk_probability is already set in state by the caller
+    return state
+
+
+def policy_node(state: RingGuardState) -> RingGuardState:
+    """
+    Policy gate: deterministic decision based on risk_probability + exposure.
+    Exposure = total_amount from features (Phase 3 will use real Razorpay data).
+    """
+    risk_prob = state["risk_probability"]
+    exposure  = state["features"].get("total_amount", 0.0)
+    decision  = policy_decide(risk_prob, exposure)
+
+    return {**state, "policy_decision": decision}
+
+
+def investigator_node(state: RingGuardState) -> RingGuardState:
+    """
+    LLM investigator: generates a narrative for high-risk clusters.
+    Only called when policy_decision is 'auto_hold' or 'escalate'.
+    Catches errors so a single LLM failure does not abort the whole pipeline.
+    """
+    try:
+        result = investigate(
+            cluster=state["cluster"],
+            features=state["features"],
+            risk_probability=state["risk_probability"],
+            policy_decision=state["policy_decision"],
+        )
+        return {
+            **state,
+            "llm_narrative": result["narrative"],
+            "llm_confidence": result["confidence_statement"],
+            "grounding_ok": result["grounding_ok"],
+        }
+    except Exception as e:
+        return {
+            **state,
+            "llm_narrative": f"[LLM_ERROR] {e}",
+            "llm_confidence": None,
+            "grounding_ok": None,
+        }
+
+
+# ---------------------------------------------------------------------------
+# Conditional routing
+# ---------------------------------------------------------------------------
+
+def should_investigate(state: RingGuardState) -> str:
+    """
+    Route to investigator only for auto_hold or escalate decisions.
+    Skip for log_only — no reason to spend an LLM call on it.
+    """
+    if state.get("policy_decision") in ("auto_hold", "escalate"):
+        return "investigate"
+    return "skip"
+
+
+# ---------------------------------------------------------------------------
+# Graph construction
+# ---------------------------------------------------------------------------
+
+def build_graph():
+    """Build and compile the RingGuard LangGraph state machine."""
+    builder = StateGraph(RingGuardState)
+
+    builder.add_node("score_node",       score_node)
+    builder.add_node("policy_node",      policy_node)
+    builder.add_node("investigator_node", investigator_node)
+
+    builder.set_entry_point("score_node")
+
+    builder.add_edge("score_node", "policy_node")
+
+    builder.add_conditional_edges(
+        "policy_node",
+        should_investigate,
+        {
+            "investigate": "investigator_node",
+            "skip":        END,
+        },
+    )
+
+    builder.add_edge("investigator_node", END)
+
+    return builder.compile()
+
+
+# ---------------------------------------------------------------------------
+# Convenience runner
+# ---------------------------------------------------------------------------
+
+def run_cluster(
+    cluster: dict,
+    features: dict,
+    risk_probability: float,
+    case_id: str = None,
+    action_taken: dict = None,
+) -> dict:
+    """
+    Run the full pipeline for a single cluster.
+
+    Returns final state dict with policy_decision, llm_narrative, etc.
+    Catches LLM errors and records them in llm_narrative rather than crashing.
+
+    Parameters
+    ----------
+    cluster        : candidate cluster record
+    features       : extracted feature dict
+    risk_probability : model score
+    case_id        : optional case ID for audit logging (e.g. "CASE08")
+    action_taken   : optional Razorpay action dict (populated by Section 3)
+    """
+    graph = build_graph()
+
+    initial_state: RingGuardState = {
+        "cluster":          cluster,
+        "features":         features,
+        "risk_probability": risk_probability,
+        "policy_decision":  None,
+        "llm_narrative":    None,
+        "llm_confidence":   None,
+        "grounding_ok":     None,
+    }
+
+    try:
+        final_state = graph.invoke(initial_state)
+    except RuntimeError as e:
+        # LLM failures (missing key, parse errors) are caught here.
+        # The policy decision was already set before the LLM call.
+        # We log the error rather than crashing the whole pipeline.
+        final_state = {
+            **initial_state,
+            "policy_decision": policy_decide(risk_probability, features.get("total_amount", 0.0)),
+            "llm_narrative":   f"[LLM_ERROR] {e}",
+            "llm_confidence":  None,
+            "grounding_ok":    None,
+        }
+
+    result = dict(final_state)
+
+    # Phase 3 Section 4.2 — write audit log entry for every run
+    if _AUDIT_ENABLED and _log_case is not None:
+        try:
+            from datetime import datetime, timezone
+            audit_case = {
+                "case_id":                   case_id or cluster.get("cluster_id", "?"),
+                "cluster_id":                cluster.get("cluster_id"),
+                "customer_ids":              cluster.get("customer_ids", []),
+                "risk_probability":          risk_probability,
+                "policy_decision":           result.get("policy_decision"),
+                "llm_narrative":             result.get("llm_narrative"),
+                "llm_confidence_statement":  result.get("llm_confidence"),
+                "action_taken":              action_taken,
+                "timestamp":                 datetime.now(timezone.utc).isoformat(),
+            }
+            _log_case(audit_case, features)
+        except Exception:
+            pass  # never crash the pipeline due to audit errors
+
+    return result
+
+
+# ---------------------------------------------------------------------------
+# CLI test — run on a single demo cluster
+# ---------------------------------------------------------------------------
+
+if __name__ == "__main__":
+    import json
+
+    demo_dir = os.path.join(BASE_DIR, "data", "synthetic")
+    scored_path  = os.path.join(demo_dir, "scored_clusters.json")
+    orders_path  = os.path.join(demo_dir, "orders.json")
+    graph_path   = os.path.join(demo_dir, "graph.gpickle")
+
+    with open(scored_path) as f:
+        scored = json.load(f)
+
+    with open(orders_path) as f:
+        orders = json.load(f)
+
+    from scoring.features import extract_features, _load_graph
+    graph_obj = _load_graph(graph_path) if os.path.exists(graph_path) else None
+
+    # Use the first high-risk cluster as a test
+    test_cluster = scored[0]
+    features = extract_features(test_cluster, orders, graph=graph_obj)
+    risk_prob = test_cluster["risk_probability"]
+
+    print(f"Testing graph on {test_cluster['cluster_id']} "
+          f"(risk={risk_prob:.4f})...")
+
+    result = run_cluster(test_cluster, features, risk_prob)
+
+    print(f"  Policy decision : {result['policy_decision']}")
+    print(f"  LLM narrative   : {result['llm_narrative']}")
+    print(f"  LLM confidence  : {result['llm_confidence']}")
+    print(f"  Grounding OK    : {result['grounding_ok']}")

agent/investigator.py

@@ -0,0 +1,264 @@
+"""
+investigator.py
+---------------
+LLM investigator node for RingGuard.
+
+Calls Groq (llama-3.3-70b-versatile) to generate a short narrative explaining
+why a cluster was flagged. Only called for auto_hold and escalate decisions
+(not log_only — no reason to spend an LLM call on something nobody will review).
+
+Requires GROQ_API_KEY in environment (from .env).
+If GROQ_API_KEY is not set, raises RuntimeError rather than silently continuing.
+
+Anti-hallucination boundary:
+  - The LLM only sees the evidence bundle (aggregated features), never raw
+    order-level or PII data.
+  - System prompt instructs the model to ONLY reference facts present in the
+    evidence bundle.
+  - Any customer ID mentioned in the narrative must be a member of the bundle's
+    customer_ids list (checked by run_grounding_check).
+  - Output validated against InvestigatorOutput Pydantic schema; retried once
+    with stricter reminder if parse fails.
+
+Usage:
+    from agent.investigator import investigate, run_grounding_check
+    result = investigate(cluster, features, risk_probability, policy_decision)
+    is_clean = run_grounding_check(result["narrative"], cluster["customer_ids"])
+"""
+
+import json
+import os
+import re
+import sys
+
+from dotenv import load_dotenv
+from pydantic import BaseModel, ValidationError
+
+load_dotenv()
+
+BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+sys.path.insert(0, BASE_DIR)
+
+
+# ---------------------------------------------------------------------------
+# Output schema
+# ---------------------------------------------------------------------------
+
+class InvestigatorOutput(BaseModel):
+    narrative: str
+    confidence_statement: str
+
+
+# ---------------------------------------------------------------------------
+# Evidence bundle builder
+# ---------------------------------------------------------------------------
+
+def build_evidence_bundle(
+    cluster: dict,
+    features: dict,
+    risk_probability: float,
+    policy_decision: str,
+) -> dict:
+    """
+    Build the evidence bundle passed to the LLM.
+    Contains only the aggregated facts listed in spec 10.3 — no raw PII.
+    """
+    return {
+        "customer_ids": cluster["customer_ids"],
+        "size": cluster["size"],
+        "strong_link_count": cluster["strong_link_count"],
+        "weak_link_count": cluster["weak_link_count"],
+        "formation_type": cluster.get("formation_type", "unknown"),
+        "return_rate": features.get("return_rate", 0.0),
+        "chargeback_rate": features.get("chargeback_rate", 0.0),
+        "shared_coupon_count": features.get("shared_coupon_count", 0),
+        "total_amount": features.get("total_amount", 0.0),
+        "risk_probability": risk_probability,
+        "policy_decision": policy_decision,
+    }
+
+
+# ---------------------------------------------------------------------------
+# System prompt
+# ---------------------------------------------------------------------------
+
+SYSTEM_PROMPT = """You are a fraud analyst assistant for RingGuard, a fraud detection system.
+
+You will receive a JSON evidence bundle describing a cluster of customers flagged as high-risk.
+
+Your task is to write a short fraud narrative and a confidence statement.
+
+STRICT RULES — violating any of these makes your response unusable:
+1. ONLY reference facts that are explicitly present in the evidence bundle.
+2. Do NOT invent customer names, order details, locations, or any fact not in the bundle.
+3. Do NOT mention any customer_id that is not in the bundle's customer_ids list.
+4. Do NOT recommend an action — the policy gate has already decided that.
+5. Do NOT speculate beyond what the evidence shows.
+6. Keep the narrative to 2-4 sentences. Keep the confidence statement to 1 sentence.
+
+Respond with ONLY a valid JSON object in this exact format (no markdown, no preamble):
+{"narrative": "...", "confidence_statement": "..."}"""
+
+
+# ---------------------------------------------------------------------------
+# Main investigate function
+# ---------------------------------------------------------------------------
+
+def _call_groq(evidence_bundle: dict, retry_strict: bool = False) -> str:
+    """
+    Call Groq API and return the raw response text. Retries on rate limits.
+
+    Phase 3 model selection:
+      - groq/compound-mini was used in Phase 2 as an interim substitute.
+      - Phase 3 plan requires replacing it with a plain, non-agentic model.
+      - Available models checked at execution time: openai/gpt-oss-120b (works,
+        returns valid JSON directly), qwen/qwen3.6-27b (uses extended thinking,
+        requires higher token budget to clear <think> block before JSON).
+      - Selected: openai/gpt-oss-120b — standard chat completion, no tool calls,
+        deterministic JSON output, sufficient token budget at 500 tokens.
+      - No `tools` parameter is passed anywhere in this function (hard requirement
+        from Phase 3 plan Section 2.2).
+    """
+    import time
+    from groq import Groq, RateLimitError
+
+    api_key = os.environ.get("GROQ_API_KEY", "")
+    if not api_key:
+        raise RuntimeError(
+            "GROQ_API_KEY is not set in environment. "
+            "Add it to .env and reload before running the investigator."
+        )
+
+    client = Groq(api_key=api_key)
+
+    user_message = json.dumps(evidence_bundle, indent=2)
+    if retry_strict:
+        user_message += (
+            "\n\nIMPORTANT: Your previous response did not parse as valid JSON. "
+            "Respond ONLY with a JSON object: "
+            '{"narrative": "...", "confidence_statement": "..."} — no other text.'
+        )
+
+    # Phase 3: openai/gpt-oss-120b — plain chat completion, no tools parameter.
+    # Configurable via GROQ_MODEL env var for future model swaps.
+    # groq/compound-mini (Phase 2 default) is not an acceptable substitute here —
+    # it is an agentic model with tool/web-search access, violating Plan Section 2.2.
+    model_name = os.environ.get("GROQ_MODEL", "openai/gpt-oss-120b")
+
+    for attempt in range(3):
+        try:
+            response = client.chat.completions.create(
+                model=model_name,
+                messages=[
+                    {"role": "system", "content": SYSTEM_PROMPT},
+                    {"role": "user", "content": user_message},
+                ],
+                temperature=0.1,  # low temperature for deterministic output
+                max_tokens=500,   # no `tools` parameter — plain chat completion
+            )
+            return response.choices[0].message.content.strip()
+        except RateLimitError as e:
+            if attempt < 2:
+                wait = 30 * (attempt + 1)
+                print(f"\n  [rate_limit] waiting {wait}s before retry {attempt+2}/3...")
+                time.sleep(wait)
+            else:
+                raise
+
+
+def _parse_response(raw: str) -> InvestigatorOutput:
+    """Parse and validate the LLM response against InvestigatorOutput schema."""
+    # Strip any accidental markdown fencing
+    raw = raw.strip()
+    if raw.startswith("```"):
+        raw = re.sub(r"^```[a-z]*\n?", "", raw)
+        raw = re.sub(r"\n?```$", "", raw)
+        raw = raw.strip()
+
+    data = json.loads(raw)
+    return InvestigatorOutput(**data)
+
+
+def investigate(
+    cluster: dict,
+    features: dict,
+    risk_probability: float,
+    policy_decision: str,
+) -> dict:
+    """
+    Call the LLM investigator for a flagged cluster.
+
+    Parameters
+    ----------
+    cluster : dict
+        Candidate cluster record (from candidate_clusters.json or scored_clusters.json).
+    features : dict
+        Feature dict for this cluster (from scoring/features.py).
+    risk_probability : float
+        Risk probability from the model.
+    policy_decision : str
+        Policy decision (must be "auto_hold" or "escalate" — callers should not
+        call this for "log_only" decisions).
+
+    Returns
+    -------
+    dict with keys: narrative, confidence_statement, grounding_ok, raw_response
+    """
+    evidence = build_evidence_bundle(cluster, features, risk_probability, policy_decision)
+
+    # First attempt
+    raw = _call_groq(evidence)
+    parse_error = None
+    output = None
+
+    try:
+        output = _parse_response(raw)
+    except (json.JSONDecodeError, ValidationError, Exception) as e:
+        parse_error = str(e)
+
+    if output is None:
+        # Retry once with stricter reminder (spec 10.5)
+        raw2 = _call_groq(evidence, retry_strict=True)
+        try:
+            output = _parse_response(raw2)
+            raw = raw2  # use the successful response
+        except (json.JSONDecodeError, ValidationError, Exception) as e2:
+            # Both attempts failed — log and raise
+            raise RuntimeError(
+                f"LLM response failed to parse after 2 attempts. "
+                f"First error: {parse_error}. Second error: {e2}. "
+                f"Raw response (last): {raw2}"
+            )
+
+    # 10.6 — grounding check: no customer ID in narrative that isn't in the bundle
+    grounding_ok = run_grounding_check(output.narrative, evidence["customer_ids"])
+
+    return {
+        "narrative": output.narrative,
+        "confidence_statement": output.confidence_statement,
+        "grounding_ok": grounding_ok,
+        "raw_response": raw,
+        "evidence_bundle": evidence,
+    }
+
+
+# ---------------------------------------------------------------------------
+# 10.6 — Grounding check
+# ---------------------------------------------------------------------------
+
+def run_grounding_check(narrative: str, customer_ids: list[str]) -> bool:
+    """
+    Check that any customer ID (CUSTXXXX pattern) mentioned in the narrative
+    is a member of customer_ids.
+
+    Returns True if the narrative is grounded (no hallucinated IDs).
+    Returns False if the narrative references a customer ID not in the bundle.
+    """
+    # Find all CUST-prefixed IDs mentioned in the narrative
+    mentioned_ids = set(re.findall(r"CUST\d+", narrative))
+    allowed_ids = set(customer_ids)
+
+    hallucinated = mentioned_ids - allowed_ids
+    if hallucinated:
+        return False
+    return True

agent/policy.py

@@ -0,0 +1,65 @@
+"""
+policy.py
+---------
+Deterministic policy gate for RingGuard.
+
+Pure function, no LLM calls, no side effects.
+This is the ONLY place in the system that decides whether a money-adjacent
+action is triggered — it stays boring and deterministic on purpose.
+
+Phase 3 threshold calibration (v3.0):
+  The original Phase 2 auto_hold threshold of ≤₹5,000 was found to be
+  incompatible with the trained model's scoring range during Phase 3 Section 1.
+  The Phase 2 model's minimum positive avg_order_amount is ₹3,249; a 3-member
+  ring with 6 orders at that level produces ≈₹19,800 total — above the old cap.
+  Updated to ≤₹25,000 to allow small-cell rings with realistic amounts through.
+  Documented in phase3_summary.md under Key Decisions.
+
+Thresholds (Phase 3, v3.0):
+  risk >= 0.85 and exposure <= 25000 →  auto_hold
+  risk >= 0.85 and exposure >  25000 →  escalate
+  0.50 <= risk < 0.85               →  escalate
+  risk < 0.50                       →  log_only
+
+Usage:
+    from agent.policy import decide
+    action = decide(risk_probability=0.91, exposure_rupees=19800.0)
+    # → "auto_hold"
+"""
+
+
+def decide(risk_probability: float, exposure_rupees: float) -> str:
+    """
+    Parameters
+    ----------
+    risk_probability : float
+        Model output probability in [0, 1].
+    exposure_rupees : float
+        Total order amount across the cluster (from features.total_amount).
+
+    Returns
+    -------
+    str — one of: "auto_hold", "escalate", "log_only"
+    """
+    if risk_probability >= 0.85 and exposure_rupees <= 25000:
+        return "auto_hold"
+    if risk_probability >= 0.85 and exposure_rupees > 25000:
+        return "escalate"
+    if 0.5 <= risk_probability < 0.85:
+        return "escalate"
+    return "log_only"
+
+
+# ---------------------------------------------------------------------------
+# Inline assertions — the money-decision boundary warrants a few spot checks
+# ---------------------------------------------------------------------------
+if __name__ == "__main__":
+    assert decide(0.91, 3000.0)  == "auto_hold", "high risk, low exposure"
+    assert decide(0.91, 30000.0) == "escalate",  "high risk, high exposure"
+    assert decide(0.85, 25000.0) == "auto_hold", "boundary: exactly 0.85, exactly 25000"
+    assert decide(0.85, 25001.0) == "escalate",  "boundary: exactly 0.85, just over 25000"
+    assert decide(0.70, 0.0)     == "escalate",  "medium risk"
+    assert decide(0.50, 9999.0)  == "escalate",  "lower boundary medium risk"
+    assert decide(0.49, 9999.0)  == "log_only",  "below medium threshold"
+    assert decide(0.0,  0.0)     == "log_only",  "zero risk"
+    print("All policy assertions passed.")

agent/run_demo.py

@@ -0,0 +1,121 @@
+"""
+run_demo.py
+-----------
+End-to-end demo run: score every cluster in data/synthetic/scored_clusters.json
+through the full LangGraph pipeline and save results to
+data/synthetic/cases_demo.json.
+
+Usage:
+    python agent/run_demo.py
+"""
+
+import json
+import os
+import sys
+import time
+from datetime import datetime, timezone
+
+BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+sys.path.insert(0, BASE_DIR)
+
+from dotenv import load_dotenv
+load_dotenv()
+
+from scoring.features import extract_features, _load_json, _load_graph, FEATURE_COLUMNS
+from agent.graph import run_cluster
+from agent.investigator import run_grounding_check
+
+
+DEMO_DIR  = os.path.join(BASE_DIR, "data", "synthetic")
+OUT_PATH  = os.path.join(DEMO_DIR, "cases_demo.json")
+
+
+def run_demo():
+    scored_path = os.path.join(DEMO_DIR, "scored_clusters.json")
+    orders_path = os.path.join(DEMO_DIR, "orders.json")
+    graph_path  = os.path.join(DEMO_DIR, "graph.gpickle")
+
+    print(f"Loading demo data from {DEMO_DIR}...")
+    scored  = _load_json(scored_path)
+    orders  = _load_json(orders_path)
+    graph   = _load_graph(graph_path) if os.path.exists(graph_path) else None
+
+    print(f"  {len(scored)} clusters to process\n")
+
+    cases = []
+    grounding_failures = []
+
+    for idx, cluster in enumerate(scored, start=1):
+        cid   = cluster["cluster_id"]
+        risk  = cluster["risk_probability"]
+        print(f"[{idx:02d}/{len(scored)}] {cid}  risk={risk:.4f}  formation={cluster.get('formation_type','?')}", end="  ", flush=True)
+
+        features = extract_features(cluster, orders, graph=graph)
+        result   = run_cluster(cluster, features, risk)
+
+        decision  = result["policy_decision"]
+        narrative = result["llm_narrative"]
+        confidence = result["llm_confidence"]
+        grounding  = result.get("grounding_ok")
+
+        print(f"-> {decision}")
+
+        # Small sleep between LLM calls to stay within free-tier TPM limits
+        # (only needed for escalate/auto_hold cases that call the LLM)
+        if decision in ("auto_hold", "escalate"):
+            time.sleep(3)
+
+        # 10.6 — grounding check for every narrative that was generated
+        if narrative and not narrative.startswith("[LLM_ERROR]"):
+            gok = run_grounding_check(narrative, cluster["customer_ids"])
+            if not gok:
+                grounding_failures.append({
+                    "cluster_id": cid,
+                    "narrative": narrative,
+                    "customer_ids": cluster["customer_ids"],
+                })
+                print(f"  [GROUNDING FAIL] narrative references customer ID not in bundle!")
+
+        case = {
+            "case_id":                 f"CASE{idx:02d}",
+            "cluster_id":              cid,
+            "customer_ids":            cluster["customer_ids"],
+            "risk_probability":        risk,
+            "policy_decision":         decision,
+            "llm_narrative":           narrative,
+            "llm_confidence_statement": confidence,
+            "action_taken":            None,   # Phase 3 populates via Razorpay MCP
+            "timestamp":               datetime.now(timezone.utc).isoformat(),
+        }
+        cases.append(case)
+
+    # Write cases_demo.json
+    with open(OUT_PATH, "w", encoding="utf-8") as f:
+        json.dump(cases, f, indent=2, ensure_ascii=False)
+    print(f"\nSaved {len(cases)} cases to {OUT_PATH}")
+
+    # Summary
+    from collections import Counter
+    decision_counts = Counter(c["policy_decision"] for c in cases)
+    llm_called      = sum(1 for c in cases if c["llm_narrative"] and not c["llm_narrative"].startswith("[LLM_ERROR]"))
+    llm_errors      = sum(1 for c in cases if c["llm_narrative"] and c["llm_narrative"].startswith("[LLM_ERROR]"))
+
+    print("\n========== Demo Run Summary ==========")
+    print(f"  Total cases          : {len(cases)}")
+    for action, count in sorted(decision_counts.items()):
+        print(f"  {action:15s}      : {count}")
+    print(f"  LLM calls made       : {llm_called}")
+    print(f"  LLM errors           : {llm_errors}")
+    print(f"  Grounding failures   : {len(grounding_failures)}")
+    print("=======================================")
+
+    if grounding_failures:
+        print("\n[WARNING] Grounding failures detected:")
+        for gf in grounding_failures:
+            print(f"  {gf['cluster_id']}: {gf['narrative']}")
+
+    return cases, grounding_failures, decision_counts
+
+
+if __name__ == "__main__":
+    run_demo()

api/.gitkeep

@@ -0,0 +1 @@
+

api/main.py

@@ -0,0 +1,357 @@
+"""
+main.py
+-------
+Phase 3 Section 5 — FastAPI dashboard for RingGuard.
+
+Endpoints:
+  GET /                  — HTML dashboard: table of all cases
+  GET /cases             — JSON list of all cases (from cases_demo.json)
+  GET /cases/{case_id}   — JSON detail for one case + full audit chain
+  GET /health            — health check
+
+The dashboard reads directly from cases_demo.json and audit_log.jsonl.
+No database, no auth — demo only.
+
+Usage:
+    uvicorn api.main:app --reload --port 7860
+"""
+
+import json
+import os
+from typing import Optional
+
+from fastapi import FastAPI, HTTPException
+from fastapi.responses import HTMLResponse, JSONResponse
+
+BASE_DIR    = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+CASES_PATH  = os.path.join(BASE_DIR, "data", "synthetic", "cases_demo.json")
+AUDIT_PATH  = os.path.join(BASE_DIR, "audit", "audit_log.jsonl")
+
+app = FastAPI(
+    title="RingGuard Dashboard",
+    description="Phase 3 — Fraud ring detection results and audit trail",
+    version="3.0.0",
+)
+
+
+# ---------------------------------------------------------------------------
+# Data loaders
+# ---------------------------------------------------------------------------
+
+def _load_cases() -> list:
+    if not os.path.exists(CASES_PATH):
+        return []
+    with open(CASES_PATH, encoding="utf-8") as f:
+        return json.load(f)
+
+
+def _load_audit() -> list:
+    if not os.path.exists(AUDIT_PATH):
+        return []
+    records = []
+    with open(AUDIT_PATH, encoding="utf-8") as f:
+        for line in f:
+            line = line.strip()
+            if line:
+                records.append(json.loads(line))
+    return records
+
+
+# ---------------------------------------------------------------------------
+# API endpoints
+# ---------------------------------------------------------------------------
+
+@app.get("/health")
+def health():
+    return {"status": "ok", "version": "3.0.0"}
+
+
+@app.get("/cases", response_class=JSONResponse)
+def get_cases():
+    """Return all cases as a JSON list."""
+    cases = _load_cases()
+    # Return a summary view (without full narrative for list)
+    summary = []
+    for c in cases:
+        summary.append({
+            "case_id":        c.get("case_id"),
+            "cluster_id":     c.get("cluster_id"),
+            "cluster_size":   len(c.get("customer_ids", [])),
+            "risk_score":     round(c.get("risk_probability", 0), 4),
+            "policy_decision": c.get("policy_decision"),
+            "action_taken":   bool(c.get("action_taken") and c["action_taken"].get("order_id")),
+            "timestamp":      c.get("timestamp"),
+        })
+    return summary
+
+
+@app.get("/cases/{case_id}", response_class=JSONResponse)
+def get_case_detail(case_id: str):
+    """Return full detail for one case, including audit chain."""
+    cases = _load_cases()
+    case = next((c for c in cases if c.get("case_id") == case_id), None)
+    if not case:
+        raise HTTPException(status_code=404, detail=f"Case {case_id} not found")
+
+    # Find audit records for this case
+    audit_records = [r for r in _load_audit() if r.get("case_id") == case_id]
+
+    return {
+        "case": case,
+        "audit_chain": audit_records,
+    }
+
+
+# ---------------------------------------------------------------------------
+# HTML dashboard
+# ---------------------------------------------------------------------------
+
+def _decision_badge(decision: str) -> str:
+    colours = {
+        "auto_hold": "#c0392b",
+        "escalate":  "#e67e22",
+        "log_only":  "#27ae60",
+    }
+    colour = colours.get(decision, "#7f8c8d")
+    return (
+        f'<span style="background:{colour};color:#fff;padding:2px 8px;'
+        f'border-radius:4px;font-size:11px;font-weight:600;">'
+        f'{decision.upper()}</span>'
+    )
+
+
+def _build_dashboard(cases: list) -> str:
+    rows = ""
+    for c in cases:
+        decision = c.get("policy_decision", "")
+        badge    = _decision_badge(decision)
+        risk     = round(c.get("risk_probability", 0) * 100, 1)
+        cluster  = c.get("cluster_id", "")
+        case_id  = c.get("case_id", "")
+        size     = len(c.get("customer_ids", []))
+        action   = c.get("action_taken") or {}
+        order_id = action.get("order_id") or "—"
+        verified = "Yes" if action.get("notes_verified") else "—"
+
+        rows += (
+            f'<tr onclick="window.location=\'/cases/{case_id}\'" style="cursor:pointer">'
+            f'<td>{case_id}</td>'
+            f'<td>{cluster}</td>'
+            f'<td style="text-align:center">{size}</td>'
+            f'<td style="text-align:right">{risk}%</td>'
+            f'<td>{badge}</td>'
+            f'<td style="font-size:11px;color:#57606a">{order_id}</td>'
+            f'<td style="text-align:center">{verified}</td>'
+            f'</tr>\n'
+        )
+
+    auto_holds = sum(1 for c in cases if c.get("policy_decision") == "auto_hold")
+    escalates  = sum(1 for c in cases if c.get("policy_decision") == "escalate")
+    log_only   = sum(1 for c in cases if c.get("policy_decision") == "log_only")
+
+    return f"""<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width, initial-scale=1">
+<title>RingGuard — Fraud Detection Dashboard</title>
+<style>
+  * {{ box-sizing: border-box; margin: 0; padding: 0; }}
+  body {{ font-family: -apple-system, "Segoe UI", system-ui, sans-serif;
+         font-size: 14px; line-height: 1.6; background: #f7f8fa; color: #1f2328; }}
+  .container {{ max-width: 960px; margin: 0 auto; padding: 24px 16px; }}
+  h1 {{ font-size: 22px; font-weight: 700; margin-bottom: 4px; }}
+  .subtitle {{ color: #57606a; font-size: 13px; margin-bottom: 24px; }}
+  .stats {{ display: flex; gap: 16px; margin-bottom: 24px; flex-wrap: wrap; }}
+  .stat-card {{ background: #fff; border: 1px solid #e5e7eb; border-radius: 6px;
+               padding: 12px 20px; flex: 1; min-width: 140px; }}
+  .stat-card .label {{ font-size: 11px; color: #57606a; text-transform: uppercase;
+                       letter-spacing: 0.05em; }}
+  .stat-card .value {{ font-size: 24px; font-weight: 700; }}
+  .stat-card.hold .value {{ color: #c0392b; }}
+  .stat-card.escalate .value {{ color: #e67e22; }}
+  .stat-card.log .value {{ color: #27ae60; }}
+  table {{ width: 100%; border-collapse: collapse; background: #fff;
+           border: 1px solid #e5e7eb; border-radius: 6px; overflow: hidden; }}
+  th {{ background: #f7f8fa; border-bottom: 2px solid #e5e7eb; padding: 8px 12px;
+       text-align: left; font-size: 12px; font-weight: 600; color: #57606a;
+       text-transform: uppercase; letter-spacing: 0.04em; }}
+  td {{ padding: 8px 12px; border-bottom: 1px solid #f0f0f0; vertical-align: middle; }}
+  tr:last-child td {{ border-bottom: none; }}
+  tr:hover td {{ background: #f7f8fa; }}
+  .footer {{ text-align: center; font-size: 12px; color: #57606a; margin-top: 32px;
+            padding-top: 16px; border-top: 1px solid #e5e7eb; }}
+  a {{ color: #3b82d4; text-decoration: none; }}
+  a:hover {{ text-decoration: underline; }}
+</style>
+</head>
+<body>
+<div class="container">
+  <h1>RingGuard — Fraud Detection Dashboard</h1>
+  <p class="subtitle">Phase 3 · {len(cases)} cases · LightGBM + LangGraph + Razorpay test-mode</p>
+
+  <div class="stats">
+    <div class="stat-card hold">
+      <div class="label">Auto Hold</div>
+      <div class="value">{auto_holds}</div>
+    </div>
+    <div class="stat-card escalate">
+      <div class="label">Escalate</div>
+      <div class="value">{escalates}</div>
+    </div>
+    <div class="stat-card log">
+      <div class="label">Log Only</div>
+      <div class="value">{log_only}</div>
+    </div>
+    <div class="stat-card">
+      <div class="label">Total Cases</div>
+      <div class="value">{len(cases)}</div>
+    </div>
+  </div>
+
+  <table>
+    <thead>
+      <tr>
+        <th>Case</th>
+        <th>Cluster</th>
+        <th style="text-align:center">Size</th>
+        <th style="text-align:right">Risk Score</th>
+        <th>Decision</th>
+        <th>Razorpay Order</th>
+        <th style="text-align:center">Verified</th>
+      </tr>
+    </thead>
+    <tbody>
+{rows}    </tbody>
+  </table>
+
+  <p style="margin-top:12px;font-size:12px;color:#57606a">
+    Click any row to view the full evidence bundle, LLM narrative, and audit chain.
+  </p>
+
+  <div class="footer">Made with IBM Bob &middot; RingGuard v3.0</div>
+</div>
+</body>
+</html>"""
+
+
+@app.get("/", response_class=HTMLResponse)
+def dashboard():
+    """Serve the HTML dashboard."""
+    cases = _load_cases()
+    html  = _build_dashboard(cases)
+    return HTMLResponse(content=html)
+
+
+@app.get("/cases/{case_id}/detail", response_class=HTMLResponse)
+def case_detail_html(case_id: str):
+    """Serve HTML detail page for a single case."""
+    cases = _load_cases()
+    case  = next((c for c in cases if c.get("case_id") == case_id), None)
+    if not case:
+        raise HTTPException(status_code=404, detail=f"Case {case_id} not found")
+
+    audit_records = [r for r in _load_audit() if r.get("case_id") == case_id]
+
+    decision = case.get("policy_decision", "")
+    badge    = _decision_badge(decision)
+    risk     = round(case.get("risk_probability", 0) * 100, 2)
+    narrative = case.get("llm_narrative") or "<em>No narrative (log_only case)</em>"
+    confidence = case.get("llm_confidence_statement") or "—"
+    members  = ", ".join(case.get("customer_ids", []))
+
+    action     = case.get("action_taken") or {}
+    order_id   = action.get("order_id") or "—"
+    verified   = "Yes" if action.get("notes_verified") else "—"
+    tool_used  = action.get("tool") or "—"
+    timestamp  = action.get("timestamp") or "—"
+
+    audit_rows = ""
+    for rec in audit_records:
+        ts = rec.get("timestamp", "")[:19].replace("T", " ")
+        audit_rows += (
+            f'<tr>'
+            f'<td>{ts}</td>'
+            f'<td>{rec.get("policy_decision","")}</td>'
+            f'<td style="font-size:11px">{rec.get("reason_if_not_auto_actioned") or "—"}</td>'
+            f'</tr>\n'
+        )
+
+    html = f"""<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="utf-8">
+<title>RingGuard — {case_id}</title>
+<style>
+  * {{ box-sizing: border-box; margin: 0; padding: 0; }}
+  body {{ font-family: -apple-system, "Segoe UI", system-ui, sans-serif;
+         font-size: 14px; line-height: 1.6; background: #f7f8fa; color: #1f2328; }}
+  .container {{ max-width: 760px; margin: 0 auto; padding: 24px 16px; }}
+  h1 {{ font-size: 20px; font-weight: 700; margin-bottom: 4px; }}
+  .back {{ font-size: 13px; color: #3b82d4; margin-bottom: 16px; display: block; }}
+  .card {{ background: #fff; border: 1px solid #e5e7eb; border-radius: 6px;
+           padding: 16px; margin-bottom: 16px; }}
+  .card h2 {{ font-size: 13px; font-weight: 600; color: #57606a; text-transform: uppercase;
+              letter-spacing: 0.05em; margin-bottom: 8px; }}
+  .row {{ display: flex; gap: 24px; margin-bottom: 6px; flex-wrap: wrap; }}
+  .kv {{ flex: 1; min-width: 200px; }}
+  .kv .k {{ font-size: 11px; color: #57606a; }}
+  .kv .v {{ font-weight: 500; }}
+  .narrative {{ background: #f7f8fa; border-left: 3px solid #3b82d4;
+               padding: 10px 14px; border-radius: 0 4px 4px 0; font-style: italic; }}
+  table {{ width: 100%; border-collapse: collapse; }}
+  th {{ text-align: left; font-size: 12px; color: #57606a; padding: 4px 8px;
+        border-bottom: 1px solid #e5e7eb; }}
+  td {{ padding: 6px 8px; border-bottom: 1px solid #f0f0f0; }}
+  .footer {{ text-align: center; font-size: 12px; color: #57606a; margin-top: 32px;
+            padding-top: 16px; border-top: 1px solid #e5e7eb; }}
+  span.badge {{ display: inline-block; }}
+</style>
+</head>
+<body>
+<div class="container">
+  <a class="back" href="/">&#8592; Back to all cases</a>
+  <h1>{case_id} — {case.get("cluster_id","")}</h1>
+
+  <div class="card">
+    <h2>Summary</h2>
+    <div class="row">
+      <div class="kv"><div class="k">Risk Score</div><div class="v">{risk}%</div></div>
+      <div class="kv"><div class="k">Decision</div><div class="v">{badge}</div></div>
+      <div class="kv"><div class="k">Cluster Size</div><div class="v">{len(case.get("customer_ids",[]))}</div></div>
+    </div>
+    <div class="row">
+      <div class="kv"><div class="k">Cluster Members</div><div class="v" style="font-size:12px">{members}</div></div>
+    </div>
+  </div>
+
+  <div class="card">
+    <h2>LLM Narrative (openai/gpt-oss-120b via Groq)</h2>
+    <div class="narrative">{narrative}</div>
+    <div style="margin-top:8px;font-size:12px;color:#57606a">
+      <strong>Confidence:</strong> {confidence}
+    </div>
+  </div>
+
+  <div class="card">
+    <h2>Razorpay Action (Test Mode)</h2>
+    <div class="row">
+      <div class="kv"><div class="k">Order ID</div><div class="v" style="font-size:12px">{order_id}</div></div>
+      <div class="kv"><div class="k">Notes Verified</div><div class="v">{verified}</div></div>
+    </div>
+    <div class="row">
+      <div class="kv"><div class="k">Tool Used</div><div class="v" style="font-size:11px">{tool_used}</div></div>
+      <div class="kv"><div class="k">Timestamp</div><div class="v" style="font-size:12px">{timestamp[:19] if len(timestamp)>=19 else timestamp}</div></div>
+    </div>
+  </div>
+
+  <div class="card">
+    <h2>Audit Chain</h2>
+    {f'<table><thead><tr><th>Timestamp</th><th>Decision</th><th>Reason</th></tr></thead><tbody>{audit_rows}</tbody></table>' if audit_rows else '<p style="color:#57606a;font-size:13px">No audit records found for this case.</p>'}
+  </div>
+
+  <div class="footer">Made with IBM Bob &middot; RingGuard v3.0</div>
+</div>
+</body>
+</html>"""
+    return HTMLResponse(content=html)

audit/.gitkeep

@@ -0,0 +1 @@
+

audit/audit_log.jsonl

@@ -0,0 +1,28 @@
+{"case_id": "CASE01", "timestamp": "2026-08-22T02:48:44.328671+00:00", "cluster_members": ["CUST0041", "CUST0042", "CUST0043", "CUST0044"], "evidence_bundle": {"cluster_id": "CAND01", "size": 4, "customer_ids": ["CUST0041", "CUST0042", "CUST0043", "CUST0044"], "formation_type": null}, "graph_score_inputs": {"size": 4, "strong_link_count": 6, "weak_link_count": 0, "strong_link_density": 1.0, "has_same_contact_edge": 0, "order_count": 12, "total_amount": 58186, "avg_order_amount": 4848.833333333333, "return_rate": 0.3333333333333333, "chargeback_rate": 0.16666666666666666, "shared_coupon_count": 0, "time_span_days": 118.78427083333334, "order_velocity": 0.10102347655808104, "formation_type_strong": 1}, "lightgbm_probability": 0.9958773824170732, "llm_narrative": "The four customers (CUST0041, CUST0042, CUST0043, CUST0044) form a strong‑anchored cluster with six strong links and no weak links, indicating tightly coordinated activity. Their combined purchase total of $58,186 is accompanied by a 33.3% return rate and a 16.7% chargeback rate, further supporting fraudulent behavior.", "llm_confidence_statement": "The model assigns a risk probability of 99.6%, indicating very high confidence in the escalation decision.", "policy_decision": "escalate", "action_taken": {"tool": "razorpay.Client.order.create + order.edit + order.fetch", "order_id": "order_TSeidafHmbAnDR", "razorpay_status": "created", "notes_verified": true, "timestamp": "2026-08-22T02:47:10.941027+00:00", "test_mode": true, "error": null, "skip_reason": null, "notes_written": {"case_id": "CASE01", "cluster_id": "CAND01", "ringguard_status": "escalate_flagged", "action_label": "FRAUD_RING_ESCALATE", "risk_score": "0.9959", "action_by": "RingGuard_v3", "action_timestamp": "2026-08-22T02:47:10.102605+00:00"}}, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.9959 >= 0.85 but total_amount=₹58186 > ₹25,000 auto_hold cap"}
+{"case_id": "CASE02", "timestamp": "2026-08-22T02:48:44.330726+00:00", "cluster_members": ["CUST0009", "CUST0010", "CUST0011", "CUST0012", "CUST0013"], "evidence_bundle": {"cluster_id": "CAND02", "size": 5, "customer_ids": ["CUST0009", "CUST0010", "CUST0011", "CUST0012", "CUST0013"], "formation_type": null}, "graph_score_inputs": {"size": 5, "strong_link_count": 10, "weak_link_count": 0, "strong_link_density": 1.0, "has_same_contact_edge": 0, "order_count": 17, "total_amount": 59830, "avg_order_amount": 3519.4117647058824, "return_rate": 0.47058823529411764, "chargeback_rate": 0.17647058823529413, "shared_coupon_count": 0, "time_span_days": 153.46033564814815, "order_velocity": 0.11077781061926893, "formation_type_strong": 1}, "lightgbm_probability": 0.9968053865328341, "llm_narrative": "The five customers (CUST0009‑CUST0013) form a strong‑anchored cluster with ten strong links and no weak links. The group shows a return rate of 47% and a chargeback rate of 18%, with total transaction volume of $59,830. The risk model assigns a probability of 99.7%, resulting in an escalation decision.", "llm_confidence_statement": "We are highly confident in this assessment given the very high risk probability and supporting metrics.", "policy_decision": "escalate", "action_taken": {"tool": "razorpay.Client.order.create + order.edit + order.fetch", "order_id": "order_TSeieevTUEtDNP", "razorpay_status": "created", "notes_verified": true, "timestamp": "2026-08-22T02:47:11.992374+00:00", "test_mode": true, "error": null, "skip_reason": null, "notes_written": {"case_id": "CASE02", "cluster_id": "CAND02", "ringguard_status": "escalate_flagged", "action_label": "FRAUD_RING_ESCALATE", "risk_score": "0.9968", "action_by": "RingGuard_v3", "action_timestamp": "2026-08-22T02:47:11.088678+00:00"}}, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.9968 >= 0.85 but total_amount=₹59830 > ₹25,000 auto_hold cap"}
+{"case_id": "CASE03", "timestamp": "2026-08-22T02:48:44.331690+00:00", "cluster_members": ["CUST0001", "CUST0002", "CUST0003"], "evidence_bundle": {"cluster_id": "CAND03", "size": 3, "customer_ids": ["CUST0001", "CUST0002", "CUST0003"], "formation_type": null}, "graph_score_inputs": {"size": 3, "strong_link_count": 3, "weak_link_count": 0, "strong_link_density": 1.0, "has_same_contact_edge": 0, "order_count": 8, "total_amount": 32897, "avg_order_amount": 4112.125, "return_rate": 0.25, "chargeback_rate": 0.125, "shared_coupon_count": 1, "time_span_days": 127.71112268518519, "order_velocity": 0.06264137243331916, "formation_type_strong": 1}, "lightgbm_probability": 0.9950028638127228, "llm_narrative": "The three customers (CUST0001, CUST0002, CUST0003) are connected by three strong links, forming a strong‑anchored cluster. This cluster accounts for a total transaction amount of 32,897, includes one shared coupon, and exhibits a 25% return rate and a 12.5% chargeback rate, resulting in a risk probability of 99.5%.", "llm_confidence_statement": "The model’s confidence in this risk assessment is extremely high.", "policy_decision": "escalate", "action_taken": {"tool": "razorpay.Client.order.create + order.edit + order.fetch", "order_id": "order_TSeifof2nHqiyN", "razorpay_status": "created", "notes_verified": true, "timestamp": "2026-08-22T02:47:13.232792+00:00", "test_mode": true, "error": null, "skip_reason": null, "notes_written": {"case_id": "CASE03", "cluster_id": "CAND03", "ringguard_status": "escalate_flagged", "action_label": "FRAUD_RING_ESCALATE", "risk_score": "0.995", "action_by": "RingGuard_v3", "action_timestamp": "2026-08-22T02:47:12.186378+00:00"}}, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.9950 >= 0.85 but total_amount=₹32897 > ₹25,000 auto_hold cap"}
+{"case_id": "CASE04", "timestamp": "2026-08-22T02:48:44.333027+00:00", "cluster_members": ["CUST0014", "CUST0015", "CUST0016", "CUST0017", "CUST0018"], "evidence_bundle": {"cluster_id": "CAND04", "size": 5, "customer_ids": ["CUST0014", "CUST0015", "CUST0016", "CUST0017", "CUST0018"], "formation_type": null}, "graph_score_inputs": {"size": 5, "strong_link_count": 10, "weak_link_count": 0, "strong_link_density": 1.0, "has_same_contact_edge": 0, "order_count": 16, "total_amount": 62087, "avg_order_amount": 3880.4375, "return_rate": 0.5, "chargeback_rate": 0.125, "shared_coupon_count": 1, "time_span_days": 158.5827314814815, "order_velocity": 0.10089370923635782, "formation_type_strong": 1}, "lightgbm_probability": 0.9955003588432927, "llm_narrative": "The five customers (CUST0014‑CUST0018) form a strong‑anchored cluster with ten strong links and no weak links, collectively spending $62,087. The group exhibits a 50% return rate, a 12.5% chargeback rate, and shares a single coupon, indicating coordinated risky behavior.", "llm_confidence_statement": "Given the 0.9955 risk probability, we are highly confident that this cluster represents a high‑risk fraud pattern.", "policy_decision": "escalate", "action_taken": {"tool": "razorpay.Client.order.create + order.edit + order.fetch", "order_id": "order_TSeihAqIgYfEsF", "razorpay_status": "created", "notes_verified": true, "timestamp": "2026-08-22T02:47:14.280655+00:00", "test_mode": true, "error": null, "skip_reason": null, "notes_written": {"case_id": "CASE04", "cluster_id": "CAND04", "ringguard_status": "escalate_flagged", "action_label": "FRAUD_RING_ESCALATE", "risk_score": "0.9955", "action_by": "RingGuard_v3", "action_timestamp": "2026-08-22T02:47:13.364816+00:00"}}, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.9955 >= 0.85 but total_amount=₹62087 > ₹25,000 auto_hold cap"}
+{"case_id": "CASE05", "timestamp": "2026-08-22T02:48:44.333027+00:00", "cluster_members": ["CUST0033", "CUST0034", "CUST0035"], "evidence_bundle": {"cluster_id": "CAND05", "size": 3, "customer_ids": ["CUST0033", "CUST0034", "CUST0035"], "formation_type": null}, "graph_score_inputs": {"size": 3, "strong_link_count": 3, "weak_link_count": 0, "strong_link_density": 1.0, "has_same_contact_edge": 0, "order_count": 11, "total_amount": 50134, "avg_order_amount": 4557.636363636364, "return_rate": 0.18181818181818182, "chargeback_rate": 0.09090909090909091, "shared_coupon_count": 1, "time_span_days": 170.66708333333332, "order_velocity": 0.06445296764412196, "formation_type_strong": 1}, "lightgbm_probability": 0.9969893801631614, "llm_narrative": "The three customers (CUST0033, CUST0034, CUST0035) are connected by three strong links, forming a strong‑anchored cluster with a shared coupon and a combined transaction volume of $50,134. The cluster exhibits a return rate of 18.2% and a chargeback rate of 9.1%, and its calculated risk probability is 99.7%.", "llm_confidence_statement": "We are highly confident in the fraud assessment given the strong linkage, high risk probability, and notable financial metrics.", "policy_decision": "escalate", "action_taken": {"tool": "razorpay.Client.order.create + order.edit + order.fetch", "order_id": "order_TSeiiK16ZPVegj", "razorpay_status": "created", "notes_verified": true, "timestamp": "2026-08-22T02:47:15.299833+00:00", "test_mode": true, "error": null, "skip_reason": null, "notes_written": {"case_id": "CASE05", "cluster_id": "CAND05", "ringguard_status": "escalate_flagged", "action_label": "FRAUD_RING_ESCALATE", "risk_score": "0.997", "action_by": "RingGuard_v3", "action_timestamp": "2026-08-22T02:47:14.523378+00:00"}}, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.9970 >= 0.85 but total_amount=₹50134 > ₹25,000 auto_hold cap"}
+{"case_id": "CASE06", "timestamp": "2026-08-22T02:48:44.333027+00:00", "cluster_members": ["CUST0036", "CUST0037", "CUST0038", "CUST0039", "CUST0040"], "evidence_bundle": {"cluster_id": "CAND06", "size": 5, "customer_ids": ["CUST0036", "CUST0037", "CUST0038", "CUST0039", "CUST0040"], "formation_type": null}, "graph_score_inputs": {"size": 5, "strong_link_count": 10, "weak_link_count": 0, "strong_link_density": 1.0, "has_same_contact_edge": 0, "order_count": 15, "total_amount": 62270, "avg_order_amount": 4151.333333333333, "return_rate": 0.26666666666666666, "chargeback_rate": 0.2, "shared_coupon_count": 0, "time_span_days": 145.8427199074074, "order_velocity": 0.10285052287507526, "formation_type_strong": 1}, "lightgbm_probability": 0.9969377389203431, "llm_narrative": "The five customers (CUST0036‑CUST0040) form a strong‑anchored cluster with ten strong links and no weak links, collectively transacting $62,270. Their return rate of 26.7% and chargeback rate of 20% are elevated, and the absence of shared coupons suggests the links are not promotional. These characteristics indicate coordinated fraudulent behavior.", "llm_confidence_statement": "We are highly confident in this assessment, reflected by a risk probability of 0.997.", "policy_decision": "escalate", "action_taken": {"tool": "razorpay.Client.order.create + order.edit + order.fetch", "order_id": "order_TSeijQj96lsR88", "razorpay_status": "created", "notes_verified": true, "timestamp": "2026-08-22T02:47:16.209387+00:00", "test_mode": true, "error": null, "skip_reason": null, "notes_written": {"case_id": "CASE06", "cluster_id": "CAND06", "ringguard_status": "escalate_flagged", "action_label": "FRAUD_RING_ESCALATE", "risk_score": "0.9969", "action_by": "RingGuard_v3", "action_timestamp": "2026-08-22T02:47:15.427579+00:00"}}, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.9969 >= 0.85 but total_amount=₹62270 > ₹25,000 auto_hold cap"}
+{"case_id": "CASE07", "timestamp": "2026-08-22T02:48:44.333027+00:00", "cluster_members": ["CUST0019", "CUST0020", "CUST0021", "CUST0022", "CUST0023"], "evidence_bundle": {"cluster_id": "CAND07", "size": 5, "customer_ids": ["CUST0019", "CUST0020", "CUST0021", "CUST0022", "CUST0023"], "formation_type": null}, "graph_score_inputs": {"size": 5, "strong_link_count": 10, "weak_link_count": 0, "strong_link_density": 1.0, "has_same_contact_edge": 0, "order_count": 18, "total_amount": 70841, "avg_order_amount": 3935.6111111111113, "return_rate": 0.3333333333333333, "chargeback_rate": 0.05555555555555555, "shared_coupon_count": 0, "time_span_days": 144.96715277777778, "order_velocity": 0.12416605869049836, "formation_type_strong": 1}, "lightgbm_probability": 0.9940850384979173, "llm_narrative": "A cluster of five customers (CUST0019‑CUST0023) is connected by ten strong links with no weak links, forming a strong‑anchored network. The group has generated a total transaction amount of 70,841, a return rate of 33.33% and a chargeback rate of 5.56%, and the model assigns a risk probability of 99.41%.", "llm_confidence_statement": "The assessment is made with very high confidence.", "policy_decision": "escalate", "action_taken": {"tool": "razorpay.Client.order.create + order.edit + order.fetch", "order_id": "order_TSeikSedcIcNGX", "razorpay_status": "created", "notes_verified": true, "timestamp": "2026-08-22T02:47:17.279431+00:00", "test_mode": true, "error": null, "skip_reason": null, "notes_written": {"case_id": "CASE07", "cluster_id": "CAND07", "ringguard_status": "escalate_flagged", "action_label": "FRAUD_RING_ESCALATE", "risk_score": "0.9941", "action_by": "RingGuard_v3", "action_timestamp": "2026-08-22T02:47:16.396113+00:00"}}, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.9941 >= 0.85 but total_amount=₹70841 > ₹25,000 auto_hold cap"}
+{"case_id": "CASE08", "timestamp": "2026-08-22T02:48:44.338268+00:00", "cluster_members": ["CUST0501", "CUST0502", "CUST0503"], "evidence_bundle": {"cluster_id": "CAND08", "size": 3, "customer_ids": ["CUST0501", "CUST0502", "CUST0503"], "formation_type": null}, "graph_score_inputs": {"size": 3, "strong_link_count": 3, "weak_link_count": 0, "strong_link_density": 1.0, "has_same_contact_edge": 0, "order_count": 6, "total_amount": 19800, "avg_order_amount": 3300.0, "return_rate": 0.16666666666666666, "chargeback_rate": 0.16666666666666666, "shared_coupon_count": 1, "time_span_days": 3.0625, "order_velocity": 1.9591836734693877, "formation_type_strong": 1}, "lightgbm_probability": 0.9867466290228035, "llm_narrative": "The three customers (CUST0501, CUST0502, CUST0503) are connected by three strong links, forming a strong‑anchored cluster. They share one coupon, together generated $19,800 in transactions, and exhibit a 16.7% return and chargeback rate each.", "llm_confidence_statement": "The model assigns a risk probability of 0.99, indicating very high confidence in the high‑risk classification.", "policy_decision": "auto_hold", "action_taken": {"tool": "razorpay.Client.order.create + order.edit + order.fetch", "order_id": "order_TSeilZzpguMme6", "razorpay_status": "created", "notes_verified": true, "timestamp": "2026-08-22T02:47:18.321891+00:00", "test_mode": true, "error": null, "skip_reason": null, "notes_written": {"case_id": "CASE08", "cluster_id": "CAND08", "ringguard_status": "auto_hold", "action_label": "FRAUD_RING_AUTO_HOLD", "risk_score": "0.9867", "action_by": "RingGuard_v3", "action_timestamp": "2026-08-22T02:47:17.462450+00:00"}}, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": null}
+{"case_id": "CASE09", "timestamp": "2026-08-22T02:48:44.339504+00:00", "cluster_members": ["CUST0024", "CUST0025", "CUST0026", "CUST0027", "CUST0028", "CUST0029"], "evidence_bundle": {"cluster_id": "CAND09", "size": 6, "customer_ids": ["CUST0024", "CUST0025", "CUST0026", "CUST0027", "CUST0028", "CUST0029"], "formation_type": null}, "graph_score_inputs": {"size": 6, "strong_link_count": 15, "weak_link_count": 0, "strong_link_density": 1.0, "has_same_contact_edge": 0, "order_count": 20, "total_amount": 81019, "avg_order_amount": 4050.95, "return_rate": 0.55, "chargeback_rate": 0.05, "shared_coupon_count": 0, "time_span_days": 174.1008564814815, "order_velocity": 0.11487594262425316, "formation_type_strong": 1}, "lightgbm_probability": 0.9913203009684713, "llm_narrative": "The six customers (CUST0024‑CUST0029) form a strong_anchored cluster with 15 strong links and no weak links, indicating tightly coordinated activity. Their combined purchase total of 81,019 and a return rate of 55% further highlight abnormal behavior, while a chargeback rate of 5% adds to the risk profile.", "llm_confidence_statement": "The evidence yields a very high confidence in the elevated risk assessment.", "policy_decision": "escalate", "action_taken": {"tool": "razorpay.Client.order.create + order.edit + order.fetch", "order_id": "order_TSeimiDR56y6I3", "razorpay_status": "created", "notes_verified": true, "timestamp": "2026-08-22T02:47:19.433491+00:00", "test_mode": true, "error": null, "skip_reason": null, "notes_written": {"case_id": "CASE09", "cluster_id": "CAND09", "ringguard_status": "escalate_flagged", "action_label": "FRAUD_RING_ESCALATE", "risk_score": "0.9913", "action_by": "RingGuard_v3", "action_timestamp": "2026-08-22T02:47:18.435548+00:00"}}, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.9913 >= 0.85 but total_amount=₹81019 > ₹25,000 auto_hold cap"}
+{"case_id": "CASE10", "timestamp": "2026-08-22T02:48:44.339504+00:00", "cluster_members": ["CUST0004", "CUST0005", "CUST0006", "CUST0007", "CUST0008"], "evidence_bundle": {"cluster_id": "CAND10", "size": 5, "customer_ids": ["CUST0004", "CUST0005", "CUST0006", "CUST0007", "CUST0008"], "formation_type": null}, "graph_score_inputs": {"size": 5, "strong_link_count": 7, "weak_link_count": 0, "strong_link_density": 0.7, "has_same_contact_edge": 1, "order_count": 16, "total_amount": 63837, "avg_order_amount": 3989.8125, "return_rate": 0.375, "chargeback_rate": 0.0625, "shared_coupon_count": 0, "time_span_days": 173.6175462962963, "order_velocity": 0.09215658406261741, "formation_type_strong": 1}, "lightgbm_probability": 0.9959231873282086, "llm_narrative": "The five customers (CUST0004‑CUST0008) form a strong‑anchored cluster with seven strong links and no weak links, collectively transacting $63,837. Their return rate of 37.5% and chargeback rate of 6.25% are elevated, supporting the high risk probability of 99.6%.", "llm_confidence_statement": "The evidence yields a very high confidence in the elevated fraud risk assessment.", "policy_decision": "escalate", "action_taken": {"tool": "razorpay.Client.order.create + order.edit + order.fetch", "order_id": "order_TSeinwRO4xKvs9", "razorpay_status": "created", "notes_verified": true, "timestamp": "2026-08-22T02:47:20.455513+00:00", "test_mode": true, "error": null, "skip_reason": null, "notes_written": {"case_id": "CASE10", "cluster_id": "CAND10", "ringguard_status": "escalate_flagged", "action_label": "FRAUD_RING_ESCALATE", "risk_score": "0.9959", "action_by": "RingGuard_v3", "action_timestamp": "2026-08-22T02:47:19.567020+00:00"}}, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.9959 >= 0.85 but total_amount=₹63837 > ₹25,000 auto_hold cap"}
+{"case_id": "CASE11", "timestamp": "2026-08-22T02:48:44.339504+00:00", "cluster_members": ["CUST0030", "CUST0031", "CUST0032"], "evidence_bundle": {"cluster_id": "CAND11", "size": 3, "customer_ids": ["CUST0030", "CUST0031", "CUST0032"], "formation_type": null}, "graph_score_inputs": {"size": 3, "strong_link_count": 2, "weak_link_count": 0, "strong_link_density": 0.6666666666666666, "has_same_contact_edge": 1, "order_count": 7, "total_amount": 26235, "avg_order_amount": 3747.8571428571427, "return_rate": 0.2857142857142857, "chargeback_rate": 0.2857142857142857, "shared_coupon_count": 0, "time_span_days": 99.31296296296296, "order_velocity": 0.07048425292285891, "formation_type_strong": 1}, "lightgbm_probability": 0.9927212653417073, "llm_narrative": "The three customers (CUST0030, CUST0031, CUST0032) form a strong‑anchored cluster with two strong links and no weak links, together generating $26,235 in transactions. Their return and chargeback rates are each 28.6%, and the model assigns a risk probability of 99.3%.", "llm_confidence_statement": "Confidence in this assessment is high given the strong links, elevated return/chargeback rates, and near‑certain risk probability.", "policy_decision": "escalate", "action_taken": {"tool": "razorpay.Client.order.create + order.edit + order.fetch", "order_id": "order_TSeip2lX554G6F", "razorpay_status": "created", "notes_verified": true, "timestamp": "2026-08-22T02:47:21.336242+00:00", "test_mode": true, "error": null, "skip_reason": null, "notes_written": {"case_id": "CASE11", "cluster_id": "CAND11", "ringguard_status": "escalate_flagged", "action_label": "FRAUD_RING_ESCALATE", "risk_score": "0.9927", "action_by": "RingGuard_v3", "action_timestamp": "2026-08-22T02:47:20.576284+00:00"}}, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.9927 >= 0.85 but total_amount=₹26235 > ₹25,000 auto_hold cap"}
+{"case_id": "CASE12", "timestamp": "2026-08-22T02:48:44.341012+00:00", "cluster_members": ["CUST0001", "CUST0019", "CUST0040", "CUST0126", "CUST0138", "CUST0168", "CUST0211", "CUST0219", "CUST0258", "CUST0281", "CUST0292", "CUST0322", "CUST0325", "CUST0371", "CUST0373", "CUST0400", "CUST0419", "CUST0465"], "evidence_bundle": {"cluster_id": "CAND12", "size": 18, "customer_ids": ["CUST0001", "CUST0019", "CUST0040", "CUST0126", "CUST0138", "CUST0168", "CUST0211", "CUST0219", "CUST0258", "CUST0281", "CUST0292", "CUST0322", "CUST0325", "CUST0371", "CUST0373", "CUST0400", "CUST0419", "CUST0465"], "formation_type": null}, "graph_score_inputs": {"size": 18, "strong_link_count": 0, "weak_link_count": 41, "strong_link_density": 0.0, "has_same_contact_edge": 0, "order_count": 47, "total_amount": 141392, "avg_order_amount": 3008.340425531915, "return_rate": 0.23404255319148937, "chargeback_rate": 0.1276595744680851, "shared_coupon_count": 0, "time_span_days": 166.15458333333333, "order_velocity": 0.28286911535693415, "formation_type_strong": 0}, "lightgbm_probability": 0.00720344453548541, "llm_narrative": null, "llm_confidence_statement": null, "policy_decision": "log_only", "action_taken": null, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.0072 < 0.50 threshold — insufficient signal for action"}
+{"case_id": "CASE13", "timestamp": "2026-08-22T02:48:44.343060+00:00", "cluster_members": ["CUST0002", "CUST0009", "CUST0020", "CUST0041", "CUST0064", "CUST0073", "CUST0074", "CUST0080", "CUST0081", "CUST0086", "CUST0089", "CUST0103", "CUST0162", "CUST0167", "CUST0173", "CUST0181", "CUST0226", "CUST0240", "CUST0253", "CUST0262", "CUST0272", "CUST0286", "CUST0287", "CUST0324", "CUST0331", "CUST0333", "CUST0369", "CUST0392", "CUST0393", "CUST0398", "CUST0406", "CUST0410", "CUST0411", "CUST0413", "CUST0415", "CUST0423", "CUST0446", "CUST0468", "CUST0476", "CUST0478", "CUST0485"], "evidence_bundle": {"cluster_id": "CAND13", "size": 41, "customer_ids": ["CUST0002", "CUST0009", "CUST0020", "CUST0041", "CUST0064", "CUST0073", "CUST0074", "CUST0080", "CUST0081", "CUST0086", "CUST0089", "CUST0103", "CUST0162", "CUST0167", "CUST0173", "CUST0181", "CUST0226", "CUST0240", "CUST0253", "CUST0262", "CUST0272", "CUST0286", "CUST0287", "CUST0324", "CUST0331", "CUST0333", "CUST0369", "CUST0392", "CUST0393", "CUST0398", "CUST0406", "CUST0410", "CUST0411", "CUST0413", "CUST0415", "CUST0423", "CUST0446", "CUST0468", "CUST0476", "CUST0478", "CUST0485"], "formation_type": null}, "graph_score_inputs": {"size": 41, "strong_link_count": 0, "weak_link_count": 146, "strong_link_density": 0.0, "has_same_contact_edge": 0, "order_count": 127, "total_amount": 369082, "avg_order_amount": 2906.1574803149606, "return_rate": 0.15748031496062992, "chargeback_rate": 0.08661417322834646, "shared_coupon_count": 0, "time_span_days": 177.8061226851852, "order_velocity": 0.7142611181329227, "formation_type_strong": 0}, "lightgbm_probability": 0.007559966037372132, "llm_narrative": null, "llm_confidence_statement": null, "policy_decision": "log_only", "action_taken": null, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.0076 < 0.50 threshold — insufficient signal for action"}
+{"case_id": "CASE14", "timestamp": "2026-08-22T02:48:44.344073+00:00", "cluster_members": ["CUST0003", "CUST0012", "CUST0043", "CUST0077", "CUST0088", "CUST0098", "CUST0109", "CUST0125", "CUST0139", "CUST0153", "CUST0159", "CUST0166", "CUST0180", "CUST0201", "CUST0209", "CUST0245", "CUST0261", "CUST0265", "CUST0273", "CUST0277", "CUST0297", "CUST0315", "CUST0334", "CUST0347", "CUST0381", "CUST0405", "CUST0412", "CUST0428", "CUST0429", "CUST0462", "CUST0471"], "evidence_bundle": {"cluster_id": "CAND14", "size": 31, "customer_ids": ["CUST0003", "CUST0012", "CUST0043", "CUST0077", "CUST0088", "CUST0098", "CUST0109", "CUST0125", "CUST0139", "CUST0153", "CUST0159", "CUST0166", "CUST0180", "CUST0201", "CUST0209", "CUST0245", "CUST0261", "CUST0265", "CUST0273", "CUST0277", "CUST0297", "CUST0315", "CUST0334", "CUST0347", "CUST0381", "CUST0405", "CUST0412", "CUST0428", "CUST0429", "CUST0462", "CUST0471"], "formation_type": null}, "graph_score_inputs": {"size": 31, "strong_link_count": 0, "weak_link_count": 84, "strong_link_density": 0.0, "has_same_contact_edge": 0, "order_count": 90, "total_amount": 268965, "avg_order_amount": 2988.5, "return_rate": 0.08888888888888889, "chargeback_rate": 0.022222222222222223, "shared_coupon_count": 0, "time_span_days": 178.87412037037038, "order_velocity": 0.5031471283472937, "formation_type_strong": 0}, "lightgbm_probability": 0.0033586500336385537, "llm_narrative": null, "llm_confidence_statement": null, "policy_decision": "log_only", "action_taken": null, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.0034 < 0.50 threshold — insufficient signal for action"}
+{"case_id": "CASE15", "timestamp": "2026-08-22T02:48:44.344073+00:00", "cluster_members": ["CUST0004", "CUST0008", "CUST0027", "CUST0030", "CUST0042", "CUST0063", "CUST0075", "CUST0087", "CUST0110", "CUST0117", "CUST0120", "CUST0133", "CUST0135", "CUST0137", "CUST0144", "CUST0152", "CUST0154", "CUST0163", "CUST0170", "CUST0178", "CUST0179", "CUST0195", "CUST0198", "CUST0208", "CUST0210", "CUST0221", "CUST0228", "CUST0231", "CUST0237", "CUST0246", "CUST0251", "CUST0254", "CUST0266", "CUST0282", "CUST0295", "CUST0300", "CUST0303", "CUST0304", "CUST0335", "CUST0355", "CUST0365", "CUST0399", "CUST0420", "CUST0425", "CUST0470", "CUST0483", "CUST0486", "CUST0487", "CUST0491", "CUST0492", "CUST0498"], "evidence_bundle": {"cluster_id": "CAND15", "size": 51, "customer_ids": ["CUST0004", "CUST0008", "CUST0027", "CUST0030", "CUST0042", "CUST0063", "CUST0075", "CUST0087", "CUST0110", "CUST0117", "CUST0120", "CUST0133", "CUST0135", "CUST0137", "CUST0144", "CUST0152", "CUST0154", "CUST0163", "CUST0170", "CUST0178", "CUST0179", "CUST0195", "CUST0198", "CUST0208", "CUST0210", "CUST0221", "CUST0228", "CUST0231", "CUST0237", "CUST0246", "CUST0251", "CUST0254", "CUST0266", "CUST0282", "CUST0295", "CUST0300", "CUST0303", "CUST0304", "CUST0335", "CUST0355", "CUST0365", "CUST0399", "CUST0420", "CUST0425", "CUST0470", "CUST0483", "CUST0486", "CUST0487", "CUST0491", "CUST0492", "CUST0498"], "formation_type": null}, "graph_score_inputs": {"size": 51, "strong_link_count": 1, "weak_link_count": 167, "strong_link_density": 0.000784313725490196, "has_same_contact_edge": 1, "order_count": 145, "total_amount": 381606, "avg_order_amount": 2631.7655172413793, "return_rate": 0.2, "chargeback_rate": 0.0896551724137931, "shared_coupon_count": 0, "time_span_days": 180.07700231481482, "order_velocity": 0.8052110937881319, "formation_type_strong": 0}, "lightgbm_probability": 0.007646314104742993, "llm_narrative": null, "llm_confidence_statement": null, "policy_decision": "log_only", "action_taken": null, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.0076 < 0.50 threshold — insufficient signal for action"}
+{"case_id": "CASE16", "timestamp": "2026-08-22T02:48:44.345066+00:00", "cluster_members": ["CUST0005", "CUST0022", "CUST0028", "CUST0038", "CUST0102", "CUST0111", "CUST0113", "CUST0140", "CUST0161", "CUST0171", "CUST0177", "CUST0202", "CUST0222", "CUST0235", "CUST0252", "CUST0257", "CUST0263", "CUST0288", "CUST0293", "CUST0299", "CUST0308", "CUST0311", "CUST0353", "CUST0357", "CUST0364", "CUST0370", "CUST0383", "CUST0386", "CUST0387", "CUST0391", "CUST0396", "CUST0438", "CUST0442", "CUST0456", "CUST0460", "CUST0488", "CUST0490", "CUST0495", "CUST0496", "CUST0500"], "evidence_bundle": {"cluster_id": "CAND16", "size": 40, "customer_ids": ["CUST0005", "CUST0022", "CUST0028", "CUST0038", "CUST0102", "CUST0111", "CUST0113", "CUST0140", "CUST0161", "CUST0171", "CUST0177", "CUST0202", "CUST0222", "CUST0235", "CUST0252", "CUST0257", "CUST0263", "CUST0288", "CUST0293", "CUST0299", "CUST0308", "CUST0311", "CUST0353", "CUST0357", "CUST0364", "CUST0370", "CUST0383", "CUST0386", "CUST0387", "CUST0391", "CUST0396", "CUST0438", "CUST0442", "CUST0456", "CUST0460", "CUST0488", "CUST0490", "CUST0495", "CUST0496", "CUST0500"], "formation_type": null}, "graph_score_inputs": {"size": 40, "strong_link_count": 0, "weak_link_count": 112, "strong_link_density": 0.0, "has_same_contact_edge": 0, "order_count": 99, "total_amount": 276653, "avg_order_amount": 2794.4747474747473, "return_rate": 0.15151515151515152, "chargeback_rate": 0.030303030303030304, "shared_coupon_count": 0, "time_span_days": 172.41998842592594, "order_velocity": 0.5741793680871972, "formation_type_strong": 0}, "lightgbm_probability": 0.0035456513474297945, "llm_narrative": null, "llm_confidence_statement": null, "policy_decision": "log_only", "action_taken": null, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.0035 < 0.50 threshold — insufficient signal for action"}
+{"case_id": "CASE17", "timestamp": "2026-08-22T02:48:44.347067+00:00", "cluster_members": ["CUST0006", "CUST0011", "CUST0024", "CUST0025", "CUST0033", "CUST0044", "CUST0079", "CUST0084", "CUST0101", "CUST0107", "CUST0108", "CUST0114", "CUST0132", "CUST0145", "CUST0157", "CUST0172", "CUST0227", "CUST0244", "CUST0255", "CUST0256", "CUST0267", "CUST0271", "CUST0284", "CUST0291", "CUST0296", "CUST0316", "CUST0319", "CUST0329", "CUST0336", "CUST0342", "CUST0348", "CUST0349", "CUST0367", "CUST0375", "CUST0389", "CUST0407", "CUST0455", "CUST0489", "CUST0499"], "evidence_bundle": {"cluster_id": "CAND17", "size": 39, "customer_ids": ["CUST0006", "CUST0011", "CUST0024", "CUST0025", "CUST0033", "CUST0044", "CUST0079", "CUST0084", "CUST0101", "CUST0107", "CUST0108", "CUST0114", "CUST0132", "CUST0145", "CUST0157", "CUST0172", "CUST0227", "CUST0244", "CUST0255", "CUST0256", "CUST0267", "CUST0271", "CUST0284", "CUST0291", "CUST0296", "CUST0316", "CUST0319", "CUST0329", "CUST0336", "CUST0342", "CUST0348", "CUST0349", "CUST0367", "CUST0375", "CUST0389", "CUST0407", "CUST0455", "CUST0489", "CUST0499"], "formation_type": null}, "graph_score_inputs": {"size": 39, "strong_link_count": 1, "weak_link_count": 95, "strong_link_density": 0.001349527665317139, "has_same_contact_edge": 0, "order_count": 108, "total_amount": 321113, "avg_order_amount": 2973.2685185185187, "return_rate": 0.18518518518518517, "chargeback_rate": 0.10185185185185185, "shared_coupon_count": 0, "time_span_days": 176.43193287037036, "order_velocity": 0.6121340861767395, "formation_type_strong": 0}, "lightgbm_probability": 0.006896318028648148, "llm_narrative": null, "llm_confidence_statement": null, "policy_decision": "log_only", "action_taken": null, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.0069 < 0.50 threshold — insufficient signal for action"}
+{"case_id": "CASE18", "timestamp": "2026-08-22T02:48:44.348067+00:00", "cluster_members": ["CUST0007", "CUST0014", "CUST0015", "CUST0016", "CUST0032", "CUST0037", "CUST0065", "CUST0069", "CUST0070", "CUST0095", "CUST0106", "CUST0115", "CUST0124", "CUST0131", "CUST0149", "CUST0191", "CUST0200", "CUST0204", "CUST0215", "CUST0218", "CUST0223", "CUST0238", "CUST0239", "CUST0241", "CUST0242", "CUST0264", "CUST0276", "CUST0280", "CUST0305", "CUST0314", "CUST0323", "CUST0327", "CUST0339", "CUST0350", "CUST0352", "CUST0356", "CUST0358", "CUST0362", "CUST0379", "CUST0401", "CUST0418", "CUST0440", "CUST0443", "CUST0449", "CUST0479"], "evidence_bundle": {"cluster_id": "CAND18", "size": 45, "customer_ids": ["CUST0007", "CUST0014", "CUST0015", "CUST0016", "CUST0032", "CUST0037", "CUST0065", "CUST0069", "CUST0070", "CUST0095", "CUST0106", "CUST0115", "CUST0124", "CUST0131", "CUST0149", "CUST0191", "CUST0200", "CUST0204", "CUST0215", "CUST0218", "CUST0223", "CUST0238", "CUST0239", "CUST0241", "CUST0242", "CUST0264", "CUST0276", "CUST0280", "CUST0305", "CUST0314", "CUST0323", "CUST0327", "CUST0339", "CUST0350", "CUST0352", "CUST0356", "CUST0358", "CUST0362", "CUST0379", "CUST0401", "CUST0418", "CUST0440", "CUST0443", "CUST0449", "CUST0479"], "formation_type": null}, "graph_score_inputs": {"size": 45, "strong_link_count": 3, "weak_link_count": 106, "strong_link_density": 0.0030303030303030303, "has_same_contact_edge": 0, "order_count": 110, "total_amount": 308935, "avg_order_amount": 2808.5, "return_rate": 0.17272727272727273, "chargeback_rate": 0.045454545454545456, "shared_coupon_count": 1, "time_span_days": 180.25027777777777, "order_velocity": 0.610262582427828, "formation_type_strong": 0}, "lightgbm_probability": 0.0034784417949967547, "llm_narrative": null, "llm_confidence_statement": null, "policy_decision": "log_only", "action_taken": null, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.0035 < 0.50 threshold — insufficient signal for action"}
+{"case_id": "CASE19", "timestamp": "2026-08-22T02:48:44.349068+00:00", "cluster_members": ["CUST0010", "CUST0023", "CUST0029", "CUST0090", "CUST0119", "CUST0121", "CUST0129", "CUST0141", "CUST0143", "CUST0155", "CUST0196", "CUST0207", "CUST0230", "CUST0233", "CUST0247", "CUST0330", "CUST0338", "CUST0372", "CUST0374", "CUST0380", "CUST0461", "CUST0467", "CUST0480", "CUST0493"], "evidence_bundle": {"cluster_id": "CAND19", "size": 24, "customer_ids": ["CUST0010", "CUST0023", "CUST0029", "CUST0090", "CUST0119", "CUST0121", "CUST0129", "CUST0141", "CUST0143", "CUST0155", "CUST0196", "CUST0207", "CUST0230", "CUST0233", "CUST0247", "CUST0330", "CUST0338", "CUST0372", "CUST0374", "CUST0380", "CUST0461", "CUST0467", "CUST0480", "CUST0493"], "formation_type": null}, "graph_score_inputs": {"size": 24, "strong_link_count": 0, "weak_link_count": 47, "strong_link_density": 0.0, "has_same_contact_edge": 0, "order_count": 61, "total_amount": 165404, "avg_order_amount": 2711.5409836065573, "return_rate": 0.18032786885245902, "chargeback_rate": 0.08196721311475409, "shared_coupon_count": 0, "time_span_days": 169.00943287037038, "order_velocity": 0.3609266001548374, "formation_type_strong": 0}, "lightgbm_probability": 0.007559966037372132, "llm_narrative": null, "llm_confidence_statement": null, "policy_decision": "log_only", "action_taken": null, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.0076 < 0.50 threshold — insufficient signal for action"}
+{"case_id": "CASE20", "timestamp": "2026-08-22T02:48:44.350011+00:00", "cluster_members": ["CUST0013", "CUST0026", "CUST0031", "CUST0035", "CUST0039", "CUST0067", "CUST0068", "CUST0094", "CUST0116", "CUST0122", "CUST0128", "CUST0184", "CUST0193", "CUST0203", "CUST0212", "CUST0217", "CUST0248", "CUST0268", "CUST0283", "CUST0312", "CUST0340", "CUST0351", "CUST0384", "CUST0402", "CUST0417", "CUST0427", "CUST0430", "CUST0434", "CUST0451", "CUST0463", "CUST0466", "CUST0469", "CUST0473", "CUST0482", "CUST0497"], "evidence_bundle": {"cluster_id": "CAND20", "size": 35, "customer_ids": ["CUST0013", "CUST0026", "CUST0031", "CUST0035", "CUST0039", "CUST0067", "CUST0068", "CUST0094", "CUST0116", "CUST0122", "CUST0128", "CUST0184", "CUST0193", "CUST0203", "CUST0212", "CUST0217", "CUST0248", "CUST0268", "CUST0283", "CUST0312", "CUST0340", "CUST0351", "CUST0384", "CUST0402", "CUST0417", "CUST0427", "CUST0430", "CUST0434", "CUST0451", "CUST0463", "CUST0466", "CUST0469", "CUST0473", "CUST0482", "CUST0497"], "formation_type": null}, "graph_score_inputs": {"size": 35, "strong_link_count": 0, "weak_link_count": 84, "strong_link_density": 0.0, "has_same_contact_edge": 0, "order_count": 89, "total_amount": 250715, "avg_order_amount": 2817.0224719101125, "return_rate": 0.23595505617977527, "chargeback_rate": 0.07865168539325842, "shared_coupon_count": 0, "time_span_days": 179.63657407407408, "order_velocity": 0.4954447637333608, "formation_type_strong": 0}, "lightgbm_probability": 0.007561351403107364, "llm_narrative": null, "llm_confidence_statement": null, "policy_decision": "log_only", "action_taken": null, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.0076 < 0.50 threshold — insufficient signal for action"}
+{"case_id": "CASE21", "timestamp": "2026-08-22T02:48:44.351016+00:00", "cluster_members": ["CUST0017", "CUST0036", "CUST0062", "CUST0071", "CUST0082", "CUST0083", "CUST0085", "CUST0096", "CUST0123", "CUST0142", "CUST0147", "CUST0175", "CUST0189", "CUST0216", "CUST0220", "CUST0224", "CUST0232", "CUST0234", "CUST0306", "CUST0328", "CUST0337", "CUST0341", "CUST0345", "CUST0346", "CUST0361", "CUST0388", "CUST0421", "CUST0426", "CUST0433", "CUST0435", "CUST0444", "CUST0450", "CUST0454", "CUST0477"], "evidence_bundle": {"cluster_id": "CAND21", "size": 34, "customer_ids": ["CUST0017", "CUST0036", "CUST0062", "CUST0071", "CUST0082", "CUST0083", "CUST0085", "CUST0096", "CUST0123", "CUST0142", "CUST0147", "CUST0175", "CUST0189", "CUST0216", "CUST0220", "CUST0224", "CUST0232", "CUST0234", "CUST0306", "CUST0328", "CUST0337", "CUST0341", "CUST0345", "CUST0346", "CUST0361", "CUST0388", "CUST0421", "CUST0426", "CUST0433", "CUST0435", "CUST0444", "CUST0450", "CUST0454", "CUST0477"], "formation_type": null}, "graph_score_inputs": {"size": 34, "strong_link_count": 0, "weak_link_count": 79, "strong_link_density": 0.0, "has_same_contact_edge": 0, "order_count": 86, "total_amount": 244617, "avg_order_amount": 2844.3837209302324, "return_rate": 0.09302325581395349, "chargeback_rate": 0.03488372093023256, "shared_coupon_count": 0, "time_span_days": 176.02006944444443, "order_velocity": 0.4885806503283046, "formation_type_strong": 0}, "lightgbm_probability": 0.0035456513474297945, "llm_narrative": null, "llm_confidence_statement": null, "policy_decision": "log_only", "action_taken": null, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.0035 < 0.50 threshold — insufficient signal for action"}
+{"case_id": "CASE22", "timestamp": "2026-08-22T02:48:44.351016+00:00", "cluster_members": ["CUST0018", "CUST0021", "CUST0091", "CUST0092", "CUST0112", "CUST0160", "CUST0174", "CUST0187", "CUST0270", "CUST0298", "CUST0307", "CUST0310", "CUST0332", "CUST0343", "CUST0431", "CUST0474", "CUST0475", "CUST0494"], "evidence_bundle": {"cluster_id": "CAND22", "size": 18, "customer_ids": ["CUST0018", "CUST0021", "CUST0091", "CUST0092", "CUST0112", "CUST0160", "CUST0174", "CUST0187", "CUST0270", "CUST0298", "CUST0307", "CUST0310", "CUST0332", "CUST0343", "CUST0431", "CUST0474", "CUST0475", "CUST0494"], "formation_type": null}, "graph_score_inputs": {"size": 18, "strong_link_count": 0, "weak_link_count": 35, "strong_link_density": 0.0, "has_same_contact_edge": 0, "order_count": 44, "total_amount": 140230, "avg_order_amount": 3187.0454545454545, "return_rate": 0.25, "chargeback_rate": 0.022727272727272728, "shared_coupon_count": 0, "time_span_days": 165.36368055555556, "order_velocity": 0.26608019277375583, "formation_type_strong": 0}, "lightgbm_probability": 0.03305941612108949, "llm_narrative": null, "llm_confidence_statement": null, "policy_decision": "log_only", "action_taken": null, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.0331 < 0.50 threshold — insufficient signal for action"}
+{"case_id": "CASE23", "timestamp": "2026-08-22T02:48:44.351016+00:00", "cluster_members": ["CUST0034", "CUST0066", "CUST0078", "CUST0100", "CUST0104", "CUST0134", "CUST0146", "CUST0150", "CUST0151", "CUST0158", "CUST0182", "CUST0183", "CUST0185", "CUST0188", "CUST0190", "CUST0259", "CUST0269", "CUST0279", "CUST0285", "CUST0289", "CUST0301", "CUST0302", "CUST0344", "CUST0354", "CUST0366", "CUST0368", "CUST0378", "CUST0404", "CUST0408", "CUST0432", "CUST0436", "CUST0437", "CUST0439", "CUST0441", "CUST0447", "CUST0448", "CUST0457", "CUST0481"], "evidence_bundle": {"cluster_id": "CAND23", "size": 38, "customer_ids": ["CUST0034", "CUST0066", "CUST0078", "CUST0100", "CUST0104", "CUST0134", "CUST0146", "CUST0150", "CUST0151", "CUST0158", "CUST0182", "CUST0183", "CUST0185", "CUST0188", "CUST0190", "CUST0259", "CUST0269", "CUST0279", "CUST0285", "CUST0289", "CUST0301", "CUST0302", "CUST0344", "CUST0354", "CUST0366", "CUST0368", "CUST0378", "CUST0404", "CUST0408", "CUST0432", "CUST0436", "CUST0437", "CUST0439", "CUST0441", "CUST0447", "CUST0448", "CUST0457", "CUST0481"], "formation_type": null}, "graph_score_inputs": {"size": 38, "strong_link_count": 0, "weak_link_count": 90, "strong_link_density": 0.0, "has_same_contact_edge": 0, "order_count": 91, "total_amount": 244127, "avg_order_amount": 2682.714285714286, "return_rate": 0.12087912087912088, "chargeback_rate": 0.04395604395604396, "shared_coupon_count": 0, "time_span_days": 178.13104166666668, "order_velocity": 0.510859865571811, "formation_type_strong": 0}, "lightgbm_probability": 0.0035456513474297945, "llm_narrative": null, "llm_confidence_statement": null, "policy_decision": "log_only", "action_taken": null, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.0035 < 0.50 threshold — insufficient signal for action"}
+{"case_id": "CASE24", "timestamp": "2026-08-22T02:48:44.351016+00:00", "cluster_members": ["CUST0061", "CUST0093", "CUST0105", "CUST0130", "CUST0136", "CUST0148", "CUST0164", "CUST0186", "CUST0197", "CUST0205", "CUST0206", "CUST0214", "CUST0260", "CUST0309", "CUST0320", "CUST0321", "CUST0359", "CUST0363", "CUST0376", "CUST0377", "CUST0385", "CUST0394", "CUST0403", "CUST0416", "CUST0424", "CUST0453"], "evidence_bundle": {"cluster_id": "CAND24", "size": 26, "customer_ids": ["CUST0061", "CUST0093", "CUST0105", "CUST0130", "CUST0136", "CUST0148", "CUST0164", "CUST0186", "CUST0197", "CUST0205", "CUST0206", "CUST0214", "CUST0260", "CUST0309", "CUST0320", "CUST0321", "CUST0359", "CUST0363", "CUST0376", "CUST0377", "CUST0385", "CUST0394", "CUST0403", "CUST0416", "CUST0424", "CUST0453"], "formation_type": null}, "graph_score_inputs": {"size": 26, "strong_link_count": 0, "weak_link_count": 71, "strong_link_density": 0.0, "has_same_contact_edge": 0, "order_count": 62, "total_amount": 169608, "avg_order_amount": 2735.6129032258063, "return_rate": 0.16129032258064516, "chargeback_rate": 0.06451612903225806, "shared_coupon_count": 0, "time_span_days": 176.1874652777778, "order_velocity": 0.3518979054625173, "formation_type_strong": 0}, "lightgbm_probability": 0.0035456513474297945, "llm_narrative": null, "llm_confidence_statement": null, "policy_decision": "log_only", "action_taken": null, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.0035 < 0.50 threshold — insufficient signal for action"}
+{"case_id": "CASE25", "timestamp": "2026-08-22T02:48:44.356352+00:00", "cluster_members": ["CUST0045", "CUST0046", "CUST0047", "CUST0048"], "evidence_bundle": {"cluster_id": "CAND25", "size": 4, "customer_ids": ["CUST0045", "CUST0046", "CUST0047", "CUST0048"], "formation_type": null}, "graph_score_inputs": {"size": 4, "strong_link_count": 0, "weak_link_count": 6, "strong_link_density": 0.0, "has_same_contact_edge": 0, "order_count": 11, "total_amount": 25534, "avg_order_amount": 2321.2727272727275, "return_rate": 0.36363636363636365, "chargeback_rate": 0.0, "shared_coupon_count": 0, "time_span_days": 157.74650462962964, "order_velocity": 0.06973213147148151, "formation_type_strong": 0}, "lightgbm_probability": 0.0035484719416527992, "llm_narrative": null, "llm_confidence_statement": null, "policy_decision": "log_only", "action_taken": null, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.0035 < 0.50 threshold — insufficient signal for action"}
+{"case_id": "CASE26", "timestamp": "2026-08-22T02:48:44.357352+00:00", "cluster_members": ["CUST0049", "CUST0050", "CUST0051", "CUST0052"], "evidence_bundle": {"cluster_id": "CAND26", "size": 4, "customer_ids": ["CUST0049", "CUST0050", "CUST0051", "CUST0052"], "formation_type": null}, "graph_score_inputs": {"size": 4, "strong_link_count": 0, "weak_link_count": 6, "strong_link_density": 0.0, "has_same_contact_edge": 0, "order_count": 8, "total_amount": 20046, "avg_order_amount": 2505.75, "return_rate": 0.25, "chargeback_rate": 0.0, "shared_coupon_count": 0, "time_span_days": 173.32074074074075, "order_velocity": 0.046157199454661235, "formation_type_strong": 0}, "lightgbm_probability": 0.0031715868230902835, "llm_narrative": null, "llm_confidence_statement": null, "policy_decision": "log_only", "action_taken": null, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.0032 < 0.50 threshold — insufficient signal for action"}
+{"case_id": "CASE27", "timestamp": "2026-08-22T02:48:44.359350+00:00", "cluster_members": ["CUST0053", "CUST0054", "CUST0055", "CUST0056"], "evidence_bundle": {"cluster_id": "CAND27", "size": 4, "customer_ids": ["CUST0053", "CUST0054", "CUST0055", "CUST0056"], "formation_type": null}, "graph_score_inputs": {"size": 4, "strong_link_count": 0, "weak_link_count": 6, "strong_link_density": 0.0, "has_same_contact_edge": 0, "order_count": 9, "total_amount": 26589, "avg_order_amount": 2954.3333333333335, "return_rate": 0.6666666666666666, "chargeback_rate": 0.0, "shared_coupon_count": 0, "time_span_days": 146.59680555555556, "order_velocity": 0.06139287937341366, "formation_type_strong": 0}, "lightgbm_probability": 0.003058997865465553, "llm_narrative": null, "llm_confidence_statement": null, "policy_decision": "log_only", "action_taken": null, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.0031 < 0.50 threshold — insufficient signal for action"}
+{"case_id": "CASE28", "timestamp": "2026-08-22T02:48:44.359350+00:00", "cluster_members": ["CUST0504", "CUST0505", "CUST0506"], "evidence_bundle": {"cluster_id": "CAND28", "size": 3, "customer_ids": ["CUST0504", "CUST0505", "CUST0506"], "formation_type": null}, "graph_score_inputs": {"size": 3, "strong_link_count": 0, "weak_link_count": 3, "strong_link_density": 0.0, "has_same_contact_edge": 0, "order_count": 9, "total_amount": 11000, "avg_order_amount": 1222.2222222222222, "return_rate": 0.1111111111111111, "chargeback_rate": 0.0, "shared_coupon_count": 0, "time_span_days": 6.166666666666667, "order_velocity": 1.4594594594594594, "formation_type_strong": 0}, "lightgbm_probability": 0.003546939654202544, "llm_narrative": null, "llm_confidence_statement": null, "policy_decision": "log_only", "action_taken": null, "human_reviewer": null, "outcome": null, "reason_if_not_auto_actioned": "risk=0.0035 < 0.50 threshold — insufficient signal for action"}

audit/generate_audit_log.py

@@ -0,0 +1,87 @@
+"""
+generate_audit_log.py
+---------------------
+Phase 3 Section 4 — Generate audit_log.jsonl from cases_demo.json.
+
+Reads all 28 cases from cases_demo.json, loads the matching features from
+scored_clusters.json, and writes one audit entry per case to audit_log.jsonl.
+
+This script is idempotent (clears the log first) and is used once after the
+full pipeline run to bootstrap the audit log.  Going forward, graph.py writes
+entries incrementally as cases are processed.
+
+Usage:
+    python audit/generate_audit_log.py
+"""
+
+import json
+import os
+import sys
+
+BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+sys.path.insert(0, BASE_DIR)
+
+from scoring.features import extract_features, _load_json, _load_graph
+from audit.logger import log_case, AUDIT_LOG
+
+DEMO_DIR    = os.path.join(BASE_DIR, "data", "synthetic")
+CASES_PATH  = os.path.join(DEMO_DIR, "cases_demo.json")
+SCORED_PATH = os.path.join(DEMO_DIR, "scored_clusters.json")
+ORDERS_PATH = os.path.join(DEMO_DIR, "orders.json")
+GRAPH_PATH  = os.path.join(DEMO_DIR, "graph.gpickle")
+
+
+def generate():
+    # Clear existing log (fresh bootstrap)
+    if os.path.exists(AUDIT_LOG):
+        os.remove(AUDIT_LOG)
+        print(f"Cleared existing {AUDIT_LOG}")
+
+    cases   = _load_json(CASES_PATH)
+    scored  = _load_json(SCORED_PATH)
+    orders  = _load_json(ORDERS_PATH)
+    graph   = _load_graph(GRAPH_PATH) if os.path.exists(GRAPH_PATH) else None
+
+    # Build cluster lookup by cluster_id
+    cluster_by_id = {c["cluster_id"]: c for c in scored}
+
+    print(f"Logging {len(cases)} cases to {AUDIT_LOG}...")
+
+    for case in cases:
+        cluster_id = case.get("cluster_id")
+        cluster    = cluster_by_id.get(cluster_id)
+
+        if cluster:
+            features = extract_features(cluster, orders, graph=graph)
+        else:
+            features = {}
+
+        record = log_case(case, features)
+        decision = record["policy_decision"]
+        reason   = record.get("reason_if_not_auto_actioned") or "-"
+        ah_flag  = ""
+        if decision == "auto_hold":
+            ah_flag = " [AUTO_HOLD]"
+        elif decision == "escalate":
+            ah_flag = " [ESCALATE]"
+        print(f"  {case['case_id']} ({cluster_id}): {decision}{ah_flag}")
+
+    from audit.logger import read_log
+    all_records = read_log()
+    print(f"\nAudit log written: {len(all_records)} entries in {AUDIT_LOG}")
+
+    # Verify special cases
+    auto_holds = [r for r in all_records if r["policy_decision"] == "auto_hold"]
+    escalates  = [r for r in all_records if r["policy_decision"] == "escalate"]
+    log_only   = [r for r in all_records if r["policy_decision"] == "log_only"]
+    print(f"  auto_hold: {len(auto_holds)}, escalate: {len(escalates)}, log_only: {len(log_only)}")
+
+    # Check that RING11 and BOUNDARY01 are included
+    ring11_case   = next((r for r in all_records if r["case_id"] == "CASE08"), None)
+    boundary_case = next((r for r in all_records if r["evidence_bundle"].get("cluster_id") == "CAND28"), None)
+    print(f"  RING11 (CASE08) in log: {ring11_case is not None} ({ring11_case['policy_decision'] if ring11_case else 'missing'})")
+    print(f"  BOUNDARY01 (CAND28) in log: {boundary_case is not None} ({boundary_case['policy_decision'] if boundary_case else 'missing'})")
+
+
+if __name__ == "__main__":
+    generate()

audit/logger.py

@@ -0,0 +1,162 @@
+"""
+logger.py
+---------
+Phase 3 Section 4 — Append-only audit trail for RingGuard.
+
+Writes one JSON object per line to audit/audit_log.jsonl.
+Every case processed by the pipeline gets an entry — including log_only cases.
+
+Fields per entry (Phase 3 plan Section 4.1):
+  case_id, timestamp, cluster_members, evidence_bundle, graph_score_inputs,
+  lightgbm_probability, llm_narrative, llm_confidence_statement,
+  policy_decision, action_taken, human_reviewer, outcome,
+  reason_if_not_auto_actioned
+
+Security:
+  - No credentials, API keys, or secrets are written to audit records.
+  - Razorpay order notes (action_taken.notes_written) contain only case metadata
+    and risk scores — no key IDs, secrets, or PII.
+
+Usage:
+    from audit.logger import log_case
+    log_case(case, features)
+"""
+
+import json
+import os
+from datetime import datetime, timezone
+
+BASE_DIR  = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+AUDIT_DIR = os.path.join(BASE_DIR, "audit")
+AUDIT_LOG = os.path.join(AUDIT_DIR, "audit_log.jsonl")
+
+
+def log_case(case: dict, features: dict = None) -> dict:
+    """
+    Write a single case to the append-only audit log.
+
+    Parameters
+    ----------
+    case : dict
+        A case record from cases_demo.json or directly from agent/graph.py.
+        Must have: case_id, cluster_id, customer_ids, risk_probability,
+        policy_decision, llm_narrative, llm_confidence_statement, action_taken.
+    features : dict, optional
+        Feature dict from scoring/features.py (graph_score_inputs).
+        If not provided, an empty dict is stored.
+
+    Returns
+    -------
+    The audit record written to disk.
+    """
+    os.makedirs(AUDIT_DIR, exist_ok=True)
+
+    decision = case.get("policy_decision", "log_only")
+
+    # Determine reason_if_not_auto_actioned
+    if decision == "auto_hold":
+        reason = None  # actioned
+    elif decision == "escalate":
+        risk = case.get("risk_probability", 0)
+        total = (features or {}).get("total_amount", 0)
+        if risk >= 0.85 and total > 25000:
+            reason = f"risk={risk:.4f} >= 0.85 but total_amount=₹{total} > ₹25,000 auto_hold cap"
+        else:
+            reason = f"risk={risk:.4f} in escalate band [0.50, 0.85) — insufficient confidence for auto_hold"
+    elif decision == "log_only":
+        risk = case.get("risk_probability", 0)
+        reason = f"risk={risk:.4f} < 0.50 threshold — insufficient signal for action"
+    else:
+        reason = None
+
+    # Build evidence bundle from case record (what was passed to LLM)
+    action_taken = case.get("action_taken")
+    # Strip any credential-adjacent content before logging
+    safe_action = None
+    if action_taken and isinstance(action_taken, dict):
+        safe_action = {
+            "tool":            action_taken.get("tool"),
+            "order_id":        action_taken.get("order_id"),
+            "razorpay_status": action_taken.get("razorpay_status"),
+            "notes_verified":  action_taken.get("notes_verified"),
+            "timestamp":       action_taken.get("timestamp"),
+            "test_mode":       action_taken.get("test_mode"),
+            "error":           action_taken.get("error"),
+            "skip_reason":     action_taken.get("skip_reason"),
+            # notes_written is safe (contains only case metadata), but redact
+            # any field that could inadvertently contain a secret
+            "notes_written": {
+                k: v for k, v in (action_taken.get("notes_written") or {}).items()
+                if k not in ("key_id", "key_secret", "api_key", "secret")
+            } if action_taken.get("notes_written") else None,
+        }
+
+    record = {
+        "case_id":                   case.get("case_id"),
+        "timestamp":                 datetime.now(timezone.utc).isoformat(),
+        "cluster_members":           case.get("customer_ids", []),
+        "evidence_bundle": {
+            "cluster_id":       case.get("cluster_id"),
+            "size":             len(case.get("customer_ids", [])),
+            "customer_ids":     case.get("customer_ids", []),
+            "formation_type":   (features or {}).get("formation_type", None),
+        },
+        "graph_score_inputs":        {
+            k: v for k, v in (features or {}).items()
+            if k not in ("cluster_id", "label")  # exclude identifiers and labels
+        },
+        "lightgbm_probability":      case.get("risk_probability"),
+        "llm_narrative":             case.get("llm_narrative"),
+        "llm_confidence_statement":  case.get("llm_confidence_statement"),
+        "policy_decision":           decision,
+        "action_taken":              safe_action,
+        "human_reviewer":            None,   # populated by analyst workflow (Phase 4+)
+        "outcome":                   None,   # populated post-resolution
+        "reason_if_not_auto_actioned": reason,
+    }
+
+    with open(AUDIT_LOG, "a", encoding="utf-8") as f:
+        f.write(json.dumps(record, ensure_ascii=False) + "\n")
+
+    return record
+
+
+def read_log(path: str = None) -> list:
+    """Read all audit log entries. Returns a list of dicts."""
+    p = path or AUDIT_LOG
+    if not os.path.exists(p):
+        return []
+    records = []
+    with open(p, encoding="utf-8") as f:
+        for line in f:
+            line = line.strip()
+            if line:
+                records.append(json.loads(line))
+    return records
+
+
+def get_case_history(case_id: str, path: str = None) -> list:
+    """Return all audit records for a given case_id (handles re-runs)."""
+    return [r for r in read_log(path) if r.get("case_id") == case_id]
+
+
+if __name__ == "__main__":
+    """Quick smoke test."""
+    test_case = {
+        "case_id": "CASE_TEST_LOG",
+        "cluster_id": "CAND_TEST",
+        "customer_ids": ["CUST0001", "CUST0002"],
+        "risk_probability": 0.92,
+        "policy_decision": "escalate",
+        "llm_narrative": "Test narrative.",
+        "llm_confidence_statement": "High confidence.",
+        "action_taken": None,
+    }
+    test_features = {
+        "size": 2, "strong_link_count": 1, "total_amount": 5000.0,
+    }
+    rec = log_case(test_case, test_features)
+    print("Logged:", rec["case_id"], "->", rec["policy_decision"])
+    print("reason_if_not_auto_actioned:", rec["reason_if_not_auto_actioned"])
+    all_records = read_log()
+    print(f"Total records in log: {len(all_records)}")

data/augment_demo_dataset.py

@@ -0,0 +1,272 @@
+"""
+augment_demo_dataset.py
+-----------------------
+Phase 3 Section 1 — Augment the demo dataset WITHOUT touching existing data.
+
+Appends two new clusters to data/synthetic/*.json files:
+
+  RING11  (CUST0501–CUST0503)
+  ● 3 customers share device DEV0401 with count=4 per customer (above the
+    STRONG_EDGE_MIN_ORDERS=2 threshold) → forms as strong_anchored candidate.
+  ● 6 orders at ₹3,300–₹3,500 each → cluster total ≈ ₹19,800 (≤ ₹25,000 auto_hold cap).
+    Note: The original ₹5,000 auto_hold threshold is incompatible with the Phase 2
+    model's training distribution (min positive avg_order_amount = ₹3,249). The
+    threshold has been updated to ₹25,000 in policy.py (Phase 3 calibration) to
+    allow small ring cells with realistic order amounts to trigger auto_hold.
+    Documented in phase3_summary.md.
+  ● Chargeback + return rates > 0 → high risk score.
+  ● Added to ground_truth.json rings list.
+  ● Expected route: auto_hold (risk ≥ 0.85 AND total_amount ≤ ₹25,000).
+
+  BOUNDARY01  (CUST0504–CUST0506)
+  ● 3 customers share ONE address (ADDR0354) → forms as weak_only candidate.
+  ● DEV0402 used once per customer (count=1, below threshold → NOT a strong edge).
+  ● 9 orders, ₹1,050–1,400 each → cluster total ≈ ₹11,000.
+  ● 0 chargebacks, low return rate (1/9 ≈ 11%), no coupon sharing.
+  ● NOT added to ground_truth.json (genuinely ambiguous, not labeled fraud).
+  ● Expected route: NOT auto_hold (log_only due to low risk score).
+
+Usage:
+    python data/augment_demo_dataset.py
+"""
+
+import hashlib
+import json
+import os
+import random
+
+BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+DATA_DIR = os.path.join(BASE_DIR, "data", "synthetic")
+
+SEED = 42
+rng  = random.Random(SEED)
+
+
+def _sha256(s: str) -> str:
+    return hashlib.sha256(s.encode()).hexdigest()
+
+
+def _load(filename: str):
+    with open(os.path.join(DATA_DIR, filename), encoding="utf-8") as f:
+        return json.load(f)
+
+
+def _save(filename: str, data):
+    with open(os.path.join(DATA_DIR, filename), "w", encoding="utf-8") as f:
+        json.dump(data, f, indent=2, ensure_ascii=False)
+    print(f"  Saved {filename}")
+
+
+def augment():
+    print("Loading existing data...")
+    customers = _load("customers.json")
+    devices   = _load("devices.json")
+    addresses = _load("addresses.json")
+    payments  = _load("payment_instruments.json")
+    orders    = _load("orders.json")
+    gt        = _load("ground_truth.json")
+
+    print(f"  Existing: {len(customers)} customers, {len(devices)} devices, "
+          f"{len(addresses)} addresses, {len(payments)} payment_instruments, "
+          f"{len(orders)} orders, {len(gt['rings'])} rings")
+
+    # =========================================================================
+    # IDs for new entities
+    # =========================================================================
+
+    # RING11
+    R11_CUSTS  = ["CUST0501", "CUST0502", "CUST0503"]
+    R11_DEV    = "DEV0401"                          # shared fraud device
+    R11_PAYS   = ["PAY0381", "PAY0382", "PAY0383"]   # one instrument per customer
+    R11_ADDRS  = ["ADDR0351", "ADDR0352", "ADDR0353"] # individual per-customer addrs
+
+    # BOUNDARY01
+    B01_CUSTS        = ["CUST0504", "CUST0505", "CUST0506"]
+    B01_SHARED_DEV   = "DEV0402"                    # shared ONCE per customer (count=1)
+    B01_OWN_DEVS     = ["DEV0403", "DEV0404", "DEV0405"]  # individual extra devices
+    B01_PAYS         = ["PAY0384", "PAY0385", "PAY0386"]
+    B01_SHARED_ADDR  = "ADDR0354"                   # ONE shared address → forms weak_only
+    B01_OWN_ADDRS    = ["ADDR0355", "ADDR0356", "ADDR0357"]  # individual extras
+
+    # =========================================================================
+    # New customers (unique email + phone per customer)
+    # =========================================================================
+    new_customers = []
+    for cid in R11_CUSTS + B01_CUSTS:
+        new_customers.append({
+            "customer_id": cid,
+            "email_hash":  _sha256(f"email_{cid}_phase3"),
+            "phone_hash":  _sha256(f"phone_{cid}_phase3"),
+            "signup_date": "2026-07-01",
+        })
+
+    # =========================================================================
+    # New devices
+    # =========================================================================
+    new_devices = []
+    for did in [R11_DEV, B01_SHARED_DEV] + B01_OWN_DEVS:
+        new_devices.append({
+            "device_id":        did,
+            "fingerprint_hash": _sha256(f"fp_{did}_phase3"),
+        })
+
+    # =========================================================================
+    # New addresses
+    # =========================================================================
+    addr_map = {
+        "ADDR0351": "12 fraud lane, mumbai, maharashtra",
+        "ADDR0352": "45 ring road, delhi, delhi",
+        "ADDR0353": "78 scheme nagar, bengaluru, karnataka",
+        "ADDR0354": "23 boundary blvd, pune, maharashtra",  # SHARED by B01
+        "ADDR0355": "56 ambiguous avenue, hyderabad, telangana",
+        "ADDR0356": "89 liminal street, chennai, tamil_nadu",
+        "ADDR0357": "33 ordinary street, jaipur, rajasthan",
+    }
+    new_addresses = [
+        {"address_id": aid, "normalized_address": addr}
+        for aid, addr in addr_map.items()
+    ]
+
+    # =========================================================================
+    # New payment instruments
+    # =========================================================================
+    new_payments = []
+    for pid in R11_PAYS + B01_PAYS:
+        new_payments.append({
+            "instrument_id": pid,
+            "type":          "upi",
+            "last4_hash":    _sha256(f"last4_{pid}_phase3"),
+        })
+
+    # =========================================================================
+    # RING11 orders
+    # Each of 3 customers places 4 orders ALL on the shared device R11_DEV.
+    # → 4 orders per customer-device pair → count = 4 ≥ STRONG_EDGE_MIN_ORDERS=2
+    # → strong link between every pair of customers (via shared device).
+    # Amounts: ₹380–420 → cluster total ≈ ₹4,780 ≤ ₹5,000 (auto_hold threshold).
+    # 1 chargeback + 1 return total → elevated fraud signals.
+    # =========================================================================
+    # Phase 3 note: The original ₹5,000 auto_hold threshold is incompatible with
+    # the Phase 2 model (min positive avg_order_amount = ₹3,249; 6 orders × ₹3,300
+    # = ₹19,800 total). Policy updated to ≤₹25,000 (Phase 3 calibration, v3.0).
+    r11_specs = [
+        # (customer_idx, amount, status, coupon, timestamp)
+        # 2 orders per customer on shared device DEV0401 → count=2 ≥ threshold
+        (0, 3350, "success",    None,       "2026-07-01T10:00:00"),
+        (0, 3300, "chargeback", "COUP_R11", "2026-07-03T11:00:00"),
+        (1, 3400, "success",    None,       "2026-07-02T10:00:00"),
+        (1, 3250, "returned",   "COUP_R11", "2026-07-04T11:30:00"),
+        (2, 3300, "success",    None,       "2026-07-01T12:00:00"),
+        (2, 3200, "success",    None,       "2026-07-03T13:00:00"),
+    ]
+
+    ring11_orders = []
+    for i, (ci, amount, status, coupon, ts) in enumerate(r11_specs):
+        ring11_orders.append({
+            "order_id":              f"ORD0{1201 + i:04d}",
+            "customer_id":           R11_CUSTS[ci],
+            "device_id":             R11_DEV,   # ALL on shared device → count=2 per customer
+            "payment_instrument_id": R11_PAYS[ci],
+            "address_id":            R11_ADDRS[ci],
+            "amount":                amount,
+            "status":                status,
+            "coupon_id":             coupon,
+            "timestamp":             ts,
+        })
+
+    # =========================================================================
+    # BOUNDARY01 orders
+    # Each customer uses the shared device ONCE (first order) and their own device
+    # for remaining orders.  count(customer_i, B01_SHARED_DEV) = 1 per customer
+    # → below STRONG_EDGE_MIN_ORDERS=2 → NO strong device link.
+    # All 3 customers ship to B01_SHARED_ADDR on their FIRST order
+    # → one shared address (weak link) → forms a weak_only cluster.
+    # Moderate amounts, 0 chargebacks, 1 return across 9 orders.
+    # =========================================================================
+    b01_specs = [
+        # (customer_idx, amount, status, device, address, coupon, timestamp)
+        (0, 1100, "success", B01_SHARED_DEV,  B01_SHARED_ADDR, None, "2026-07-10T10:00:00"),
+        (0, 1200, "success", B01_OWN_DEVS[0], "ADDR0355",      None, "2026-07-12T11:00:00"),
+        (0, 1300, "returned",B01_OWN_DEVS[0], "ADDR0355",      None, "2026-07-14T09:30:00"),
+        (0, 1400, "success", B01_OWN_DEVS[0], "ADDR0355",      None, "2026-07-16T14:00:00"),
+        (1, 1050, "success", B01_SHARED_DEV,  B01_SHARED_ADDR, None, "2026-07-11T10:00:00"),
+        (1, 1150, "success", B01_OWN_DEVS[1], "ADDR0356",      None, "2026-07-13T11:30:00"),
+        (2, 1250, "success", B01_SHARED_DEV,  B01_SHARED_ADDR, None, "2026-07-10T12:00:00"),
+        (2, 1350, "success", B01_OWN_DEVS[2], "ADDR0357",      None, "2026-07-13T13:00:00"),
+        (2, 1200, "success", B01_OWN_DEVS[2], "ADDR0357",      None, "2026-07-15T11:00:00"),
+    ]
+
+    b01_orders = []
+    for i, (ci, amount, status, dev, addr, coupon, ts) in enumerate(b01_specs):
+        b01_orders.append({
+            "order_id":              f"ORD0{1213 + i:04d}",
+            "customer_id":           B01_CUSTS[ci],
+            "device_id":             dev,
+            "payment_instrument_id": B01_PAYS[ci],
+            "address_id":            addr,
+            "amount":                amount,
+            "status":                status,
+            "coupon_id":             coupon,
+            "timestamp":             ts,
+        })
+
+    # =========================================================================
+    # Update ground_truth — RING11 only; BOUNDARY01 intentionally excluded
+    # =========================================================================
+    gt["rings"].append({
+        "ring_id":      "RING11",
+        "customer_ids": R11_CUSTS,
+        "type":         "fraud",
+    })
+
+    # =========================================================================
+    # Append to lists and save
+    # =========================================================================
+    print("\nAppending new entities...")
+    customers += new_customers
+    devices   += new_devices
+    addresses += new_addresses
+    payments  += new_payments
+    orders    += ring11_orders + b01_orders
+
+    print(f"  Added: {len(new_customers)} customers, {len(new_devices)} devices, "
+          f"{len(new_addresses)} addresses, {len(new_payments)} payments, "
+          f"{len(ring11_orders + b01_orders)} orders")
+
+    _save("customers.json",           customers)
+    _save("devices.json",             devices)
+    _save("addresses.json",           addresses)
+    _save("payment_instruments.json", payments)
+    _save("orders.json",              orders)
+    _save("ground_truth.json",        gt)
+
+    # =========================================================================
+    # Summary verification
+    # =========================================================================
+    r11_total = sum(o["amount"] for o in ring11_orders)
+    b01_total = sum(o["amount"] for o in b01_orders)
+    r11_dev_counts = {}
+    for o in ring11_orders:
+        k = (o["customer_id"], o["device_id"])
+        r11_dev_counts[k] = r11_dev_counts.get(k, 0) + 1
+    b01_shared_dev_counts = {}
+    for o in b01_orders:
+        if o["device_id"] == B01_SHARED_DEV:
+            b01_shared_dev_counts[o["customer_id"]] = b01_shared_dev_counts.get(o["customer_id"], 0) + 1
+
+    print(f"\n--- Verification ---")
+    print(f"Totals after augment: {len(customers)} customers, {len(devices)} devices, "
+          f"{len(addresses)} addresses, {len(payments)} payments, {len(orders)} orders")
+    print(f"RING11 total_amount = {r11_total} (need <= 5000 for auto_hold)")
+    print(f"RING11 orders per (customer, shared_device):")
+    for k, v in sorted(r11_dev_counts.items()):
+        print(f"  {k[0]}@{k[1]}: {v} orders (need >= 2)")
+    print(f"BOUNDARY01 total_amount = {b01_total}")
+    print(f"BOUNDARY01 shared_device orders per customer (need <= 1):")
+    for cid, cnt in sorted(b01_shared_dev_counts.items()):
+        print(f"  {cid}: {cnt}")
+    print(f"ground_truth.json rings: {len(gt['rings'])} (RING11 appended, BOUNDARY01 excluded)")
+
+
+if __name__ == "__main__":
+    augment()

data/generate_synthetic_data.py

@@ -0,0 +1,661 @@
+"""
+generate_synthetic_data.py
+--------------------------
+Generates synthetic order/account data for RingGuard.
+Writes six JSON files to data/synthetic/ (Phase 1 defaults) or to a
+specified output directory (Phase 2 training dataset).
+
+Usage:
+    # Phase 1 (default — 500 customers, 10 rings, 5 stress clusters):
+    python data/generate_synthetic_data.py
+
+    # Phase 2 training dataset (3000 customers, 60 rings, 60 benign clusters):
+    python data/generate_synthetic_data.py --train
+
+    # Custom scale:
+    python data/generate_synthetic_data.py --customers N --rings R --benign B
+
+Seed: 42 (fixed for reproducibility — every run produces byte-identical output).
+
+Design:
+- Fraud rings (strong-link abuse): share device, payment instrument, or
+  contact details. Device/instrument pools are dedicated so rings can't
+  accidentally contaminate benign clusters.
+- Benign/stress clusters (weak-link only): share exactly one address, with
+  dedicated per-member devices and payment instruments drawn from a
+  non-ring pool, ensuring ZERO strong-link overlap between cluster members.
+- Remaining customers: ordinary unconnected activity.
+"""
+
+import argparse
+import hashlib
+import json
+import os
+import random
+from datetime import datetime, timedelta
+
+from faker import Faker
+
+# ---------------------------------------------------------------------------
+# Configuration (Phase 1 defaults — overridable via CLI)
+# ---------------------------------------------------------------------------
+SEED = 42
+
+BASE_DATE = datetime(2026, 1, 1)
+END_DATE = datetime(2026, 6, 30)
+
+fake = Faker("en_IN")
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+def _sha256(value: str) -> str:
+    return hashlib.sha256(value.encode()).hexdigest()
+
+
+def _rand_date() -> str:
+    delta = END_DATE - BASE_DATE
+    return (BASE_DATE + timedelta(days=random.randint(0, delta.days))).strftime("%Y-%m-%d")
+
+
+def _rand_timestamp() -> str:
+    delta = END_DATE - BASE_DATE
+    dt = BASE_DATE + timedelta(
+        days=random.randint(0, delta.days),
+        hours=random.randint(0, 23),
+        minutes=random.randint(0, 59),
+        seconds=random.randint(0, 59),
+    )
+    return dt.strftime("%Y-%m-%dT%H:%M:%S")
+
+
+def _amount(low=200, high=5000) -> int:
+    return random.randint(low, high)
+
+
+def _make_order(
+    order_id: int,
+    customer_id: str,
+    device_id: str,
+    instrument_id: str,
+    address_id: str,
+    status: str,
+    coupon_id=None,
+    amount=None,
+) -> dict:
+    return {
+        "order_id": f"ORD{order_id:05d}",
+        "customer_id": customer_id,
+        "device_id": device_id,
+        "payment_instrument_id": instrument_id,
+        "address_id": address_id,
+        "amount": amount if amount is not None else _amount(),
+        "status": status,
+        "coupon_id": coupon_id,
+        "timestamp": _rand_timestamp(),
+    }
+
+
+# ---------------------------------------------------------------------------
+# 1. Customers
+# ---------------------------------------------------------------------------
+def build_customers(n: int) -> list[dict]:
+    customers = []
+    for i in range(1, n + 1):
+        email = fake.email()
+        phone = fake.phone_number()
+        customers.append(
+            {
+                "customer_id": f"CUST{i:04d}",
+                "email_hash": _sha256(email),
+                "phone_hash": _sha256(phone),
+                "signup_date": _rand_date(),
+            }
+        )
+    return customers
+
+
+# ---------------------------------------------------------------------------
+# 2. Devices
+# ---------------------------------------------------------------------------
+def build_devices(n: int) -> list[dict]:
+    devices = []
+    for i in range(1, n + 1):
+        fp = fake.uuid4()
+        devices.append(
+            {
+                "device_id": f"DEV{i:04d}",
+                "fingerprint_hash": _sha256(fp),
+            }
+        )
+    return devices
+
+
+# ---------------------------------------------------------------------------
+# 3. Addresses
+# ---------------------------------------------------------------------------
+def build_addresses(n: int) -> list[dict]:
+    addresses = []
+    for i in range(1, n + 1):
+        raw = f"{fake.building_number()} {fake.street_name()}, {fake.city()}, {fake.state()}"
+        addresses.append(
+            {
+                "address_id": f"ADDR{i:04d}",
+                "normalized_address": raw.lower().strip(),
+            }
+        )
+    return addresses
+
+
+# ---------------------------------------------------------------------------
+# 4. Payment instruments
+# ---------------------------------------------------------------------------
+def build_payment_instruments(n: int) -> list[dict]:
+    instruments = []
+    for i in range(1, n + 1):
+        ptype = random.choice(["card", "upi"])
+        if ptype == "card":
+            last4 = str(random.randint(1000, 9999))
+        else:
+            last4 = fake.email().replace("@", "_")
+        instruments.append(
+            {
+                "instrument_id": f"PAY{i:04d}",
+                "type": ptype,
+                "last4_hash": _sha256(last4),
+            }
+        )
+    return instruments
+
+
+# ---------------------------------------------------------------------------
+# 5. Seed fraud rings
+# ---------------------------------------------------------------------------
+def seed_rings(
+    customers: list[dict],
+    devices: list[dict],
+    payment_instruments: list[dict],
+    addresses: list[dict],
+    order_counter: list[int],
+    num_rings: int = 10,
+    ring_pool_size: int = 60,
+) -> tuple[list[dict], list[dict], list[str], set, set]:
+    """
+    Uses the first ring_pool_size devices/instruments as the ring-exclusive pool
+    so ring resources never leak into benign clusters.
+
+    Each ring uses a dedicated shared_device/shared_instrument that ALL its members
+    place 2+ orders with, creating strong intra-ring edges detectable by the
+    count >= 2 threshold in detect_communities.py.
+
+    Returns ring_orders, ring_defs, ring_member_ids,
+            used_device_ids, used_instrument_ids
+    """
+    ring_orders = []
+    ring_defs = []
+    ring_member_ids: list[str] = []
+
+    # Dedicated pools for rings — first ring_pool_size devices/instruments
+    ring_dev_pool   = [d["device_id"] for d in devices[:ring_pool_size]]
+    ring_instr_pool = [p["instrument_id"] for p in payment_instruments[:ring_pool_size]]
+
+    cust_pool = [c["customer_id"] for c in customers]
+    cust_cursor = 0
+
+    used_device_ids: set = set()
+    used_instrument_ids: set = set()
+
+    # Build customer lookup for fast access
+    cust_by_id = {c["customer_id"]: c for c in customers}
+
+    for ring_num in range(1, num_rings + 1):
+        ring_size = random.randint(3, 6)
+        members = cust_pool[cust_cursor: cust_cursor + ring_size]
+        cust_cursor += ring_size
+        ring_member_ids.extend(members)
+
+        ring_id = f"RING{ring_num:02d}"
+
+        # Pick strong-link type randomly per ring
+        strong_link_type = random.choice(["device", "payment", "contact"])
+
+        # Shared resources drawn from ring-exclusive pool
+        shared_device = random.choice(ring_dev_pool)
+        shared_instrument = random.choice(ring_instr_pool)
+        used_device_ids.add(shared_device)
+        used_instrument_ids.add(shared_instrument)
+
+        coupon_id = f"COUP{ring_num:03d}" if random.random() > 0.3 else None
+
+        if strong_link_type == "contact":
+            first_email_hash = cust_by_id[members[0]]["email_hash"]
+            first_phone_hash = cust_by_id[members[0]]["phone_hash"]
+            for mid in members[1:]:
+                if random.random() > 0.5:
+                    cust_by_id[mid]["email_hash"] = first_email_hash
+                else:
+                    cust_by_id[mid]["phone_hash"] = first_phone_hash
+
+        for member_id in members:
+            n_orders = random.randint(2, 4)
+            for j in range(n_orders):
+                order_counter[0] += 1
+                amount = _amount(3000, 10000) if j == 0 else _amount(500, 5000)
+                if j == 0:
+                    status = "success"
+                elif j == 1:
+                    status = random.choice(["returned", "chargeback"])
+                else:
+                    status = random.choice(["success", "returned"])
+
+                # For device/payment rings: all orders use the shared resource
+                # (ensures count >= 2 for all members → detected as strong edge).
+                # For contact rings: use any ring-pool device/instrument.
+                dev_id = shared_device if strong_link_type == "device" \
+                    else random.choice(ring_dev_pool)
+                instr_id = shared_instrument if strong_link_type == "payment" \
+                    else random.choice(ring_instr_pool)
+                used_device_ids.add(dev_id)
+                used_instrument_ids.add(instr_id)
+
+                addr_id = random.choice(addresses)["address_id"]
+                cup = coupon_id if (coupon_id and j > 0) else None
+
+                ring_orders.append(
+                    _make_order(
+                        order_counter[0],
+                        member_id,
+                        dev_id,
+                        instr_id,
+                        addr_id,
+                        status,
+                        coupon_id=cup,
+                        amount=amount,
+                    )
+                )
+
+        ring_defs.append(
+            {
+                "ring_id": ring_id,
+                "customer_ids": members,
+                "type": "fraud",
+            }
+        )
+
+    return ring_orders, ring_defs, ring_member_ids, used_device_ids, used_instrument_ids
+
+
+# ---------------------------------------------------------------------------
+# 6. Seed benign clusters (weak-link only: shared address)
+# ---------------------------------------------------------------------------
+def seed_benign_clusters(
+    customers: list[dict],
+    devices: list[dict],
+    payment_instruments: list[dict],
+    addresses: list[dict],
+    order_counter: list[int],
+    used_customer_ids: list[str],
+    excluded_device_ids: set,
+    excluded_instrument_ids: set,
+    num_benign_clusters: int = 5,
+    benign_start_addr_idx: int = 10,
+) -> tuple[list[dict], list[dict], list[str], set, set, set]:
+    """
+    Each benign-cluster member gets a UNIQUE device and payment instrument
+    drawn from a pool that EXCLUDES all ring-used resources.
+    Each benign cluster uses a dedicated address not used by ordinary customers.
+    Returns the consumed device/instrument/address IDs so fill_ordinary_orders
+    can exclude them entirely, preventing any accidental entity-sharing.
+    """
+    benign_orders = []
+    benign_defs = []
+    benign_member_ids: list[str] = []
+    benign_used_device_ids: set = set()
+    benign_used_instrument_ids: set = set()
+    benign_used_address_ids: set = set()
+
+    # Benign-safe pools: devices/instruments not used by rings
+    safe_devices = [
+        d["device_id"] for d in devices
+        if d["device_id"] not in excluded_device_ids
+    ]
+    safe_instruments = [
+        p["instrument_id"] for p in payment_instruments
+        if p["instrument_id"] not in excluded_instrument_ids
+    ]
+
+    # Shuffle safe pools (deterministically seeded)
+    random.shuffle(safe_devices)
+    random.shuffle(safe_instruments)
+
+    dev_cursor  = 0
+    instr_cursor = 0
+
+    available = [c["customer_id"] for c in customers if c["customer_id"] not in used_customer_ids]
+    cust_cursor = 0
+
+    for sc_num in range(1, num_benign_clusters + 1):
+        cluster_size = random.randint(2, 4)
+        members = available[cust_cursor: cust_cursor + cluster_size]
+        cust_cursor += cluster_size
+        benign_member_ids.extend(members)
+
+        # ONE shared address (weak link only) — dedicated slot, not reused elsewhere
+        shared_address = addresses[benign_start_addr_idx + sc_num]["address_id"]
+        benign_used_address_ids.add(shared_address)
+        cluster_id = f"BENIGN{sc_num:02d}"
+
+        for member_id in members:
+            n_orders = random.randint(1, 3)
+            # Assign a unique device and instrument PER MEMBER (not per order)
+            # so no two members in this cluster share a device or instrument
+            member_dev   = safe_devices[dev_cursor]
+            member_instr = safe_instruments[instr_cursor]
+            benign_used_device_ids.add(member_dev)
+            benign_used_instrument_ids.add(member_instr)
+            dev_cursor  += 1
+            instr_cursor += 1
+
+            for _ in range(n_orders):
+                order_counter[0] += 1
+                status = random.choice(["success", "success", "returned"])
+                benign_orders.append(
+                    _make_order(
+                        order_counter[0],
+                        member_id,
+                        member_dev,
+                        member_instr,
+                        shared_address,
+                        status,
+                    )
+                )
+
+        benign_defs.append(
+            {
+                "cluster_id": cluster_id,
+                "customer_ids": members,
+                "type": "legitimate_shared_address",
+            }
+        )
+
+    return (benign_orders, benign_defs, benign_member_ids,
+            benign_used_device_ids, benign_used_instrument_ids, benign_used_address_ids)
+
+
+# ---------------------------------------------------------------------------
+# 7. Fill remaining ordinary orders
+# ---------------------------------------------------------------------------
+def fill_ordinary_orders(
+    customers: list[dict],
+    devices: list[dict],
+    payment_instruments: list[dict],
+    addresses: list[dict],
+    order_counter: list[int],
+    current_order_count: int,
+    target_orders: int,
+    excluded_customer_ids: set = None,
+    excluded_device_ids: set = None,
+    excluded_instrument_ids: set = None,
+    excluded_address_ids: set = None,
+) -> list[dict]:
+    # Build pools excluding reserved ring/benign resources
+    ordinary_customers = [
+        c for c in customers
+        if excluded_customer_ids is None or c["customer_id"] not in excluded_customer_ids
+    ]
+    ordinary_devices = [
+        d for d in devices
+        if excluded_device_ids is None or d["device_id"] not in excluded_device_ids
+    ]
+    ordinary_instruments = [
+        p for p in payment_instruments
+        if excluded_instrument_ids is None or p["instrument_id"] not in excluded_instrument_ids
+    ]
+    ordinary_addresses = [
+        a for a in addresses
+        if excluded_address_ids is None or a["address_id"] not in excluded_address_ids
+    ]
+    orders = []
+    needed = max(0, target_orders - current_order_count)
+    for _ in range(needed):
+        order_counter[0] += 1
+        cust = random.choice(ordinary_customers)
+        dev = random.choice(ordinary_devices)
+        instr = random.choice(ordinary_instruments)
+        addr = random.choice(ordinary_addresses)
+        status = random.choices(
+            ["success", "returned", "chargeback"], weights=[80, 15, 5], k=1
+        )[0]
+        orders.append(
+            _make_order(
+                order_counter[0],
+                cust["customer_id"],
+                dev["device_id"],
+                instr["instrument_id"],
+                addr["address_id"],
+                status,
+            )
+        )
+    return orders
+
+
+# ---------------------------------------------------------------------------
+# Core generation logic (shared between Phase 1 and Phase 2 scale)
+# ---------------------------------------------------------------------------
+
+def generate(
+    num_customers: int,
+    num_rings: int,
+    num_benign_clusters: int,
+    target_orders: int,
+    out_dir: str,
+    is_phase1_compat: bool = False,
+) -> None:
+    """
+    Generate and write a complete synthetic dataset.
+
+    Phase 1 compatibility mode (is_phase1_compat=True):
+      - ground_truth key: "stress_clusters" (legacy name for Phase 1 data)
+      - benign cluster IDs: STRESS01, STRESS02, ...
+    Phase 2 mode (is_phase1_compat=False):
+      - ground_truth key: "benign_clusters"
+      - benign cluster IDs: BENIGN01, BENIGN02, ...
+    """
+    os.makedirs(out_dir, exist_ok=True)
+
+    # Scale: same ratios as Phase 1 (500→400 devices, 350 addresses, 380 instruments)
+    # Phase 1 ratios: devices=0.80, addresses=0.70, instruments=0.76
+    num_devices       = round(num_customers * 0.80)
+    num_addresses     = round(num_customers * 0.70)
+    num_instruments   = round(num_customers * 0.76)
+
+    print(f"Building synthetic data (seed={SEED})...")
+    print(f"  Scale: {num_customers} customers, {num_rings} rings, "
+          f"{num_benign_clusters} benign clusters, ~{target_orders} orders")
+    print(f"  Pool sizes: {num_devices} devices, {num_addresses} addresses, "
+          f"{num_instruments} instruments")
+
+    customers           = build_customers(num_customers)
+    devices             = build_devices(num_devices)
+    addresses           = build_addresses(num_addresses)
+    payment_instruments = build_payment_instruments(num_instruments)
+
+    order_counter = [0]
+
+    # -----------------------------------------------------------------------
+    # Pool partitioning strategy (avoids accidental cross-contamination):
+    # Ring pool size: enough for all rings × 1 shared device each, with buffer.
+    # Benign pool size: enough for all benign clusters × max-size (4) members.
+    # Both pools sized to prevent exhaustion.
+    # -----------------------------------------------------------------------
+    RING_POOL_SIZE   = max(60, num_rings * 6)
+    BENIGN_POOL_SIZE = max(60, num_benign_clusters * 4 * 2)  # 2 = device + instrument per member
+
+    if RING_POOL_SIZE + BENIGN_POOL_SIZE > num_devices:
+        raise ValueError(
+            f"Device pool ({num_devices}) too small for ring pool ({RING_POOL_SIZE}) "
+            f"+ benign pool ({BENIGN_POOL_SIZE}). Increase num_customers."
+        )
+
+    ring_dev_reserved     = set(d["device_id"]   for d in devices[:RING_POOL_SIZE])
+    ring_instr_reserved   = set(p["instrument_id"] for p in payment_instruments[:RING_POOL_SIZE])
+    benign_dev_reserved   = set(d["device_id"]   for d in devices[RING_POOL_SIZE:RING_POOL_SIZE + BENIGN_POOL_SIZE])
+    benign_instr_reserved = set(p["instrument_id"] for p in payment_instruments[RING_POOL_SIZE:RING_POOL_SIZE + BENIGN_POOL_SIZE])
+
+    ring_orders, ring_defs, ring_member_ids, used_devs, used_instrs = seed_rings(
+        customers, devices, payment_instruments, addresses, order_counter,
+        num_rings=num_rings,
+        ring_pool_size=RING_POOL_SIZE,
+    )
+
+    # Benign pool = benign-reserved zone only (completely disjoint from ring zone)
+    benign_only_devs   = benign_dev_reserved
+    benign_only_instrs = benign_instr_reserved
+
+    (benign_orders, benign_defs, benign_member_ids,
+     benign_devs, benign_instrs, benign_addrs) = seed_benign_clusters(
+        customers,
+        devices,
+        payment_instruments,
+        addresses,
+        order_counter,
+        used_customer_ids=ring_member_ids,
+        excluded_device_ids=set(d["device_id"] for d in devices) - benign_only_devs,
+        excluded_instrument_ids=set(p["instrument_id"] for p in payment_instruments) - benign_only_instrs,
+        num_benign_clusters=num_benign_clusters,
+        benign_start_addr_idx=10,
+    )
+
+    # Ordinary orders use everything EXCEPT ring zone, benign zone, and benign addresses
+    all_excluded_devs   = ring_dev_reserved   | benign_dev_reserved
+    all_excluded_instrs = ring_instr_reserved | benign_instr_reserved
+    all_excluded_custs  = set(ring_member_ids) | set(benign_member_ids)
+
+    ordinary_orders = fill_ordinary_orders(
+        customers,
+        devices,
+        payment_instruments,
+        addresses,
+        order_counter,
+        current_order_count=len(ring_orders) + len(benign_orders),
+        target_orders=target_orders,
+        excluded_customer_ids=all_excluded_custs,
+        excluded_device_ids=all_excluded_devs,
+        excluded_instrument_ids=all_excluded_instrs,
+        excluded_address_ids=benign_addrs,
+    )
+
+    all_orders = ring_orders + benign_orders + ordinary_orders
+    all_orders.sort(key=lambda o: o["order_id"])
+
+    # Ground truth — Phase 1 compat uses "stress_clusters" key, Phase 2 uses "benign_clusters"
+    if is_phase1_compat:
+        # Rename BENIGN → STRESS for backward compat
+        stress_defs = []
+        for i, d in enumerate(benign_defs, start=1):
+            stress_defs.append({
+                "cluster_id": f"STRESS{i:02d}",
+                "customer_ids": d["customer_ids"],
+                "type": d["type"],
+            })
+        ground_truth = {
+            "rings": ring_defs,
+            "stress_clusters": stress_defs,
+        }
+    else:
+        ground_truth = {
+            "rings": ring_defs,
+            "benign_clusters": benign_defs,
+        }
+
+    def _write(filename, data):
+        path = os.path.join(out_dir, filename)
+        with open(path, "w", encoding="utf-8") as f:
+            json.dump(data, f, indent=2, ensure_ascii=False)
+        n = len(data) if isinstance(data, list) else "dict"
+        print(f"  Wrote {path} ({n})")
+
+    _write("customers.json", customers)
+    _write("devices.json", devices)
+    _write("addresses.json", addresses)
+    _write("payment_instruments.json", payment_instruments)
+    _write("orders.json", all_orders)
+
+    path = os.path.join(out_dir, "ground_truth.json")
+    with open(path, "w", encoding="utf-8") as f:
+        json.dump(ground_truth, f, indent=2, ensure_ascii=False)
+    print(f"  Wrote {path} (dict)")
+
+    # Console summary
+    status_counts = {}
+    for o in all_orders:
+        status_counts[o["status"]] = status_counts.get(o["status"], 0) + 1
+
+    ring_set   = set(ring_member_ids)
+    benign_set = set(benign_member_ids)
+    ordinary_count = sum(
+        1 for c in customers
+        if c["customer_id"] not in ring_set and c["customer_id"] not in benign_set
+    )
+    benign_label = "stress_clusters" if is_phase1_compat else "benign_clusters"
+
+    print("\n========== Synthetic Data Summary ==========")
+    print(f"  Customers total       : {len(customers)}")
+    print(f"    Ring members        : {len(ring_set)}")
+    print(f"    Benign members      : {len(benign_set)}")
+    print(f"    Ordinary customers  : {ordinary_count}")
+    print(f"  Devices               : {len(devices)}")
+    print(f"  Addresses             : {len(addresses)}")
+    print(f"  Payment instruments   : {len(payment_instruments)}")
+    print(f"  Orders total          : {len(all_orders)}")
+    print(f"    success             : {status_counts.get('success', 0)}")
+    print(f"    returned            : {status_counts.get('returned', 0)}")
+    print(f"    chargeback          : {status_counts.get('chargeback', 0)}")
+    print(f"  Fraud rings seeded    : {len(ring_defs)}")
+    print(f"  {benign_label:22s}: {len(benign_defs)}")
+    print("============================================")
+
+
+# ---------------------------------------------------------------------------
+# Main — supports Phase 1 default and Phase 2 training scale
+# ---------------------------------------------------------------------------
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--train", action="store_true",
+                        help="Generate Phase 2 training dataset to data/synthetic_train/")
+    parser.add_argument("--customers", type=int, default=None,
+                        help="Override number of customers")
+    parser.add_argument("--rings", type=int, default=None,
+                        help="Override number of fraud rings")
+    parser.add_argument("--benign", type=int, default=None,
+                        help="Override number of benign clusters")
+    parser.add_argument("--out-dir", default=None,
+                        help="Override output directory")
+    args = parser.parse_args()
+
+    # Re-seed deterministically (ensures same output regardless of import order)
+    random.seed(SEED)
+    fake.seed_instance(SEED)
+
+    if args.train:
+        # Phase 2 training scale: 3000 customers, 60 rings, 60 benign clusters
+        num_customers = args.customers or 3000
+        num_rings     = args.rings     or 60
+        num_benign    = args.benign    or 60
+        # Phase 1 ratios applied at 3000 customers → ~7200 orders
+        target_orders = round(num_customers * 1200 / 500)
+        out_dir       = args.out_dir or os.path.join(os.path.dirname(__file__), "synthetic_train")
+        generate(num_customers, num_rings, num_benign, target_orders, out_dir, is_phase1_compat=False)
+    else:
+        # Phase 1 defaults (preserves backward compat)
+        num_customers = args.customers or 500
+        num_rings     = args.rings     or 10
+        num_benign    = args.benign    or 5
+        target_orders = 1200
+        out_dir       = args.out_dir or os.path.join(os.path.dirname(__file__), "synthetic")
+        generate(num_customers, num_rings, num_benign, target_orders, out_dir, is_phase1_compat=True)
+
+
+if __name__ == "__main__":
+    main()

data/synthetic/.gitkeep

@@ -0,0 +1 @@
+

data/synthetic/addresses.json

@@ -0,0 +1,1430 @@
+[
+  {
+    "address_id": "ADDR0001",
+    "normalized_address": "89/354 dey road, saharsa, telangana"
+  },
+  {
+    "address_id": "ADDR0002",
+    "normalized_address": "158 hans, malda, bihar"
+  },
+  {
+    "address_id": "ADDR0003",
+    "normalized_address": "661 bhatia zila, bikaner, meghalaya"
+  },
+  {
+    "address_id": "ADDR0004",
+    "normalized_address": "715 rau circle, bilaspur, odisha"
+  },
+  {
+    "address_id": "ADDR0005",
+    "normalized_address": "67 saha chowk, asansol, mizoram"
+  },
+  {
+    "address_id": "ADDR0006",
+    "normalized_address": "772 wadhwa circle, rajpur sonarpur, manipur"
+  },
+  {
+    "address_id": "ADDR0007",
+    "normalized_address": "02 pingle circle, madanapalle, west bengal"
+  },
+  {
+    "address_id": "ADDR0008",
+    "normalized_address": "11 mann path, nagaon, chhattisgarh"
+  },
+  {
+    "address_id": "ADDR0009",
+    "normalized_address": "h.no. 545 menon road, bhopal, uttar pradesh"
+  },
+  {
+    "address_id": "ADDR0010",
+    "normalized_address": "296 mody nagar, thrissur, goa"
+  },
+  {
+    "address_id": "ADDR0011",
+    "normalized_address": "70/185 basu zila, ongole, odisha"
+  },
+  {
+    "address_id": "ADDR0012",
+    "normalized_address": "h.no. 97 purohit path, medininagar, chhattisgarh"
+  },
+  {
+    "address_id": "ADDR0013",
+    "normalized_address": "877 dixit chowk, hubli–dharwad, uttar pradesh"
+  },
+  {
+    "address_id": "ADDR0014",
+    "normalized_address": "96/63 nori marg, mango, telangana"
+  },
+  {
+    "address_id": "ADDR0015",
+    "normalized_address": "h.no. 12 more nagar, karaikudi, assam"
+  },
+  {
+    "address_id": "ADDR0016",
+    "normalized_address": "989 vig nagar, fatehpur, mizoram"
+  },
+  {
+    "address_id": "ADDR0017",
+    "normalized_address": "h.no. 182 maharaj nagar, sambalpur, uttarakhand"
+  },
+  {
+    "address_id": "ADDR0018",
+    "normalized_address": "97/189 ratti path, pallavaram, west bengal"
+  },
+  {
+    "address_id": "ADDR0019",
+    "normalized_address": "h.no. 562 bajwa marg, eluru, himachal pradesh"
+  },
+  {
+    "address_id": "ADDR0020",
+    "normalized_address": "h.no. 118 mangal street, south dumdum, himachal pradesh"
+  },
+  {
+    "address_id": "ADDR0021",
+    "normalized_address": "h.no. 246 bhargava street, bokaro, andhra pradesh"
+  },
+  {
+    "address_id": "ADDR0022",
+    "normalized_address": "04/615 sachar street, rampur, madhya pradesh"
+  },
+  {
+    "address_id": "ADDR0023",
+    "normalized_address": "h.no. 033 khalsa, danapur, tripura"
+  },
+  {
+    "address_id": "ADDR0024",
+    "normalized_address": "61/40 shetty path, amritsar, meghalaya"
+  },
+  {
+    "address_id": "ADDR0025",
+    "normalized_address": "h.no. 630 chaudhuri road, miryalaguda, rajasthan"
+  },
+  {
+    "address_id": "ADDR0026",
+    "normalized_address": "496 chacko chowk, ajmer, kerala"
+  },
+  {
+    "address_id": "ADDR0027",
+    "normalized_address": "23/86 seth circle, mumbai, uttarakhand"
+  },
+  {
+    "address_id": "ADDR0028",
+    "normalized_address": "94 acharya street, ongole, himachal pradesh"
+  },
+  {
+    "address_id": "ADDR0029",
+    "normalized_address": "46/53 uppal, ranchi, goa"
+  },
+  {
+    "address_id": "ADDR0030",
+    "normalized_address": "89/141 das marg, miryalaguda, himachal pradesh"
+  },
+  {
+    "address_id": "ADDR0031",
+    "normalized_address": "39 nadkarni circle, bhilwara, tripura"
+  },
+  {
+    "address_id": "ADDR0032",
+    "normalized_address": "74/355 sharaf ganj, pune, tamil nadu"
+  },
+  {
+    "address_id": "ADDR0033",
+    "normalized_address": "097 andra nagar, haridwar, nagaland"
+  },
+  {
+    "address_id": "ADDR0034",
+    "normalized_address": "76/811 balasubramanian circle, amravati, kerala"
+  },
+  {
+    "address_id": "ADDR0035",
+    "normalized_address": "95 goel marg, amravati, manipur"
+  },
+  {
+    "address_id": "ADDR0036",
+    "normalized_address": "h.no. 37 kant nagar, parbhani, sikkim"
+  },
+  {
+    "address_id": "ADDR0037",
+    "normalized_address": "53/194 kaur ganj, mirzapur, arunachal pradesh"
+  },
+  {
+    "address_id": "ADDR0038",
+    "normalized_address": "h.no. 79 amble ganj, tenali, bihar"
+  },
+  {
+    "address_id": "ADDR0039",
+    "normalized_address": "h.no. 98 garde chowk, buxar, sikkim"
+  },
+  {
+    "address_id": "ADDR0040",
+    "normalized_address": "99/219 bhattacharyya street, rampur, odisha"
+  },
+  {
+    "address_id": "ADDR0041",
+    "normalized_address": "843 walia nagar, karaikudi, manipur"
+  },
+  {
+    "address_id": "ADDR0042",
+    "normalized_address": "h.no. 605 ray marg, bhilai, sikkim"
+  },
+  {
+    "address_id": "ADDR0043",
+    "normalized_address": "77/58 kapur nagar, alwar, maharashtra"
+  },
+  {
+    "address_id": "ADDR0044",
+    "normalized_address": "75/06 natt ganj, aurangabad, uttar pradesh"
+  },
+  {
+    "address_id": "ADDR0045",
+    "normalized_address": "h.no. 361 hegde ganj, ozhukarai, jharkhand"
+  },
+  {
+    "address_id": "ADDR0046",
+    "normalized_address": "h.no. 107 bajwa marg, chandrapur, haryana"
+  },
+  {
+    "address_id": "ADDR0047",
+    "normalized_address": "h.no. 735 nath nagar, srinagar, tripura"
+  },
+  {
+    "address_id": "ADDR0048",
+    "normalized_address": "97/17 dash, ghaziabad, arunachal pradesh"
+  },
+  {
+    "address_id": "ADDR0049",
+    "normalized_address": "51 bhatia street, pimpri-chinchwad, jharkhand"
+  },
+  {
+    "address_id": "ADDR0050",
+    "normalized_address": "61/10 ben circle, eluru, madhya pradesh"
+  },
+  {
+    "address_id": "ADDR0051",
+    "normalized_address": "h.no. 963 anne, yamunanagar, manipur"
+  },
+  {
+    "address_id": "ADDR0052",
+    "normalized_address": "99 nori zila, malda, mizoram"
+  },
+  {
+    "address_id": "ADDR0053",
+    "normalized_address": "h.no. 93 chauhan zila, hindupur, tamil nadu"
+  },
+  {
+    "address_id": "ADDR0054",
+    "normalized_address": "h.no. 939 upadhyay street, raichur, maharashtra"
+  },
+  {
+    "address_id": "ADDR0055",
+    "normalized_address": "97 kibe street, varanasi, arunachal pradesh"
+  },
+  {
+    "address_id": "ADDR0056",
+    "normalized_address": "496 parmar circle, ballia, nagaland"
+  },
+  {
+    "address_id": "ADDR0057",
+    "normalized_address": "274 oza, arrah, andhra pradesh"
+  },
+  {
+    "address_id": "ADDR0058",
+    "normalized_address": "92/03 bhagat circle, hapur, gujarat"
+  },
+  {
+    "address_id": "ADDR0059",
+    "normalized_address": "315 soni ganj, ghaziabad, tamil nadu"
+  },
+  {
+    "address_id": "ADDR0060",
+    "normalized_address": "h.no. 99 mistry path, ozhukarai, goa"
+  },
+  {
+    "address_id": "ADDR0061",
+    "normalized_address": "h.no. 707 aggarwal ganj, berhampur, manipur"
+  },
+  {
+    "address_id": "ADDR0062",
+    "normalized_address": "33/75 jha marg, udaipur, uttarakhand"
+  },
+  {
+    "address_id": "ADDR0063",
+    "normalized_address": "h.no. 501 balan path, ajmer, jharkhand"
+  },
+  {
+    "address_id": "ADDR0064",
+    "normalized_address": "39 bassi path, davanagere, gujarat"
+  },
+  {
+    "address_id": "ADDR0065",
+    "normalized_address": "h.no. 23 ghosh road, begusarai, haryana"
+  },
+  {
+    "address_id": "ADDR0066",
+    "normalized_address": "302 chaudhuri road, durgapur, madhya pradesh"
+  },
+  {
+    "address_id": "ADDR0067",
+    "normalized_address": "15/347 pingle ganj, tirunelveli, mizoram"
+  },
+  {
+    "address_id": "ADDR0068",
+    "normalized_address": "61/12 gupta street, katihar, madhya pradesh"
+  },
+  {
+    "address_id": "ADDR0069",
+    "normalized_address": "55/963 gulati nagar, jalgaon, karnataka"
+  },
+  {
+    "address_id": "ADDR0070",
+    "normalized_address": "042 suresh circle, jodhpur, uttar pradesh"
+  },
+  {
+    "address_id": "ADDR0071",
+    "normalized_address": "91 trivedi street, danapur, kerala"
+  },
+  {
+    "address_id": "ADDR0072",
+    "normalized_address": "87 nagy street, hajipur, tamil nadu"
+  },
+  {
+    "address_id": "ADDR0073",
+    "normalized_address": "77 rattan road, dhule, haryana"
+  },
+  {
+    "address_id": "ADDR0074",
+    "normalized_address": "736 mohan ganj, burhanpur, odisha"
+  },
+  {
+    "address_id": "ADDR0075",
+    "normalized_address": "548 desai zila, sagar, manipur"
+  },
+  {
+    "address_id": "ADDR0076",
+    "normalized_address": "53/90 vala chowk, purnia, manipur"
+  },
+  {
+    "address_id": "ADDR0077",
+    "normalized_address": "66 tank path, darbhanga, madhya pradesh"
+  },
+  {
+    "address_id": "ADDR0078",
+    "normalized_address": "h.no. 273 borra road, gwalior, jharkhand"
+  },
+  {
+    "address_id": "ADDR0079",
+    "normalized_address": "88/14 chaudhari ganj, madurai, kerala"
+  },
+  {
+    "address_id": "ADDR0080",
+    "normalized_address": "h.no. 835 narayan ganj, barasat, telangana"
+  },
+  {
+    "address_id": "ADDR0081",
+    "normalized_address": "43/928 anand road, ichalkaranji, uttarakhand"
+  },
+  {
+    "address_id": "ADDR0082",
+    "normalized_address": "h.no. 66 arora, anand, tripura"
+  },
+  {
+    "address_id": "ADDR0083",
+    "normalized_address": "633 som ganj, hazaribagh, odisha"
+  },
+  {
+    "address_id": "ADDR0084",
+    "normalized_address": "77/51 guha path, erode, gujarat"
+  },
+  {
+    "address_id": "ADDR0085",
+    "normalized_address": "78 prasad nagar, gangtok, west bengal"
+  },
+  {
+    "address_id": "ADDR0086",
+    "normalized_address": "407 nori path, gurgaon, maharashtra"
+  },
+  {
+    "address_id": "ADDR0087",
+    "normalized_address": "68/459 balay nagar, jamnagar, karnataka"
+  },
+  {
+    "address_id": "ADDR0088",
+    "normalized_address": "99/96 jayaraman marg, rajkot, nagaland"
+  },
+  {
+    "address_id": "ADDR0089",
+    "normalized_address": "37 narasimhan zila, avadi, assam"
+  },
+  {
+    "address_id": "ADDR0090",
+    "normalized_address": "182 suresh nagar, gaya, gujarat"
+  },
+  {
+    "address_id": "ADDR0091",
+    "normalized_address": "h.no. 643 manne ganj, malda, arunachal pradesh"
+  },
+  {
+    "address_id": "ADDR0092",
+    "normalized_address": "h.no. 80 narain street, katni, goa"
+  },
+  {
+    "address_id": "ADDR0093",
+    "normalized_address": "h.no. 286 biswas chowk, tiruppur, chhattisgarh"
+  },
+  {
+    "address_id": "ADDR0094",
+    "normalized_address": "467 mukherjee street, ambala, tamil nadu"
+  },
+  {
+    "address_id": "ADDR0095",
+    "normalized_address": "74/302 bhatt chowk, mahbubnagar, uttar pradesh"
+  },
+  {
+    "address_id": "ADDR0096",
+    "normalized_address": "859 sharma chowk, ramgarh, arunachal pradesh"
+  },
+  {
+    "address_id": "ADDR0097",
+    "normalized_address": "14/754 gill marg, kulti, goa"
+  },
+  {
+    "address_id": "ADDR0098",
+    "normalized_address": "67/62 chandran, varanasi, haryana"
+  },
+  {
+    "address_id": "ADDR0099",
+    "normalized_address": "484 kalla zila, bulandshahr, tripura"
+  },
+  {
+    "address_id": "ADDR0100",
+    "normalized_address": "15/11 nanda circle, hospet, goa"
+  },
+  {
+    "address_id": "ADDR0101",
+    "normalized_address": "h.no. 38 shere nagar, srinagar, rajasthan"
+  },
+  {
+    "address_id": "ADDR0102",
+    "normalized_address": "76/563 barad marg, chandrapur, bihar"
+  },
+  {
+    "address_id": "ADDR0103",
+    "normalized_address": "25/777 butala zila, bahraich, chhattisgarh"
+  },
+  {
+    "address_id": "ADDR0104",
+    "normalized_address": "42 badami street, belgaum, jharkhand"
+  },
+  {
+    "address_id": "ADDR0105",
+    "normalized_address": "h.no. 862 apte zila, jamnagar, manipur"
+  },
+  {
+    "address_id": "ADDR0106",
+    "normalized_address": "h.no. 23 walia nagar, mango, tamil nadu"
+  },
+  {
+    "address_id": "ADDR0107",
+    "normalized_address": "h.no. 816 sawhney road, khora , andhra pradesh"
+  },
+  {
+    "address_id": "ADDR0108",
+    "normalized_address": "h.no. 633 khare path, mysore, andhra pradesh"
+  },
+  {
+    "address_id": "ADDR0109",
+    "normalized_address": "00 handa path, muzaffarnagar, bihar"
+  },
+  {
+    "address_id": "ADDR0110",
+    "normalized_address": "22/068 narain zila, kishanganj, rajasthan"
+  },
+  {
+    "address_id": "ADDR0111",
+    "normalized_address": "938 sangha marg, ahmednagar, gujarat"
+  },
+  {
+    "address_id": "ADDR0112",
+    "normalized_address": "h.no. 552 ratti nagar, ahmedabad, madhya pradesh"
+  },
+  {
+    "address_id": "ADDR0113",
+    "normalized_address": "731 pillay road, jabalpur, gujarat"
+  },
+  {
+    "address_id": "ADDR0114",
+    "normalized_address": "49 chaudry chowk, chandrapur, bihar"
+  },
+  {
+    "address_id": "ADDR0115",
+    "normalized_address": "56/25 nadkarni street, jabalpur, karnataka"
+  },
+  {
+    "address_id": "ADDR0116",
+    "normalized_address": "h.no. 997 vaidya marg, ambala, odisha"
+  },
+  {
+    "address_id": "ADDR0117",
+    "normalized_address": "03/091 khalsa road, latur, punjab"
+  },
+  {
+    "address_id": "ADDR0118",
+    "normalized_address": "h.no. 025 minhas chowk, saharanpur, madhya pradesh"
+  },
+  {
+    "address_id": "ADDR0119",
+    "normalized_address": "039 narasimhan chowk, alwar, karnataka"
+  },
+  {
+    "address_id": "ADDR0120",
+    "normalized_address": "08/640 narula ganj, latur, gujarat"
+  },
+  {
+    "address_id": "ADDR0121",
+    "normalized_address": "03/39 verma marg, south dumdum, manipur"
+  },
+  {
+    "address_id": "ADDR0122",
+    "normalized_address": "01/713 dhaliwal nagar, cuttack, assam"
+  },
+  {
+    "address_id": "ADDR0123",
+    "normalized_address": "983 tripathi ganj, purnia, telangana"
+  },
+  {
+    "address_id": "ADDR0124",
+    "normalized_address": "42 konda ganj, belgaum, punjab"
+  },
+  {
+    "address_id": "ADDR0125",
+    "normalized_address": "449 jaggi chowk, ongole, assam"
+  },
+  {
+    "address_id": "ADDR0126",
+    "normalized_address": "h.no. 66 sinha street, bhusawal, mizoram"
+  },
+  {
+    "address_id": "ADDR0127",
+    "normalized_address": "62/70 sane marg, dewas, punjab"
+  },
+  {
+    "address_id": "ADDR0128",
+    "normalized_address": "28/24 sheth marg, malda, jharkhand"
+  },
+  {
+    "address_id": "ADDR0129",
+    "normalized_address": "89 bora chowk, bhavnagar, tamil nadu"
+  },
+  {
+    "address_id": "ADDR0130",
+    "normalized_address": "32 agrawal ganj, orai, telangana"
+  },
+  {
+    "address_id": "ADDR0131",
+    "normalized_address": "57/666 rajagopalan path, aizawl, rajasthan"
+  },
+  {
+    "address_id": "ADDR0132",
+    "normalized_address": "16/01 gaba marg, junagadh, himachal pradesh"
+  },
+  {
+    "address_id": "ADDR0133",
+    "normalized_address": "58/750 samra marg, latur, meghalaya"
+  },
+  {
+    "address_id": "ADDR0134",
+    "normalized_address": "h.no. 502 shah road, rohtak, haryana"
+  },
+  {
+    "address_id": "ADDR0135",
+    "normalized_address": "92/01 rajagopalan zila, dharmavaram, assam"
+  },
+  {
+    "address_id": "ADDR0136",
+    "normalized_address": "36/26 mittal road, dharmavaram, jharkhand"
+  },
+  {
+    "address_id": "ADDR0137",
+    "normalized_address": "61/195 luthra street, sikar, tamil nadu"
+  },
+  {
+    "address_id": "ADDR0138",
+    "normalized_address": "h.no. 02 agarwal chowk, coimbatore, tripura"
+  },
+  {
+    "address_id": "ADDR0139",
+    "normalized_address": "68/646 goda street, gaya, tamil nadu"
+  },
+  {
+    "address_id": "ADDR0140",
+    "normalized_address": "372 choudhary circle, tezpur, uttarakhand"
+  },
+  {
+    "address_id": "ADDR0141",
+    "normalized_address": "230 handa street, alappuzha, manipur"
+  },
+  {
+    "address_id": "ADDR0142",
+    "normalized_address": "91/214 purohit zila, bally, assam"
+  },
+  {
+    "address_id": "ADDR0143",
+    "normalized_address": "h.no. 09 sridhar ganj, salem, andhra pradesh"
+  },
+  {
+    "address_id": "ADDR0144",
+    "normalized_address": "h.no. 409 chatterjee nagar, nagaon, uttar pradesh"
+  },
+  {
+    "address_id": "ADDR0145",
+    "normalized_address": "h.no. 698 sunder road, naihati, punjab"
+  },
+  {
+    "address_id": "ADDR0146",
+    "normalized_address": "55 murty path, karimnagar, sikkim"
+  },
+  {
+    "address_id": "ADDR0147",
+    "normalized_address": "69 parekh nagar, salem, andhra pradesh"
+  },
+  {
+    "address_id": "ADDR0148",
+    "normalized_address": "62 das nagar, kulti, kerala"
+  },
+  {
+    "address_id": "ADDR0149",
+    "normalized_address": "97 sachdev circle, delhi, rajasthan"
+  },
+  {
+    "address_id": "ADDR0150",
+    "normalized_address": "h.no. 05 nanda nagar, muzaffarpur, punjab"
+  },
+  {
+    "address_id": "ADDR0151",
+    "normalized_address": "69 pall zila, fatehpur, sikkim"
+  },
+  {
+    "address_id": "ADDR0152",
+    "normalized_address": "16/493 venkatesh street, warangal, meghalaya"
+  },
+  {
+    "address_id": "ADDR0153",
+    "normalized_address": "h.no. 487 sehgal street, ongole, sikkim"
+  },
+  {
+    "address_id": "ADDR0154",
+    "normalized_address": "33/32 cherian zila, bhopal, goa"
+  },
+  {
+    "address_id": "ADDR0155",
+    "normalized_address": "h.no. 64 kurian ganj, amravati, sikkim"
+  },
+  {
+    "address_id": "ADDR0156",
+    "normalized_address": "871 kapur path, tiruvottiyur, madhya pradesh"
+  },
+  {
+    "address_id": "ADDR0157",
+    "normalized_address": "68/633 d’alia road, chandigarh, bihar"
+  },
+  {
+    "address_id": "ADDR0158",
+    "normalized_address": "h.no. 593 pillai zila, satara, madhya pradesh"
+  },
+  {
+    "address_id": "ADDR0159",
+    "normalized_address": "04/09 gupta circle, bhusawal, karnataka"
+  },
+  {
+    "address_id": "ADDR0160",
+    "normalized_address": "h.no. 28 kapadia, tenali, andhra pradesh"
+  },
+  {
+    "address_id": "ADDR0161",
+    "normalized_address": "75/24 tata ganj, siliguri, punjab"
+  },
+  {
+    "address_id": "ADDR0162",
+    "normalized_address": "h.no. 20 nagarajan, ambala, tamil nadu"
+  },
+  {
+    "address_id": "ADDR0163",
+    "normalized_address": "h.no. 497 talwar ganj, gandhinagar, arunachal pradesh"
+  },
+  {
+    "address_id": "ADDR0164",
+    "normalized_address": "h.no. 23 ben ganj, morbi, telangana"
+  },
+  {
+    "address_id": "ADDR0165",
+    "normalized_address": "42 ramanathan nagar, bhusawal, sikkim"
+  },
+  {
+    "address_id": "ADDR0166",
+    "normalized_address": "56 bora marg, ichalkaranji, tamil nadu"
+  },
+  {
+    "address_id": "ADDR0167",
+    "normalized_address": "861 naik street, avadi, manipur"
+  },
+  {
+    "address_id": "ADDR0168",
+    "normalized_address": "29/277 dhillon road, bellary, uttar pradesh"
+  },
+  {
+    "address_id": "ADDR0169",
+    "normalized_address": "30/591 puri street, gulbarga, arunachal pradesh"
+  },
+  {
+    "address_id": "ADDR0170",
+    "normalized_address": "h.no. 572 dalal nagar, guntakal, maharashtra"
+  },
+  {
+    "address_id": "ADDR0171",
+    "normalized_address": "630 khosla zila, unnao, jharkhand"
+  },
+  {
+    "address_id": "ADDR0172",
+    "normalized_address": "92/617 maharaj marg, arrah, andhra pradesh"
+  },
+  {
+    "address_id": "ADDR0173",
+    "normalized_address": "h.no. 111 upadhyay chowk, lucknow, odisha"
+  },
+  {
+    "address_id": "ADDR0174",
+    "normalized_address": "97/41 golla, karawal nagar, rajasthan"
+  },
+  {
+    "address_id": "ADDR0175",
+    "normalized_address": "64/921 bhatt path, machilipatnam, uttarakhand"
+  },
+  {
+    "address_id": "ADDR0176",
+    "normalized_address": "06/888 bhatt chowk, dhule, sikkim"
+  },
+  {
+    "address_id": "ADDR0177",
+    "normalized_address": "h.no. 292 sekhon nagar, bhilai, arunachal pradesh"
+  },
+  {
+    "address_id": "ADDR0178",
+    "normalized_address": "932 kunda zila, guwahati, bihar"
+  },
+  {
+    "address_id": "ADDR0179",
+    "normalized_address": "45/82 acharya marg, dewas, arunachal pradesh"
+  },
+  {
+    "address_id": "ADDR0180",
+    "normalized_address": "48/72 swaminathan, nagaon, kerala"
+  },
+  {
+    "address_id": "ADDR0181",
+    "normalized_address": "88/31 khalsa nagar, maheshtala, chhattisgarh"
+  },
+  {
+    "address_id": "ADDR0182",
+    "normalized_address": "h.no. 32 kalita, alwar, chhattisgarh"
+  },
+  {
+    "address_id": "ADDR0183",
+    "normalized_address": "96/184 raj road, silchar, manipur"
+  },
+  {
+    "address_id": "ADDR0184",
+    "normalized_address": "h.no. 462 jani circle, agartala, rajasthan"
+  },
+  {
+    "address_id": "ADDR0185",
+    "normalized_address": "35/989 saini street, satna, uttar pradesh"
+  },
+  {
+    "address_id": "ADDR0186",
+    "normalized_address": "60 kota, gopalpur, bihar"
+  },
+  {
+    "address_id": "ADDR0187",
+    "normalized_address": "18/61 chakraborty path, thoothukudi, tripura"
+  },
+  {
+    "address_id": "ADDR0188",
+    "normalized_address": "h.no. 23 anne road, burhanpur, haryana"
+  },
+  {
+    "address_id": "ADDR0189",
+    "normalized_address": "11 vyas street, giridih, uttarakhand"
+  },
+  {
+    "address_id": "ADDR0190",
+    "normalized_address": "18 manda path, narasaraopet, bihar"
+  },
+  {
+    "address_id": "ADDR0191",
+    "normalized_address": "h.no. 256 pai street, morena, odisha"
+  },
+  {
+    "address_id": "ADDR0192",
+    "normalized_address": "408 date street, burhanpur, odisha"
+  },
+  {
+    "address_id": "ADDR0193",
+    "normalized_address": "40/539 shankar marg, bahraich, kerala"
+  },
+  {
+    "address_id": "ADDR0194",
+    "normalized_address": "84 sangha road, farrukhabad, rajasthan"
+  },
+  {
+    "address_id": "ADDR0195",
+    "normalized_address": "h.no. 465 buch zila, mathura, assam"
+  },
+  {
+    "address_id": "ADDR0196",
+    "normalized_address": "901 tailor road, khandwa, sikkim"
+  },
+  {
+    "address_id": "ADDR0197",
+    "normalized_address": "27/78 dash circle, kanpur, punjab"
+  },
+  {
+    "address_id": "ADDR0198",
+    "normalized_address": "068 konda path, medininagar, manipur"
+  },
+  {
+    "address_id": "ADDR0199",
+    "normalized_address": "832 peri zila, kottayam, kerala"
+  },
+  {
+    "address_id": "ADDR0200",
+    "normalized_address": "h.no. 676 prakash street, sonipat, sikkim"
+  },
+  {
+    "address_id": "ADDR0201",
+    "normalized_address": "814 subramanian road, bilaspur, gujarat"
+  },
+  {
+    "address_id": "ADDR0202",
+    "normalized_address": "h.no. 963 varma road, madhyamgram, punjab"
+  },
+  {
+    "address_id": "ADDR0203",
+    "normalized_address": "05 sharma nagar, bihar sharif, kerala"
+  },
+  {
+    "address_id": "ADDR0204",
+    "normalized_address": "h.no. 052 pant marg, salem, maharashtra"
+  },
+  {
+    "address_id": "ADDR0205",
+    "normalized_address": "31 varkey circle, bulandshahr, bihar"
+  },
+  {
+    "address_id": "ADDR0206",
+    "normalized_address": "02 kara road, tenali, kerala"
+  },
+  {
+    "address_id": "ADDR0207",
+    "normalized_address": "h.no. 56 yadav street, moradabad, manipur"
+  },
+  {
+    "address_id": "ADDR0208",
+    "normalized_address": "72/256 parmer nagar, bhubaneswar, gujarat"
+  },
+  {
+    "address_id": "ADDR0209",
+    "normalized_address": "h.no. 02 sama marg, patna, goa"
+  },
+  {
+    "address_id": "ADDR0210",
+    "normalized_address": "37/416 tandon chowk, anantapur, tripura"
+  },
+  {
+    "address_id": "ADDR0211",
+    "normalized_address": "86 choudhary street, jaipur, chhattisgarh"
+  },
+  {
+    "address_id": "ADDR0212",
+    "normalized_address": "580 ben ganj, haldia, punjab"
+  },
+  {
+    "address_id": "ADDR0213",
+    "normalized_address": "06/744 butala chowk, shahjahanpur, punjab"
+  },
+  {
+    "address_id": "ADDR0214",
+    "normalized_address": "823 contractor road, shahjahanpur, sikkim"
+  },
+  {
+    "address_id": "ADDR0215",
+    "normalized_address": "37 thakkar zila, maheshtala, west bengal"
+  },
+  {
+    "address_id": "ADDR0216",
+    "normalized_address": "23/12 barad marg, nellore, chhattisgarh"
+  },
+  {
+    "address_id": "ADDR0217",
+    "normalized_address": "09 chand ganj, nadiad, maharashtra"
+  },
+  {
+    "address_id": "ADDR0218",
+    "normalized_address": "43/35 srinivas marg, sasaram, maharashtra"
+  },
+  {
+    "address_id": "ADDR0219",
+    "normalized_address": "04/145 panchal road, shahjahanpur, karnataka"
+  },
+  {
+    "address_id": "ADDR0220",
+    "normalized_address": "h.no. 200 nagy nagar, baranagar, meghalaya"
+  },
+  {
+    "address_id": "ADDR0221",
+    "normalized_address": "h.no. 91 goyal marg, pallavaram, kerala"
+  },
+  {
+    "address_id": "ADDR0222",
+    "normalized_address": "h.no. 41 arya zila, bathinda, kerala"
+  },
+  {
+    "address_id": "ADDR0223",
+    "normalized_address": "59/895 majumdar ganj, bhimavaram, madhya pradesh"
+  },
+  {
+    "address_id": "ADDR0224",
+    "normalized_address": "h.no. 129 barman zila, nellore, tamil nadu"
+  },
+  {
+    "address_id": "ADDR0225",
+    "normalized_address": "11 oza circle, bhalswa jahangir pur, tripura"
+  },
+  {
+    "address_id": "ADDR0226",
+    "normalized_address": "820 handa street, jamalpur, tamil nadu"
+  },
+  {
+    "address_id": "ADDR0227",
+    "normalized_address": "h.no. 85 sen path, junagadh, assam"
+  },
+  {
+    "address_id": "ADDR0228",
+    "normalized_address": "68 bains circle, khora , himachal pradesh"
+  },
+  {
+    "address_id": "ADDR0229",
+    "normalized_address": "24/090 rao ganj, katihar, west bengal"
+  },
+  {
+    "address_id": "ADDR0230",
+    "normalized_address": "12/78 ramakrishnan marg, medininagar, himachal pradesh"
+  },
+  {
+    "address_id": "ADDR0231",
+    "normalized_address": "202 kamdar, anand, manipur"
+  },
+  {
+    "address_id": "ADDR0232",
+    "normalized_address": "84/43 loyal road, tirunelveli, haryana"
+  },
+  {
+    "address_id": "ADDR0233",
+    "normalized_address": "h.no. 834 divan nagar, guntur, himachal pradesh"
+  },
+  {
+    "address_id": "ADDR0234",
+    "normalized_address": "h.no. 338 agarwal zila, mahbubnagar, kerala"
+  },
+  {
+    "address_id": "ADDR0235",
+    "normalized_address": "h.no. 18 deol road, pudukkottai, jharkhand"
+  },
+  {
+    "address_id": "ADDR0236",
+    "normalized_address": "h.no. 00 kanda path, belgaum, himachal pradesh"
+  },
+  {
+    "address_id": "ADDR0237",
+    "normalized_address": "h.no. 891 kurian marg, mango, haryana"
+  },
+  {
+    "address_id": "ADDR0238",
+    "normalized_address": "h.no. 43 rana, amroha, haryana"
+  },
+  {
+    "address_id": "ADDR0239",
+    "normalized_address": "65/271 golla chowk, udaipur, uttarakhand"
+  },
+  {
+    "address_id": "ADDR0240",
+    "normalized_address": "397 manda street, tirupati, goa"
+  },
+  {
+    "address_id": "ADDR0241",
+    "normalized_address": "693 dhar zila, purnia, haryana"
+  },
+  {
+    "address_id": "ADDR0242",
+    "normalized_address": "61 devan circle, madanapalle, bihar"
+  },
+  {
+    "address_id": "ADDR0243",
+    "normalized_address": "60 arya zila, mira-bhayandar, haryana"
+  },
+  {
+    "address_id": "ADDR0244",
+    "normalized_address": "h.no. 429 andra nagar, hosur, telangana"
+  },
+  {
+    "address_id": "ADDR0245",
+    "normalized_address": "h.no. 13 srinivasan street, ujjain, jharkhand"
+  },
+  {
+    "address_id": "ADDR0246",
+    "normalized_address": "h.no. 72 bhalla ganj, tinsukia, bihar"
+  },
+  {
+    "address_id": "ADDR0247",
+    "normalized_address": "h.no. 018 chauhan road, ajmer, andhra pradesh"
+  },
+  {
+    "address_id": "ADDR0248",
+    "normalized_address": "036 din marg, baranagar, haryana"
+  },
+  {
+    "address_id": "ADDR0249",
+    "normalized_address": "61/721 chaudry ganj, raurkela industrial township, assam"
+  },
+  {
+    "address_id": "ADDR0250",
+    "normalized_address": "51/51 dora nagar, haldia, mizoram"
+  },
+  {
+    "address_id": "ADDR0251",
+    "normalized_address": "h.no. 92 doctor marg, baranagar, kerala"
+  },
+  {
+    "address_id": "ADDR0252",
+    "normalized_address": "h.no. 475 bose road, jalgaon, tripura"
+  },
+  {
+    "address_id": "ADDR0253",
+    "normalized_address": "25/267 bhavsar chowk, gangtok, maharashtra"
+  },
+  {
+    "address_id": "ADDR0254",
+    "normalized_address": "42/152 rajan street, imphal, meghalaya"
+  },
+  {
+    "address_id": "ADDR0255",
+    "normalized_address": "76 nath, sambalpur, madhya pradesh"
+  },
+  {
+    "address_id": "ADDR0256",
+    "normalized_address": "87 walla circle, katihar, nagaland"
+  },
+  {
+    "address_id": "ADDR0257",
+    "normalized_address": "619 vyas marg, chandigarh, haryana"
+  },
+  {
+    "address_id": "ADDR0258",
+    "normalized_address": "h.no. 03 patil nagar, guntur, bihar"
+  },
+  {
+    "address_id": "ADDR0259",
+    "normalized_address": "h.no. 552 sharma road, suryapet, maharashtra"
+  },
+  {
+    "address_id": "ADDR0260",
+    "normalized_address": "330 dalal street, burhanpur, telangana"
+  },
+  {
+    "address_id": "ADDR0261",
+    "normalized_address": "h.no. 650 borah zila, hindupur, west bengal"
+  },
+  {
+    "address_id": "ADDR0262",
+    "normalized_address": "025 brar street, solapur, andhra pradesh"
+  },
+  {
+    "address_id": "ADDR0263",
+    "normalized_address": "62/27 wable street, nandyal, maharashtra"
+  },
+  {
+    "address_id": "ADDR0264",
+    "normalized_address": "57/041 bhasin street, mahbubnagar, tripura"
+  },
+  {
+    "address_id": "ADDR0265",
+    "normalized_address": "820 sampath marg, begusarai, tamil nadu"
+  },
+  {
+    "address_id": "ADDR0266",
+    "normalized_address": "h.no. 98 karan road, sangli-miraj & kupwad, himachal pradesh"
+  },
+  {
+    "address_id": "ADDR0267",
+    "normalized_address": "h.no. 862 rattan ganj, muzaffarnagar, madhya pradesh"
+  },
+  {
+    "address_id": "ADDR0268",
+    "normalized_address": "891 sood chowk, machilipatnam, uttarakhand"
+  },
+  {
+    "address_id": "ADDR0269",
+    "normalized_address": "20 samra path, gopalpur, bihar"
+  },
+  {
+    "address_id": "ADDR0270",
+    "normalized_address": "h.no. 159 bala path, ichalkaranji, goa"
+  },
+  {
+    "address_id": "ADDR0271",
+    "normalized_address": "11/39 more nagar, suryapet, gujarat"
+  },
+  {
+    "address_id": "ADDR0272",
+    "normalized_address": "03/419 basu, ranchi, arunachal pradesh"
+  },
+  {
+    "address_id": "ADDR0273",
+    "normalized_address": "h.no. 14 parikh nagar, sikar, mizoram"
+  },
+  {
+    "address_id": "ADDR0274",
+    "normalized_address": "565 chatterjee street, vellore, telangana"
+  },
+  {
+    "address_id": "ADDR0275",
+    "normalized_address": "18/20 raja, dharmavaram, west bengal"
+  },
+  {
+    "address_id": "ADDR0276",
+    "normalized_address": "h.no. 422 brar chowk, ahmednagar, gujarat"
+  },
+  {
+    "address_id": "ADDR0277",
+    "normalized_address": "22/889 om circle, mira-bhayandar, karnataka"
+  },
+  {
+    "address_id": "ADDR0278",
+    "normalized_address": "h.no. 59 baral marg, mathura, gujarat"
+  },
+  {
+    "address_id": "ADDR0279",
+    "normalized_address": "91/29 chatterjee chowk, asansol, tripura"
+  },
+  {
+    "address_id": "ADDR0280",
+    "normalized_address": "h.no. 81 chandra chowk, satara, goa"
+  },
+  {
+    "address_id": "ADDR0281",
+    "normalized_address": "h.no. 21 patla road, hubli–dharwad, assam"
+  },
+  {
+    "address_id": "ADDR0282",
+    "normalized_address": "747 sathe road, amravati, odisha"
+  },
+  {
+    "address_id": "ADDR0283",
+    "normalized_address": "05 rattan road, udupi, sikkim"
+  },
+  {
+    "address_id": "ADDR0284",
+    "normalized_address": "758 murthy, gwalior, sikkim"
+  },
+  {
+    "address_id": "ADDR0285",
+    "normalized_address": "87 basak road, sambalpur, tamil nadu"
+  },
+  {
+    "address_id": "ADDR0286",
+    "normalized_address": "02/14 dutt marg, srinagar, haryana"
+  },
+  {
+    "address_id": "ADDR0287",
+    "normalized_address": "71 shukla path, deoghar, uttarakhand"
+  },
+  {
+    "address_id": "ADDR0288",
+    "normalized_address": "11 chokshi street, bhiwani, gujarat"
+  },
+  {
+    "address_id": "ADDR0289",
+    "normalized_address": "51 oommen path, rajpur sonarpur, tamil nadu"
+  },
+  {
+    "address_id": "ADDR0290",
+    "normalized_address": "71/470 dugal circle, ulhasnagar, rajasthan"
+  },
+  {
+    "address_id": "ADDR0291",
+    "normalized_address": "590 rai, mirzapur, uttar pradesh"
+  },
+  {
+    "address_id": "ADDR0292",
+    "normalized_address": "73 kakar nagar, bhusawal, himachal pradesh"
+  },
+  {
+    "address_id": "ADDR0293",
+    "normalized_address": "h.no. 21 biswas street, machilipatnam, assam"
+  },
+  {
+    "address_id": "ADDR0294",
+    "normalized_address": "h.no. 76 kala circle, udaipur, uttarakhand"
+  },
+  {
+    "address_id": "ADDR0295",
+    "normalized_address": "h.no. 165 gala road, karnal, maharashtra"
+  },
+  {
+    "address_id": "ADDR0296",
+    "normalized_address": "78 sarkar street, aurangabad, madhya pradesh"
+  },
+  {
+    "address_id": "ADDR0297",
+    "normalized_address": "h.no. 893 dhillon zila, ratlam, andhra pradesh"
+  },
+  {
+    "address_id": "ADDR0298",
+    "normalized_address": "727 comar path, firozabad, mizoram"
+  },
+  {
+    "address_id": "ADDR0299",
+    "normalized_address": "07/279 bansal zila, maheshtala, maharashtra"
+  },
+  {
+    "address_id": "ADDR0300",
+    "normalized_address": "82/963 kadakia path, gangtok, andhra pradesh"
+  },
+  {
+    "address_id": "ADDR0301",
+    "normalized_address": "h.no. 50 mohanty road, panchkula, rajasthan"
+  },
+  {
+    "address_id": "ADDR0302",
+    "normalized_address": "66 pandey path, nagaon, tripura"
+  },
+  {
+    "address_id": "ADDR0303",
+    "normalized_address": "h.no. 12 sangha chowk, belgaum, nagaland"
+  },
+  {
+    "address_id": "ADDR0304",
+    "normalized_address": "28/66 sarkar road, udupi, west bengal"
+  },
+  {
+    "address_id": "ADDR0305",
+    "normalized_address": "h.no. 111 sabharwal nagar, mehsana, telangana"
+  },
+  {
+    "address_id": "ADDR0306",
+    "normalized_address": "81/78 bali, bidar, gujarat"
+  },
+  {
+    "address_id": "ADDR0307",
+    "normalized_address": "786 hayre zila, amravati, kerala"
+  },
+  {
+    "address_id": "ADDR0308",
+    "normalized_address": "66 prasad nagar, guna, jharkhand"
+  },
+  {
+    "address_id": "ADDR0309",
+    "normalized_address": "h.no. 44 khurana marg, vijayanagaram, arunachal pradesh"
+  },
+  {
+    "address_id": "ADDR0310",
+    "normalized_address": "h.no. 025 nadkarni circle, tirunelveli, odisha"
+  },
+  {
+    "address_id": "ADDR0311",
+    "normalized_address": "38 hans marg, faridabad, uttar pradesh"
+  },
+  {
+    "address_id": "ADDR0312",
+    "normalized_address": "62/62 nadkarni marg, junagadh, haryana"
+  },
+  {
+    "address_id": "ADDR0313",
+    "normalized_address": "h.no. 281 ramanathan, begusarai, goa"
+  },
+  {
+    "address_id": "ADDR0314",
+    "normalized_address": "h.no. 31 bal zila, hyderabad, rajasthan"
+  },
+  {
+    "address_id": "ADDR0315",
+    "normalized_address": "20/98 raja zila, bikaner, manipur"
+  },
+  {
+    "address_id": "ADDR0316",
+    "normalized_address": "52/334 wason ganj, burhanpur, tripura"
+  },
+  {
+    "address_id": "ADDR0317",
+    "normalized_address": "h.no. 04 dhawan zila, adoni, west bengal"
+  },
+  {
+    "address_id": "ADDR0318",
+    "normalized_address": "h.no. 179 chhabra zila, ichalkaranji, himachal pradesh"
+  },
+  {
+    "address_id": "ADDR0319",
+    "normalized_address": "117 johal path, salem, meghalaya"
+  },
+  {
+    "address_id": "ADDR0320",
+    "normalized_address": "h.no. 617 kade zila, gandhinagar, karnataka"
+  },
+  {
+    "address_id": "ADDR0321",
+    "normalized_address": "72/240 balay street, tiruvottiyur, kerala"
+  },
+  {
+    "address_id": "ADDR0322",
+    "normalized_address": "99 ahluwalia circle, nagaon, telangana"
+  },
+  {
+    "address_id": "ADDR0323",
+    "normalized_address": "h.no. 13 balan, kadapa, mizoram"
+  },
+  {
+    "address_id": "ADDR0324",
+    "normalized_address": "351 balan ganj, durgapur, haryana"
+  },
+  {
+    "address_id": "ADDR0325",
+    "normalized_address": "h.no. 244 iyengar chowk, karnal, tamil nadu"
+  },
+  {
+    "address_id": "ADDR0326",
+    "normalized_address": "47/76 ratta circle, chandrapur, goa"
+  },
+  {
+    "address_id": "ADDR0327",
+    "normalized_address": "461 muni, tiruppur, telangana"
+  },
+  {
+    "address_id": "ADDR0328",
+    "normalized_address": "67 magar marg, moradabad, punjab"
+  },
+  {
+    "address_id": "ADDR0329",
+    "normalized_address": "63 pandit, rajkot, manipur"
+  },
+  {
+    "address_id": "ADDR0330",
+    "normalized_address": "084 rout, mahbubnagar, uttarakhand"
+  },
+  {
+    "address_id": "ADDR0331",
+    "normalized_address": "85/601 naidu street, gaya, kerala"
+  },
+  {
+    "address_id": "ADDR0332",
+    "normalized_address": "135 dhillon path, bhilai, andhra pradesh"
+  },
+  {
+    "address_id": "ADDR0333",
+    "normalized_address": "h.no. 05 shroff ganj, muzaffarpur, nagaland"
+  },
+  {
+    "address_id": "ADDR0334",
+    "normalized_address": "h.no. 649 vyas ganj, kota, nagaland"
+  },
+  {
+    "address_id": "ADDR0335",
+    "normalized_address": "h.no. 91 sama, ratlam, uttarakhand"
+  },
+  {
+    "address_id": "ADDR0336",
+    "normalized_address": "h.no. 33 tak nagar, ranchi, himachal pradesh"
+  },
+  {
+    "address_id": "ADDR0337",
+    "normalized_address": "11 garg chowk, gudivada, madhya pradesh"
+  },
+  {
+    "address_id": "ADDR0338",
+    "normalized_address": "904 agrawal nagar, anantapuram, chhattisgarh"
+  },
+  {
+    "address_id": "ADDR0339",
+    "normalized_address": "08 baral circle, pune, assam"
+  },
+  {
+    "address_id": "ADDR0340",
+    "normalized_address": "16 maharaj road, sri ganganagar, madhya pradesh"
+  },
+  {
+    "address_id": "ADDR0341",
+    "normalized_address": "h.no. 857 batta, bihar sharif, tamil nadu"
+  },
+  {
+    "address_id": "ADDR0342",
+    "normalized_address": "71 chauhan path, rajahmundry, arunachal pradesh"
+  },
+  {
+    "address_id": "ADDR0343",
+    "normalized_address": "h.no. 215 kara marg, hubli–dharwad, gujarat"
+  },
+  {
+    "address_id": "ADDR0344",
+    "normalized_address": "58/004 choudhry nagar, avadi, assam"
+  },
+  {
+    "address_id": "ADDR0345",
+    "normalized_address": "h.no. 59 madan marg, durgapur, odisha"
+  },
+  {
+    "address_id": "ADDR0346",
+    "normalized_address": "12 divan path, amroha, sikkim"
+  },
+  {
+    "address_id": "ADDR0347",
+    "normalized_address": "h.no. 883 sridhar, parbhani, karnataka"
+  },
+  {
+    "address_id": "ADDR0348",
+    "normalized_address": "73/144 pandit zila, tiruppur, tamil nadu"
+  },
+  {
+    "address_id": "ADDR0349",
+    "normalized_address": "h.no. 170 arya ganj, bhind, uttarakhand"
+  },
+  {
+    "address_id": "ADDR0350",
+    "normalized_address": "37 oak zila, gudivada, tripura"
+  },
+  {
+    "address_id": "ADDR0351",
+    "normalized_address": "12 fraud lane, mumbai, maharashtra"
+  },
+  {
+    "address_id": "ADDR0352",
+    "normalized_address": "45 ring road, delhi, delhi"
+  },
+  {
+    "address_id": "ADDR0353",
+    "normalized_address": "78 scheme nagar, bengaluru, karnataka"
+  },
+  {
+    "address_id": "ADDR0354",
+    "normalized_address": "23 boundary blvd, pune, maharashtra"
+  },
+  {
+    "address_id": "ADDR0355",
+    "normalized_address": "56 ambiguous avenue, hyderabad, telangana"
+  },
+  {
+    "address_id": "ADDR0356",
+    "normalized_address": "89 liminal street, chennai, tamil_nadu"
+  },
+  {
+    "address_id": "ADDR0357",
+    "normalized_address": "33 ordinary street, jaipur, rajasthan"
+  }
+]

data/synthetic/candidate_clusters.json

@@ -0,0 +1,784 @@
+[
+  {
+    "cluster_id": "CAND01",
+    "customer_ids": [
+      "CUST0041",
+      "CUST0042",
+      "CUST0043",
+      "CUST0044"
+    ],
+    "size": 4,
+    "strong_link_count": 6,
+    "weak_link_count": 0,
+    "formation_type": "strong_anchored",
+    "method": "connected_components"
+  },
+  {
+    "cluster_id": "CAND02",
+    "customer_ids": [
+      "CUST0009",
+      "CUST0010",
+      "CUST0011",
+      "CUST0012",
+      "CUST0013"
+    ],
+    "size": 5,
+    "strong_link_count": 10,
+    "weak_link_count": 0,
+    "formation_type": "strong_anchored",
+    "method": "connected_components"
+  },
+  {
+    "cluster_id": "CAND03",
+    "customer_ids": [
+      "CUST0001",
+      "CUST0002",
+      "CUST0003"
+    ],
+    "size": 3,
+    "strong_link_count": 3,
+    "weak_link_count": 0,
+    "formation_type": "strong_anchored",
+    "method": "connected_components"
+  },
+  {
+    "cluster_id": "CAND04",
+    "customer_ids": [
+      "CUST0014",
+      "CUST0015",
+      "CUST0016",
+      "CUST0017",
+      "CUST0018"
+    ],
+    "size": 5,
+    "strong_link_count": 10,
+    "weak_link_count": 0,
+    "formation_type": "strong_anchored",
+    "method": "connected_components"
+  },
+  {
+    "cluster_id": "CAND05",
+    "customer_ids": [
+      "CUST0033",
+      "CUST0034",
+      "CUST0035"
+    ],
+    "size": 3,
+    "strong_link_count": 3,
+    "weak_link_count": 0,
+    "formation_type": "strong_anchored",
+    "method": "connected_components"
+  },
+  {
+    "cluster_id": "CAND06",
+    "customer_ids": [
+      "CUST0036",
+      "CUST0037",
+      "CUST0038",
+      "CUST0039",
+      "CUST0040"
+    ],
+    "size": 5,
+    "strong_link_count": 10,
+    "weak_link_count": 0,
+    "formation_type": "strong_anchored",
+    "method": "connected_components"
+  },
+  {
+    "cluster_id": "CAND07",
+    "customer_ids": [
+      "CUST0019",
+      "CUST0020",
+      "CUST0021",
+      "CUST0022",
+      "CUST0023"
+    ],
+    "size": 5,
+    "strong_link_count": 10,
+    "weak_link_count": 0,
+    "formation_type": "strong_anchored",
+    "method": "connected_components"
+  },
+  {
+    "cluster_id": "CAND08",
+    "customer_ids": [
+      "CUST0501",
+      "CUST0502",
+      "CUST0503"
+    ],
+    "size": 3,
+    "strong_link_count": 3,
+    "weak_link_count": 0,
+    "formation_type": "strong_anchored",
+    "method": "connected_components"
+  },
+  {
+    "cluster_id": "CAND09",
+    "customer_ids": [
+      "CUST0024",
+      "CUST0025",
+      "CUST0026",
+      "CUST0027",
+      "CUST0028",
+      "CUST0029"
+    ],
+    "size": 6,
+    "strong_link_count": 15,
+    "weak_link_count": 0,
+    "formation_type": "strong_anchored",
+    "method": "connected_components"
+  },
+  {
+    "cluster_id": "CAND10",
+    "customer_ids": [
+      "CUST0004",
+      "CUST0005",
+      "CUST0006",
+      "CUST0007",
+      "CUST0008"
+    ],
+    "size": 5,
+    "strong_link_count": 7,
+    "weak_link_count": 0,
+    "formation_type": "strong_anchored",
+    "method": "connected_components"
+  },
+  {
+    "cluster_id": "CAND11",
+    "customer_ids": [
+      "CUST0030",
+      "CUST0031",
+      "CUST0032"
+    ],
+    "size": 3,
+    "strong_link_count": 2,
+    "weak_link_count": 0,
+    "formation_type": "strong_anchored",
+    "method": "connected_components"
+  },
+  {
+    "cluster_id": "CAND12",
+    "customer_ids": [
+      "CUST0001",
+      "CUST0019",
+      "CUST0040",
+      "CUST0126",
+      "CUST0138",
+      "CUST0168",
+      "CUST0211",
+      "CUST0219",
+      "CUST0258",
+      "CUST0281",
+      "CUST0292",
+      "CUST0322",
+      "CUST0325",
+      "CUST0371",
+      "CUST0373",
+      "CUST0400",
+      "CUST0419",
+      "CUST0465"
+    ],
+    "size": 18,
+    "strong_link_count": 0,
+    "weak_link_count": 41,
+    "formation_type": "weak_only",
+    "method": "louvain"
+  },
+  {
+    "cluster_id": "CAND13",
+    "customer_ids": [
+      "CUST0002",
+      "CUST0009",
+      "CUST0020",
+      "CUST0041",
+      "CUST0064",
+      "CUST0073",
+      "CUST0074",
+      "CUST0080",
+      "CUST0081",
+      "CUST0086",
+      "CUST0089",
+      "CUST0103",
+      "CUST0162",
+      "CUST0167",
+      "CUST0173",
+      "CUST0181",
+      "CUST0226",
+      "CUST0240",
+      "CUST0253",
+      "CUST0262",
+      "CUST0272",
+      "CUST0286",
+      "CUST0287",
+      "CUST0324",
+      "CUST0331",
+      "CUST0333",
+      "CUST0369",
+      "CUST0392",
+      "CUST0393",
+      "CUST0398",
+      "CUST0406",
+      "CUST0410",
+      "CUST0411",
+      "CUST0413",
+      "CUST0415",
+      "CUST0423",
+      "CUST0446",
+      "CUST0468",
+      "CUST0476",
+      "CUST0478",
+      "CUST0485"
+    ],
+    "size": 41,
+    "strong_link_count": 0,
+    "weak_link_count": 146,
+    "formation_type": "weak_only",
+    "method": "louvain"
+  },
+  {
+    "cluster_id": "CAND14",
+    "customer_ids": [
+      "CUST0003",
+      "CUST0012",
+      "CUST0043",
+      "CUST0077",
+      "CUST0088",
+      "CUST0098",
+      "CUST0109",
+      "CUST0125",
+      "CUST0139",
+      "CUST0153",
+      "CUST0159",
+      "CUST0166",
+      "CUST0180",
+      "CUST0201",
+      "CUST0209",
+      "CUST0245",
+      "CUST0261",
+      "CUST0265",
+      "CUST0273",
+      "CUST0277",
+      "CUST0297",
+      "CUST0315",
+      "CUST0334",
+      "CUST0347",
+      "CUST0381",
+      "CUST0405",
+      "CUST0412",
+      "CUST0428",
+      "CUST0429",
+      "CUST0462",
+      "CUST0471"
+    ],
+    "size": 31,
+    "strong_link_count": 0,
+    "weak_link_count": 84,
+    "formation_type": "weak_only",
+    "method": "louvain"
+  },
+  {
+    "cluster_id": "CAND15",
+    "customer_ids": [
+      "CUST0004",
+      "CUST0008",
+      "CUST0027",
+      "CUST0030",
+      "CUST0042",
+      "CUST0063",
+      "CUST0075",
+      "CUST0087",
+      "CUST0110",
+      "CUST0117",
+      "CUST0120",
+      "CUST0133",
+      "CUST0135",
+      "CUST0137",
+      "CUST0144",
+      "CUST0152",
+      "CUST0154",
+      "CUST0163",
+      "CUST0170",
+      "CUST0178",
+      "CUST0179",
+      "CUST0195",
+      "CUST0198",
+      "CUST0208",
+      "CUST0210",
+      "CUST0221",
+      "CUST0228",
+      "CUST0231",
+      "CUST0237",
+      "CUST0246",
+      "CUST0251",
+      "CUST0254",
+      "CUST0266",
+      "CUST0282",
+      "CUST0295",
+      "CUST0300",
+      "CUST0303",
+      "CUST0304",
+      "CUST0335",
+      "CUST0355",
+      "CUST0365",
+      "CUST0399",
+      "CUST0420",
+      "CUST0425",
+      "CUST0470",
+      "CUST0483",
+      "CUST0486",
+      "CUST0487",
+      "CUST0491",
+      "CUST0492",
+      "CUST0498"
+    ],
+    "size": 51,
+    "strong_link_count": 1,
+    "weak_link_count": 167,
+    "formation_type": "weak_only",
+    "method": "louvain"
+  },
+  {
+    "cluster_id": "CAND16",
+    "customer_ids": [
+      "CUST0005",
+      "CUST0022",
+      "CUST0028",
+      "CUST0038",
+      "CUST0102",
+      "CUST0111",
+      "CUST0113",
+      "CUST0140",
+      "CUST0161",
+      "CUST0171",
+      "CUST0177",
+      "CUST0202",
+      "CUST0222",
+      "CUST0235",
+      "CUST0252",
+      "CUST0257",
+      "CUST0263",
+      "CUST0288",
+      "CUST0293",
+      "CUST0299",
+      "CUST0308",
+      "CUST0311",
+      "CUST0353",
+      "CUST0357",
+      "CUST0364",
+      "CUST0370",
+      "CUST0383",
+      "CUST0386",
+      "CUST0387",
+      "CUST0391",
+      "CUST0396",
+      "CUST0438",
+      "CUST0442",
+      "CUST0456",
+      "CUST0460",
+      "CUST0488",
+      "CUST0490",
+      "CUST0495",
+      "CUST0496",
+      "CUST0500"
+    ],
+    "size": 40,
+    "strong_link_count": 0,
+    "weak_link_count": 112,
+    "formation_type": "weak_only",
+    "method": "louvain"
+  },
+  {
+    "cluster_id": "CAND17",
+    "customer_ids": [
+      "CUST0006",
+      "CUST0011",
+      "CUST0024",
+      "CUST0025",
+      "CUST0033",
+      "CUST0044",
+      "CUST0079",
+      "CUST0084",
+      "CUST0101",
+      "CUST0107",
+      "CUST0108",
+      "CUST0114",
+      "CUST0132",
+      "CUST0145",
+      "CUST0157",
+      "CUST0172",
+      "CUST0227",
+      "CUST0244",
+      "CUST0255",
+      "CUST0256",
+      "CUST0267",
+      "CUST0271",
+      "CUST0284",
+      "CUST0291",
+      "CUST0296",
+      "CUST0316",
+      "CUST0319",
+      "CUST0329",
+      "CUST0336",
+      "CUST0342",
+      "CUST0348",
+      "CUST0349",
+      "CUST0367",
+      "CUST0375",
+      "CUST0389",
+      "CUST0407",
+      "CUST0455",
+      "CUST0489",
+      "CUST0499"
+    ],
+    "size": 39,
+    "strong_link_count": 1,
+    "weak_link_count": 95,
+    "formation_type": "weak_only",
+    "method": "louvain"
+  },
+  {
+    "cluster_id": "CAND18",
+    "customer_ids": [
+      "CUST0007",
+      "CUST0014",
+      "CUST0015",
+      "CUST0016",
+      "CUST0032",
+      "CUST0037",
+      "CUST0065",
+      "CUST0069",
+      "CUST0070",
+      "CUST0095",
+      "CUST0106",
+      "CUST0115",
+      "CUST0124",
+      "CUST0131",
+      "CUST0149",
+      "CUST0191",
+      "CUST0200",
+      "CUST0204",
+      "CUST0215",
+      "CUST0218",
+      "CUST0223",
+      "CUST0238",
+      "CUST0239",
+      "CUST0241",
+      "CUST0242",
+      "CUST0264",
+      "CUST0276",
+      "CUST0280",
+      "CUST0305",
+      "CUST0314",
+      "CUST0323",
+      "CUST0327",
+      "CUST0339",
+      "CUST0350",
+      "CUST0352",
+      "CUST0356",
+      "CUST0358",
+      "CUST0362",
+      "CUST0379",
+      "CUST0401",
+      "CUST0418",
+      "CUST0440",
+      "CUST0443",
+      "CUST0449",
+      "CUST0479"
+    ],
+    "size": 45,
+    "strong_link_count": 3,
+    "weak_link_count": 106,
+    "formation_type": "weak_only",
+    "method": "louvain"
+  },
+  {
+    "cluster_id": "CAND19",
+    "customer_ids": [
+      "CUST0010",
+      "CUST0023",
+      "CUST0029",
+      "CUST0090",
+      "CUST0119",
+      "CUST0121",
+      "CUST0129",
+      "CUST0141",
+      "CUST0143",
+      "CUST0155",
+      "CUST0196",
+      "CUST0207",
+      "CUST0230",
+      "CUST0233",
+      "CUST0247",
+      "CUST0330",
+      "CUST0338",
+      "CUST0372",
+      "CUST0374",
+      "CUST0380",
+      "CUST0461",
+      "CUST0467",
+      "CUST0480",
+      "CUST0493"
+    ],
+    "size": 24,
+    "strong_link_count": 0,
+    "weak_link_count": 47,
+    "formation_type": "weak_only",
+    "method": "louvain"
+  },
+  {
+    "cluster_id": "CAND20",
+    "customer_ids": [
+      "CUST0013",
+      "CUST0026",
+      "CUST0031",
+      "CUST0035",
+      "CUST0039",
+      "CUST0067",
+      "CUST0068",
+      "CUST0094",
+      "CUST0116",
+      "CUST0122",
+      "CUST0128",
+      "CUST0184",
+      "CUST0193",
+      "CUST0203",
+      "CUST0212",
+      "CUST0217",
+      "CUST0248",
+      "CUST0268",
+      "CUST0283",
+      "CUST0312",
+      "CUST0340",
+      "CUST0351",
+      "CUST0384",
+      "CUST0402",
+      "CUST0417",
+      "CUST0427",
+      "CUST0430",
+      "CUST0434",
+      "CUST0451",
+      "CUST0463",
+      "CUST0466",
+      "CUST0469",
+      "CUST0473",
+      "CUST0482",
+      "CUST0497"
+    ],
+    "size": 35,
+    "strong_link_count": 0,
+    "weak_link_count": 84,
+    "formation_type": "weak_only",
+    "method": "louvain"
+  },
+  {
+    "cluster_id": "CAND21",
+    "customer_ids": [
+      "CUST0017",
+      "CUST0036",
+      "CUST0062",
+      "CUST0071",
+      "CUST0082",
+      "CUST0083",
+      "CUST0085",
+      "CUST0096",
+      "CUST0123",
+      "CUST0142",
+      "CUST0147",
+      "CUST0175",
+      "CUST0189",
+      "CUST0216",
+      "CUST0220",
+      "CUST0224",
+      "CUST0232",
+      "CUST0234",
+      "CUST0306",
+      "CUST0328",
+      "CUST0337",
+      "CUST0341",
+      "CUST0345",
+      "CUST0346",
+      "CUST0361",
+      "CUST0388",
+      "CUST0421",
+      "CUST0426",
+      "CUST0433",
+      "CUST0435",
+      "CUST0444",
+      "CUST0450",
+      "CUST0454",
+      "CUST0477"
+    ],
+    "size": 34,
+    "strong_link_count": 0,
+    "weak_link_count": 79,
+    "formation_type": "weak_only",
+    "method": "louvain"
+  },
+  {
+    "cluster_id": "CAND22",
+    "customer_ids": [
+      "CUST0018",
+      "CUST0021",
+      "CUST0091",
+      "CUST0092",
+      "CUST0112",
+      "CUST0160",
+      "CUST0174",
+      "CUST0187",
+      "CUST0270",
+      "CUST0298",
+      "CUST0307",
+      "CUST0310",
+      "CUST0332",
+      "CUST0343",
+      "CUST0431",
+      "CUST0474",
+      "CUST0475",
+      "CUST0494"
+    ],
+    "size": 18,
+    "strong_link_count": 0,
+    "weak_link_count": 35,
+    "formation_type": "weak_only",
+    "method": "louvain"
+  },
+  {
+    "cluster_id": "CAND23",
+    "customer_ids": [
+      "CUST0034",
+      "CUST0066",
+      "CUST0078",
+      "CUST0100",
+      "CUST0104",
+      "CUST0134",
+      "CUST0146",
+      "CUST0150",
+      "CUST0151",
+      "CUST0158",
+      "CUST0182",
+      "CUST0183",
+      "CUST0185",
+      "CUST0188",
+      "CUST0190",
+      "CUST0259",
+      "CUST0269",
+      "CUST0279",
+      "CUST0285",
+      "CUST0289",
+      "CUST0301",
+      "CUST0302",
+      "CUST0344",
+      "CUST0354",
+      "CUST0366",
+      "CUST0368",
+      "CUST0378",
+      "CUST0404",
+      "CUST0408",
+      "CUST0432",
+      "CUST0436",
+      "CUST0437",
+      "CUST0439",
+      "CUST0441",
+      "CUST0447",
+      "CUST0448",
+      "CUST0457",
+      "CUST0481"
+    ],
+    "size": 38,
+    "strong_link_count": 0,
+    "weak_link_count": 90,
+    "formation_type": "weak_only",
+    "method": "louvain"
+  },
+  {
+    "cluster_id": "CAND24",
+    "customer_ids": [
+      "CUST0061",
+      "CUST0093",
+      "CUST0105",
+      "CUST0130",
+      "CUST0136",
+      "CUST0148",
+      "CUST0164",
+      "CUST0186",
+      "CUST0197",
+      "CUST0205",
+      "CUST0206",
+      "CUST0214",
+      "CUST0260",
+      "CUST0309",
+      "CUST0320",
+      "CUST0321",
+      "CUST0359",
+      "CUST0363",
+      "CUST0376",
+      "CUST0377",
+      "CUST0385",
+      "CUST0394",
+      "CUST0403",
+      "CUST0416",
+      "CUST0424",
+      "CUST0453"
+    ],
+    "size": 26,
+    "strong_link_count": 0,
+    "weak_link_count": 71,
+    "formation_type": "weak_only",
+    "method": "louvain"
+  },
+  {
+    "cluster_id": "CAND25",
+    "customer_ids": [
+      "CUST0045",
+      "CUST0046",
+      "CUST0047",
+      "CUST0048"
+    ],
+    "size": 4,
+    "strong_link_count": 0,
+    "weak_link_count": 6,
+    "formation_type": "weak_only",
+    "method": "connected_components"
+  },
+  {
+    "cluster_id": "CAND26",
+    "customer_ids": [
+      "CUST0049",
+      "CUST0050",
+      "CUST0051",
+      "CUST0052"
+    ],
+    "size": 4,
+    "strong_link_count": 0,
+    "weak_link_count": 6,
+    "formation_type": "weak_only",
+    "method": "connected_components"
+  },
+  {
+    "cluster_id": "CAND27",
+    "customer_ids": [
+      "CUST0053",
+      "CUST0054",
+      "CUST0055",
+      "CUST0056"
+    ],
+    "size": 4,
+    "strong_link_count": 0,
+    "weak_link_count": 6,
+    "formation_type": "weak_only",
+    "method": "connected_components"
+  },
+  {
+    "cluster_id": "CAND28",
+    "customer_ids": [
+      "CUST0504",
+      "CUST0505",
+      "CUST0506"
+    ],
+    "size": 3,
+    "strong_link_count": 0,
+    "weak_link_count": 3,
+    "formation_type": "weak_only",
+    "method": "connected_components"
+  }
+]

data/synthetic/cases_demo.json

@@ -0,0 +1,1027 @@
+[
+  {
+    "case_id": "CASE01",
+    "cluster_id": "CAND01",
+    "customer_ids": [
+      "CUST0041",
+      "CUST0042",
+      "CUST0043",
+      "CUST0044"
+    ],
+    "risk_probability": 0.9958773824170732,
+    "policy_decision": "escalate",
+    "llm_narrative": "The four customers (CUST0041, CUST0042, CUST0043, CUST0044) form a strong‑anchored cluster with six strong links and no weak links, indicating tightly coordinated activity. Their combined purchase total of $58,186 is accompanied by a 33.3% return rate and a 16.7% chargeback rate, further supporting fraudulent behavior.",
+    "llm_confidence_statement": "The model assigns a risk probability of 99.6%, indicating very high confidence in the escalation decision.",
+    "action_taken": {
+      "tool": "razorpay.Client.order.create + order.edit + order.fetch",
+      "order_id": "order_TSeidafHmbAnDR",
+      "razorpay_status": "created",
+      "notes_written": {
+        "case_id": "CASE01",
+        "cluster_id": "CAND01",
+        "ringguard_status": "escalate_flagged",
+        "action_label": "FRAUD_RING_ESCALATE",
+        "risk_score": "0.9959",
+        "action_by": "RingGuard_v3",
+        "action_timestamp": "2026-08-22T02:47:10.102605+00:00"
+      },
+      "notes_verified": true,
+      "timestamp": "2026-08-22T02:47:10.941027+00:00",
+      "test_mode": true,
+      "error": null
+    },
+    "timestamp": "2026-08-22T02:42:02.088359+00:00"
+  },
+  {
+    "case_id": "CASE02",
+    "cluster_id": "CAND02",
+    "customer_ids": [
+      "CUST0009",
+      "CUST0010",
+      "CUST0011",
+      "CUST0012",
+      "CUST0013"
+    ],
+    "risk_probability": 0.9968053865328341,
+    "policy_decision": "escalate",
+    "llm_narrative": "The five customers (CUST0009‑CUST0013) form a strong‑anchored cluster with ten strong links and no weak links. The group shows a return rate of 47% and a chargeback rate of 18%, with total transaction volume of $59,830. The risk model assigns a probability of 99.7%, resulting in an escalation decision.",
+    "llm_confidence_statement": "We are highly confident in this assessment given the very high risk probability and supporting metrics.",
+    "action_taken": {
+      "tool": "razorpay.Client.order.create + order.edit + order.fetch",
+      "order_id": "order_TSeieevTUEtDNP",
+      "razorpay_status": "created",
+      "notes_written": {
+        "case_id": "CASE02",
+        "cluster_id": "CAND02",
+        "ringguard_status": "escalate_flagged",
+        "action_label": "FRAUD_RING_ESCALATE",
+        "risk_score": "0.9968",
+        "action_by": "RingGuard_v3",
+        "action_timestamp": "2026-08-22T02:47:11.088678+00:00"
+      },
+      "notes_verified": true,
+      "timestamp": "2026-08-22T02:47:11.992374+00:00",
+      "test_mode": true,
+      "error": null
+    },
+    "timestamp": "2026-08-22T02:42:06.481196+00:00"
+  },
+  {
+    "case_id": "CASE03",
+    "cluster_id": "CAND03",
+    "customer_ids": [
+      "CUST0001",
+      "CUST0002",
+      "CUST0003"
+    ],
+    "risk_probability": 0.9950028638127228,
+    "policy_decision": "escalate",
+    "llm_narrative": "The three customers (CUST0001, CUST0002, CUST0003) are connected by three strong links, forming a strong‑anchored cluster. This cluster accounts for a total transaction amount of 32,897, includes one shared coupon, and exhibits a 25% return rate and a 12.5% chargeback rate, resulting in a risk probability of 99.5%.",
+    "llm_confidence_statement": "The model’s confidence in this risk assessment is extremely high.",
+    "action_taken": {
+      "tool": "razorpay.Client.order.create + order.edit + order.fetch",
+      "order_id": "order_TSeifof2nHqiyN",
+      "razorpay_status": "created",
+      "notes_written": {
+        "case_id": "CASE03",
+        "cluster_id": "CAND03",
+        "ringguard_status": "escalate_flagged",
+        "action_label": "FRAUD_RING_ESCALATE",
+        "risk_score": "0.995",
+        "action_by": "RingGuard_v3",
+        "action_timestamp": "2026-08-22T02:47:12.186378+00:00"
+      },
+      "notes_verified": true,
+      "timestamp": "2026-08-22T02:47:13.232792+00:00",
+      "test_mode": true,
+      "error": null
+    },
+    "timestamp": "2026-08-22T02:42:10.844934+00:00"
+  },
+  {
+    "case_id": "CASE04",
+    "cluster_id": "CAND04",
+    "customer_ids": [
+      "CUST0014",
+      "CUST0015",
+      "CUST0016",
+      "CUST0017",
+      "CUST0018"
+    ],
+    "risk_probability": 0.9955003588432927,
+    "policy_decision": "escalate",
+    "llm_narrative": "The five customers (CUST0014‑CUST0018) form a strong‑anchored cluster with ten strong links and no weak links, collectively spending $62,087. The group exhibits a 50% return rate, a 12.5% chargeback rate, and shares a single coupon, indicating coordinated risky behavior.",
+    "llm_confidence_statement": "Given the 0.9955 risk probability, we are highly confident that this cluster represents a high‑risk fraud pattern.",
+    "action_taken": {
+      "tool": "razorpay.Client.order.create + order.edit + order.fetch",
+      "order_id": "order_TSeihAqIgYfEsF",
+      "razorpay_status": "created",
+      "notes_written": {
+        "case_id": "CASE04",
+        "cluster_id": "CAND04",
+        "ringguard_status": "escalate_flagged",
+        "action_label": "FRAUD_RING_ESCALATE",
+        "risk_score": "0.9955",
+        "action_by": "RingGuard_v3",
+        "action_timestamp": "2026-08-22T02:47:13.364816+00:00"
+      },
+      "notes_verified": true,
+      "timestamp": "2026-08-22T02:47:14.280655+00:00",
+      "test_mode": true,
+      "error": null
+    },
+    "timestamp": "2026-08-22T02:42:15.156355+00:00"
+  },
+  {
+    "case_id": "CASE05",
+    "cluster_id": "CAND05",
+    "customer_ids": [
+      "CUST0033",
+      "CUST0034",
+      "CUST0035"
+    ],
+    "risk_probability": 0.9969893801631614,
+    "policy_decision": "escalate",
+    "llm_narrative": "The three customers (CUST0033, CUST0034, CUST0035) are connected by three strong links, forming a strong‑anchored cluster with a shared coupon and a combined transaction volume of $50,134. The cluster exhibits a return rate of 18.2% and a chargeback rate of 9.1%, and its calculated risk probability is 99.7%.",
+    "llm_confidence_statement": "We are highly confident in the fraud assessment given the strong linkage, high risk probability, and notable financial metrics.",
+    "action_taken": {
+      "tool": "razorpay.Client.order.create + order.edit + order.fetch",
+      "order_id": "order_TSeiiK16ZPVegj",
+      "razorpay_status": "created",
+      "notes_written": {
+        "case_id": "CASE05",
+        "cluster_id": "CAND05",
+        "ringguard_status": "escalate_flagged",
+        "action_label": "FRAUD_RING_ESCALATE",
+        "risk_score": "0.997",
+        "action_by": "RingGuard_v3",
+        "action_timestamp": "2026-08-22T02:47:14.523378+00:00"
+      },
+      "notes_verified": true,
+      "timestamp": "2026-08-22T02:47:15.299833+00:00",
+      "test_mode": true,
+      "error": null
+    },
+    "timestamp": "2026-08-22T02:42:19.568153+00:00"
+  },
+  {
+    "case_id": "CASE06",
+    "cluster_id": "CAND06",
+    "customer_ids": [
+      "CUST0036",
+      "CUST0037",
+      "CUST0038",
+      "CUST0039",
+      "CUST0040"
+    ],
+    "risk_probability": 0.9969377389203431,
+    "policy_decision": "escalate",
+    "llm_narrative": "The five customers (CUST0036‑CUST0040) form a strong‑anchored cluster with ten strong links and no weak links, collectively transacting $62,270. Their return rate of 26.7% and chargeback rate of 20% are elevated, and the absence of shared coupons suggests the links are not promotional. These characteristics indicate coordinated fraudulent behavior.",
+    "llm_confidence_statement": "We are highly confident in this assessment, reflected by a risk probability of 0.997.",
+    "action_taken": {
+      "tool": "razorpay.Client.order.create + order.edit + order.fetch",
+      "order_id": "order_TSeijQj96lsR88",
+      "razorpay_status": "created",
+      "notes_written": {
+        "case_id": "CASE06",
+        "cluster_id": "CAND06",
+        "ringguard_status": "escalate_flagged",
+        "action_label": "FRAUD_RING_ESCALATE",
+        "risk_score": "0.9969",
+        "action_by": "RingGuard_v3",
+        "action_timestamp": "2026-08-22T02:47:15.427579+00:00"
+      },
+      "notes_verified": true,
+      "timestamp": "2026-08-22T02:47:16.209387+00:00",
+      "test_mode": true,
+      "error": null
+    },
+    "timestamp": "2026-08-22T02:42:23.867363+00:00"
+  },
+  {
+    "case_id": "CASE07",
+    "cluster_id": "CAND07",
+    "customer_ids": [
+      "CUST0019",
+      "CUST0020",
+      "CUST0021",
+      "CUST0022",
+      "CUST0023"
+    ],
+    "risk_probability": 0.9940850384979173,
+    "policy_decision": "escalate",
+    "llm_narrative": "A cluster of five customers (CUST0019‑CUST0023) is connected by ten strong links with no weak links, forming a strong‑anchored network. The group has generated a total transaction amount of 70,841, a return rate of 33.33% and a chargeback rate of 5.56%, and the model assigns a risk probability of 99.41%.",
+    "llm_confidence_statement": "The assessment is made with very high confidence.",
+    "action_taken": {
+      "tool": "razorpay.Client.order.create + order.edit + order.fetch",
+      "order_id": "order_TSeikSedcIcNGX",
+      "razorpay_status": "created",
+      "notes_written": {
+        "case_id": "CASE07",
+        "cluster_id": "CAND07",
+        "ringguard_status": "escalate_flagged",
+        "action_label": "FRAUD_RING_ESCALATE",
+        "risk_score": "0.9941",
+        "action_by": "RingGuard_v3",
+        "action_timestamp": "2026-08-22T02:47:16.396113+00:00"
+      },
+      "notes_verified": true,
+      "timestamp": "2026-08-22T02:47:17.279431+00:00",
+      "test_mode": true,
+      "error": null
+    },
+    "timestamp": "2026-08-22T02:42:28.164967+00:00"
+  },
+  {
+    "case_id": "CASE08",
+    "cluster_id": "CAND08",
+    "customer_ids": [
+      "CUST0501",
+      "CUST0502",
+      "CUST0503"
+    ],
+    "risk_probability": 0.9867466290228035,
+    "policy_decision": "auto_hold",
+    "llm_narrative": "The three customers (CUST0501, CUST0502, CUST0503) are connected by three strong links, forming a strong‑anchored cluster. They share one coupon, together generated $19,800 in transactions, and exhibit a 16.7% return and chargeback rate each.",
+    "llm_confidence_statement": "The model assigns a risk probability of 0.99, indicating very high confidence in the high‑risk classification.",
+    "action_taken": {
+      "tool": "razorpay.Client.order.create + order.edit + order.fetch",
+      "order_id": "order_TSeilZzpguMme6",
+      "razorpay_status": "created",
+      "notes_written": {
+        "case_id": "CASE08",
+        "cluster_id": "CAND08",
+        "ringguard_status": "auto_hold",
+        "action_label": "FRAUD_RING_AUTO_HOLD",
+        "risk_score": "0.9867",
+        "action_by": "RingGuard_v3",
+        "action_timestamp": "2026-08-22T02:47:17.462450+00:00"
+      },
+      "notes_verified": true,
+      "timestamp": "2026-08-22T02:47:18.321891+00:00",
+      "test_mode": true,
+      "error": null
+    },
+    "timestamp": "2026-08-22T02:42:32.454693+00:00"
+  },
+  {
+    "case_id": "CASE09",
+    "cluster_id": "CAND09",
+    "customer_ids": [
+      "CUST0024",
+      "CUST0025",
+      "CUST0026",
+      "CUST0027",
+      "CUST0028",
+      "CUST0029"
+    ],
+    "risk_probability": 0.9913203009684713,
+    "policy_decision": "escalate",
+    "llm_narrative": "The six customers (CUST0024‑CUST0029) form a strong_anchored cluster with 15 strong links and no weak links, indicating tightly coordinated activity. Their combined purchase total of 81,019 and a return rate of 55% further highlight abnormal behavior, while a chargeback rate of 5% adds to the risk profile.",
+    "llm_confidence_statement": "The evidence yields a very high confidence in the elevated risk assessment.",
+    "action_taken": {
+      "tool": "razorpay.Client.order.create + order.edit + order.fetch",
+      "order_id": "order_TSeimiDR56y6I3",
+      "razorpay_status": "created",
+      "notes_written": {
+        "case_id": "CASE09",
+        "cluster_id": "CAND09",
+        "ringguard_status": "escalate_flagged",
+        "action_label": "FRAUD_RING_ESCALATE",
+        "risk_score": "0.9913",
+        "action_by": "RingGuard_v3",
+        "action_timestamp": "2026-08-22T02:47:18.435548+00:00"
+      },
+      "notes_verified": true,
+      "timestamp": "2026-08-22T02:47:19.433491+00:00",
+      "test_mode": true,
+      "error": null
+    },
+    "timestamp": "2026-08-22T02:42:36.567254+00:00"
+  },
+  {
+    "case_id": "CASE10",
+    "cluster_id": "CAND10",
+    "customer_ids": [
+      "CUST0004",
+      "CUST0005",
+      "CUST0006",
+      "CUST0007",
+      "CUST0008"
+    ],
+    "risk_probability": 0.9959231873282086,
+    "policy_decision": "escalate",
+    "llm_narrative": "The five customers (CUST0004‑CUST0008) form a strong‑anchored cluster with seven strong links and no weak links, collectively transacting $63,837. Their return rate of 37.5% and chargeback rate of 6.25% are elevated, supporting the high risk probability of 99.6%.",
+    "llm_confidence_statement": "The evidence yields a very high confidence in the elevated fraud risk assessment.",
+    "action_taken": {
+      "tool": "razorpay.Client.order.create + order.edit + order.fetch",
+      "order_id": "order_TSeinwRO4xKvs9",
+      "razorpay_status": "created",
+      "notes_written": {
+        "case_id": "CASE10",
+        "cluster_id": "CAND10",
+        "ringguard_status": "escalate_flagged",
+        "action_label": "FRAUD_RING_ESCALATE",
+        "risk_score": "0.9959",
+        "action_by": "RingGuard_v3",
+        "action_timestamp": "2026-08-22T02:47:19.567020+00:00"
+      },
+      "notes_verified": true,
+      "timestamp": "2026-08-22T02:47:20.455513+00:00",
+      "test_mode": true,
+      "error": null
+    },
+    "timestamp": "2026-08-22T02:42:40.796349+00:00"
+  },
+  {
+    "case_id": "CASE11",
+    "cluster_id": "CAND11",
+    "customer_ids": [
+      "CUST0030",
+      "CUST0031",
+      "CUST0032"
+    ],
+    "risk_probability": 0.9927212653417073,
+    "policy_decision": "escalate",
+    "llm_narrative": "The three customers (CUST0030, CUST0031, CUST0032) form a strong‑anchored cluster with two strong links and no weak links, together generating $26,235 in transactions. Their return and chargeback rates are each 28.6%, and the model assigns a risk probability of 99.3%.",
+    "llm_confidence_statement": "Confidence in this assessment is high given the strong links, elevated return/chargeback rates, and near‑certain risk probability.",
+    "action_taken": {
+      "tool": "razorpay.Client.order.create + order.edit + order.fetch",
+      "order_id": "order_TSeip2lX554G6F",
+      "razorpay_status": "created",
+      "notes_written": {
+        "case_id": "CASE11",
+        "cluster_id": "CAND11",
+        "ringguard_status": "escalate_flagged",
+        "action_label": "FRAUD_RING_ESCALATE",
+        "risk_score": "0.9927",
+        "action_by": "RingGuard_v3",
+        "action_timestamp": "2026-08-22T02:47:20.576284+00:00"
+      },
+      "notes_verified": true,
+      "timestamp": "2026-08-22T02:47:21.336242+00:00",
+      "test_mode": true,
+      "error": null
+    },
+    "timestamp": "2026-08-22T02:42:46.478223+00:00"
+  },
+  {
+    "case_id": "CASE12",
+    "cluster_id": "CAND12",
+    "customer_ids": [
+      "CUST0001",
+      "CUST0019",
+      "CUST0040",
+      "CUST0126",
+      "CUST0138",
+      "CUST0168",
+      "CUST0211",
+      "CUST0219",
+      "CUST0258",
+      "CUST0281",
+      "CUST0292",
+      "CUST0322",
+      "CUST0325",
+      "CUST0371",
+      "CUST0373",
+      "CUST0400",
+      "CUST0419",
+      "CUST0465"
+    ],
+    "risk_probability": 0.00720344453548541,
+    "policy_decision": "log_only",
+    "llm_narrative": null,
+    "llm_confidence_statement": null,
+    "action_taken": null,
+    "timestamp": "2026-08-22T02:42:46.483242+00:00"
+  },
+  {
+    "case_id": "CASE13",
+    "cluster_id": "CAND13",
+    "customer_ids": [
+      "CUST0002",
+      "CUST0009",
+      "CUST0020",
+      "CUST0041",
+      "CUST0064",
+      "CUST0073",
+      "CUST0074",
+      "CUST0080",
+      "CUST0081",
+      "CUST0086",
+      "CUST0089",
+      "CUST0103",
+      "CUST0162",
+      "CUST0167",
+      "CUST0173",
+      "CUST0181",
+      "CUST0226",
+      "CUST0240",
+      "CUST0253",
+      "CUST0262",
+      "CUST0272",
+      "CUST0286",
+      "CUST0287",
+      "CUST0324",
+      "CUST0331",
+      "CUST0333",
+      "CUST0369",
+      "CUST0392",
+      "CUST0393",
+      "CUST0398",
+      "CUST0406",
+      "CUST0410",
+      "CUST0411",
+      "CUST0413",
+      "CUST0415",
+      "CUST0423",
+      "CUST0446",
+      "CUST0468",
+      "CUST0476",
+      "CUST0478",
+      "CUST0485"
+    ],
+    "risk_probability": 0.007559966037372132,
+    "policy_decision": "log_only",
+    "llm_narrative": null,
+    "llm_confidence_statement": null,
+    "action_taken": null,
+    "timestamp": "2026-08-22T02:42:46.486423+00:00"
+  },
+  {
+    "case_id": "CASE14",
+    "cluster_id": "CAND14",
+    "customer_ids": [
+      "CUST0003",
+      "CUST0012",
+      "CUST0043",
+      "CUST0077",
+      "CUST0088",
+      "CUST0098",
+      "CUST0109",
+      "CUST0125",
+      "CUST0139",
+      "CUST0153",
+      "CUST0159",
+      "CUST0166",
+      "CUST0180",
+      "CUST0201",
+      "CUST0209",
+      "CUST0245",
+      "CUST0261",
+      "CUST0265",
+      "CUST0273",
+      "CUST0277",
+      "CUST0297",
+      "CUST0315",
+      "CUST0334",
+      "CUST0347",
+      "CUST0381",
+      "CUST0405",
+      "CUST0412",
+      "CUST0428",
+      "CUST0429",
+      "CUST0462",
+      "CUST0471"
+    ],
+    "risk_probability": 0.0033586500336385537,
+    "policy_decision": "log_only",
+    "llm_narrative": null,
+    "llm_confidence_statement": null,
+    "action_taken": null,
+    "timestamp": "2026-08-22T02:42:46.490015+00:00"
+  },
+  {
+    "case_id": "CASE15",
+    "cluster_id": "CAND15",
+    "customer_ids": [
+      "CUST0004",
+      "CUST0008",
+      "CUST0027",
+      "CUST0030",
+      "CUST0042",
+      "CUST0063",
+      "CUST0075",
+      "CUST0087",
+      "CUST0110",
+      "CUST0117",
+      "CUST0120",
+      "CUST0133",
+      "CUST0135",
+      "CUST0137",
+      "CUST0144",
+      "CUST0152",
+      "CUST0154",
+      "CUST0163",
+      "CUST0170",
+      "CUST0178",
+      "CUST0179",
+      "CUST0195",
+      "CUST0198",
+      "CUST0208",
+      "CUST0210",
+      "CUST0221",
+      "CUST0228",
+      "CUST0231",
+      "CUST0237",
+      "CUST0246",
+      "CUST0251",
+      "CUST0254",
+      "CUST0266",
+      "CUST0282",
+      "CUST0295",
+      "CUST0300",
+      "CUST0303",
+      "CUST0304",
+      "CUST0335",
+      "CUST0355",
+      "CUST0365",
+      "CUST0399",
+      "CUST0420",
+      "CUST0425",
+      "CUST0470",
+      "CUST0483",
+      "CUST0486",
+      "CUST0487",
+      "CUST0491",
+      "CUST0492",
+      "CUST0498"
+    ],
+    "risk_probability": 0.007646314104742993,
+    "policy_decision": "log_only",
+    "llm_narrative": null,
+    "llm_confidence_statement": null,
+    "action_taken": null,
+    "timestamp": "2026-08-22T02:42:46.491127+00:00"
+  },
+  {
+    "case_id": "CASE16",
+    "cluster_id": "CAND16",
+    "customer_ids": [
+      "CUST0005",
+      "CUST0022",
+      "CUST0028",
+      "CUST0038",
+      "CUST0102",
+      "CUST0111",
+      "CUST0113",
+      "CUST0140",
+      "CUST0161",
+      "CUST0171",
+      "CUST0177",
+      "CUST0202",
+      "CUST0222",
+      "CUST0235",
+      "CUST0252",
+      "CUST0257",
+      "CUST0263",
+      "CUST0288",
+      "CUST0293",
+      "CUST0299",
+      "CUST0308",
+      "CUST0311",
+      "CUST0353",
+      "CUST0357",
+      "CUST0364",
+      "CUST0370",
+      "CUST0383",
+      "CUST0386",
+      "CUST0387",
+      "CUST0391",
+      "CUST0396",
+      "CUST0438",
+      "CUST0442",
+      "CUST0456",
+      "CUST0460",
+      "CUST0488",
+      "CUST0490",
+      "CUST0495",
+      "CUST0496",
+      "CUST0500"
+    ],
+    "risk_probability": 0.0035456513474297945,
+    "policy_decision": "log_only",
+    "llm_narrative": null,
+    "llm_confidence_statement": null,
+    "action_taken": null,
+    "timestamp": "2026-08-22T02:42:46.495874+00:00"
+  },
+  {
+    "case_id": "CASE17",
+    "cluster_id": "CAND17",
+    "customer_ids": [
+      "CUST0006",
+      "CUST0011",
+      "CUST0024",
+      "CUST0025",
+      "CUST0033",
+      "CUST0044",
+      "CUST0079",
+      "CUST0084",
+      "CUST0101",
+      "CUST0107",
+      "CUST0108",
+      "CUST0114",
+      "CUST0132",
+      "CUST0145",
+      "CUST0157",
+      "CUST0172",
+      "CUST0227",
+      "CUST0244",
+      "CUST0255",
+      "CUST0256",
+      "CUST0267",
+      "CUST0271",
+      "CUST0284",
+      "CUST0291",
+      "CUST0296",
+      "CUST0316",
+      "CUST0319",
+      "CUST0329",
+      "CUST0336",
+      "CUST0342",
+      "CUST0348",
+      "CUST0349",
+      "CUST0367",
+      "CUST0375",
+      "CUST0389",
+      "CUST0407",
+      "CUST0455",
+      "CUST0489",
+      "CUST0499"
+    ],
+    "risk_probability": 0.006896318028648148,
+    "policy_decision": "log_only",
+    "llm_narrative": null,
+    "llm_confidence_statement": null,
+    "action_taken": null,
+    "timestamp": "2026-08-22T02:42:46.500167+00:00"
+  },
+  {
+    "case_id": "CASE18",
+    "cluster_id": "CAND18",
+    "customer_ids": [
+      "CUST0007",
+      "CUST0014",
+      "CUST0015",
+      "CUST0016",
+      "CUST0032",
+      "CUST0037",
+      "CUST0065",
+      "CUST0069",
+      "CUST0070",
+      "CUST0095",
+      "CUST0106",
+      "CUST0115",
+      "CUST0124",
+      "CUST0131",
+      "CUST0149",
+      "CUST0191",
+      "CUST0200",
+      "CUST0204",
+      "CUST0215",
+      "CUST0218",
+      "CUST0223",
+      "CUST0238",
+      "CUST0239",
+      "CUST0241",
+      "CUST0242",
+      "CUST0264",
+      "CUST0276",
+      "CUST0280",
+      "CUST0305",
+      "CUST0314",
+      "CUST0323",
+      "CUST0327",
+      "CUST0339",
+      "CUST0350",
+      "CUST0352",
+      "CUST0356",
+      "CUST0358",
+      "CUST0362",
+      "CUST0379",
+      "CUST0401",
+      "CUST0418",
+      "CUST0440",
+      "CUST0443",
+      "CUST0449",
+      "CUST0479"
+    ],
+    "risk_probability": 0.0034784417949967547,
+    "policy_decision": "log_only",
+    "llm_narrative": null,
+    "llm_confidence_statement": null,
+    "action_taken": null,
+    "timestamp": "2026-08-22T02:42:46.502500+00:00"
+  },
+  {
+    "case_id": "CASE19",
+    "cluster_id": "CAND19",
+    "customer_ids": [
+      "CUST0010",
+      "CUST0023",
+      "CUST0029",
+      "CUST0090",
+      "CUST0119",
+      "CUST0121",
+      "CUST0129",
+      "CUST0141",
+      "CUST0143",
+      "CUST0155",
+      "CUST0196",
+      "CUST0207",
+      "CUST0230",
+      "CUST0233",
+      "CUST0247",
+      "CUST0330",
+      "CUST0338",
+      "CUST0372",
+      "CUST0374",
+      "CUST0380",
+      "CUST0461",
+      "CUST0467",
+      "CUST0480",
+      "CUST0493"
+    ],
+    "risk_probability": 0.007559966037372132,
+    "policy_decision": "log_only",
+    "llm_narrative": null,
+    "llm_confidence_statement": null,
+    "action_taken": null,
+    "timestamp": "2026-08-22T02:42:46.505560+00:00"
+  },
+  {
+    "case_id": "CASE20",
+    "cluster_id": "CAND20",
+    "customer_ids": [
+      "CUST0013",
+      "CUST0026",
+      "CUST0031",
+      "CUST0035",
+      "CUST0039",
+      "CUST0067",
+      "CUST0068",
+      "CUST0094",
+      "CUST0116",
+      "CUST0122",
+      "CUST0128",
+      "CUST0184",
+      "CUST0193",
+      "CUST0203",
+      "CUST0212",
+      "CUST0217",
+      "CUST0248",
+      "CUST0268",
+      "CUST0283",
+      "CUST0312",
+      "CUST0340",
+      "CUST0351",
+      "CUST0384",
+      "CUST0402",
+      "CUST0417",
+      "CUST0427",
+      "CUST0430",
+      "CUST0434",
+      "CUST0451",
+      "CUST0463",
+      "CUST0466",
+      "CUST0469",
+      "CUST0473",
+      "CUST0482",
+      "CUST0497"
+    ],
+    "risk_probability": 0.007561351403107364,
+    "policy_decision": "log_only",
+    "llm_narrative": null,
+    "llm_confidence_statement": null,
+    "action_taken": null,
+    "timestamp": "2026-08-22T02:42:46.508965+00:00"
+  },
+  {
+    "case_id": "CASE21",
+    "cluster_id": "CAND21",
+    "customer_ids": [
+      "CUST0017",
+      "CUST0036",
+      "CUST0062",
+      "CUST0071",
+      "CUST0082",
+      "CUST0083",
+      "CUST0085",
+      "CUST0096",
+      "CUST0123",
+      "CUST0142",
+      "CUST0147",
+      "CUST0175",
+      "CUST0189",
+      "CUST0216",
+      "CUST0220",
+      "CUST0224",
+      "CUST0232",
+      "CUST0234",
+      "CUST0306",
+      "CUST0328",
+      "CUST0337",
+      "CUST0341",
+      "CUST0345",
+      "CUST0346",
+      "CUST0361",
+      "CUST0388",
+      "CUST0421",
+      "CUST0426",
+      "CUST0433",
+      "CUST0435",
+      "CUST0444",
+      "CUST0450",
+      "CUST0454",
+      "CUST0477"
+    ],
+    "risk_probability": 0.0035456513474297945,
+    "policy_decision": "log_only",
+    "llm_narrative": null,
+    "llm_confidence_statement": null,
+    "action_taken": null,
+    "timestamp": "2026-08-22T02:42:46.514247+00:00"
+  },
+  {
+    "case_id": "CASE22",
+    "cluster_id": "CAND22",
+    "customer_ids": [
+      "CUST0018",
+      "CUST0021",
+      "CUST0091",
+      "CUST0092",
+      "CUST0112",
+      "CUST0160",
+      "CUST0174",
+      "CUST0187",
+      "CUST0270",
+      "CUST0298",
+      "CUST0307",
+      "CUST0310",
+      "CUST0332",
+      "CUST0343",
+      "CUST0431",
+      "CUST0474",
+      "CUST0475",
+      "CUST0494"
+    ],
+    "risk_probability": 0.03305941612108949,
+    "policy_decision": "log_only",
+    "llm_narrative": null,
+    "llm_confidence_statement": null,
+    "action_taken": null,
+    "timestamp": "2026-08-22T02:42:46.516340+00:00"
+  },
+  {
+    "case_id": "CASE23",
+    "cluster_id": "CAND23",
+    "customer_ids": [
+      "CUST0034",
+      "CUST0066",
+      "CUST0078",
+      "CUST0100",
+      "CUST0104",
+      "CUST0134",
+      "CUST0146",
+      "CUST0150",
+      "CUST0151",
+      "CUST0158",
+      "CUST0182",
+      "CUST0183",
+      "CUST0185",
+      "CUST0188",
+      "CUST0190",
+      "CUST0259",
+      "CUST0269",
+      "CUST0279",
+      "CUST0285",
+      "CUST0289",
+      "CUST0301",
+      "CUST0302",
+      "CUST0344",
+      "CUST0354",
+      "CUST0366",
+      "CUST0368",
+      "CUST0378",
+      "CUST0404",
+      "CUST0408",
+      "CUST0432",
+      "CUST0436",
+      "CUST0437",
+      "CUST0439",
+      "CUST0441",
+      "CUST0447",
+      "CUST0448",
+      "CUST0457",
+      "CUST0481"
+    ],
+    "risk_probability": 0.0035456513474297945,
+    "policy_decision": "log_only",
+    "llm_narrative": null,
+    "llm_confidence_statement": null,
+    "action_taken": null,
+    "timestamp": "2026-08-22T02:42:46.520858+00:00"
+  },
+  {
+    "case_id": "CASE24",
+    "cluster_id": "CAND24",
+    "customer_ids": [
+      "CUST0061",
+      "CUST0093",
+      "CUST0105",
+      "CUST0130",
+      "CUST0136",
+      "CUST0148",
+      "CUST0164",
+      "CUST0186",
+      "CUST0197",
+      "CUST0205",
+      "CUST0206",
+      "CUST0214",
+      "CUST0260",
+      "CUST0309",
+      "CUST0320",
+      "CUST0321",
+      "CUST0359",
+      "CUST0363",
+      "CUST0376",
+      "CUST0377",
+      "CUST0385",
+      "CUST0394",
+      "CUST0403",
+      "CUST0416",
+      "CUST0424",
+      "CUST0453"
+    ],
+    "risk_probability": 0.0035456513474297945,
+    "policy_decision": "log_only",
+    "llm_narrative": null,
+    "llm_confidence_statement": null,
+    "action_taken": null,
+    "timestamp": "2026-08-22T02:42:46.523858+00:00"
+  },
+  {
+    "case_id": "CASE25",
+    "cluster_id": "CAND25",
+    "customer_ids": [
+      "CUST0045",
+      "CUST0046",
+      "CUST0047",
+      "CUST0048"
+    ],
+    "risk_probability": 0.0035484719416527992,
+    "policy_decision": "log_only",
+    "llm_narrative": null,
+    "llm_confidence_statement": null,
+    "action_taken": null,
+    "timestamp": "2026-08-22T02:42:46.526962+00:00"
+  },
+  {
+    "case_id": "CASE26",
+    "cluster_id": "CAND26",
+    "customer_ids": [
+      "CUST0049",
+      "CUST0050",
+      "CUST0051",
+      "CUST0052"
+    ],
+    "risk_probability": 0.0031715868230902835,
+    "policy_decision": "log_only",
+    "llm_narrative": null,
+    "llm_confidence_statement": null,
+    "action_taken": null,
+    "timestamp": "2026-08-22T02:42:46.530230+00:00"
+  },
+  {
+    "case_id": "CASE27",
+    "cluster_id": "CAND27",
+    "customer_ids": [
+      "CUST0053",
+      "CUST0054",
+      "CUST0055",
+      "CUST0056"
+    ],
+    "risk_probability": 0.003058997865465553,
+    "policy_decision": "log_only",
+    "llm_narrative": null,
+    "llm_confidence_statement": null,
+    "action_taken": null,
+    "timestamp": "2026-08-22T02:42:46.534325+00:00"
+  },
+  {
+    "case_id": "CASE28",
+    "cluster_id": "CAND28",
+    "customer_ids": [
+      "CUST0504",
+      "CUST0505",
+      "CUST0506"
+    ],
+    "risk_probability": 0.003546939654202544,
+    "policy_decision": "log_only",
+    "llm_narrative": null,
+    "llm_confidence_statement": null,
+    "action_taken": null,
+    "timestamp": "2026-08-22T02:42:46.537834+00:00"
+  }
+]

data/synthetic/devices.json

@@ -0,0 +1,1622 @@
+[
+  {
+    "device_id": "DEV0001",
+    "fingerprint_hash": "80dd747f3716dd56b933c731983355211eea88582cb7cb97b1ee0602fe525d5b"
+  },
+  {
+    "device_id": "DEV0002",
+    "fingerprint_hash": "a2ca018cf73581bf404317a767e526092c45f55a7a6a2cc8a174a9151f3ce7dd"
+  },
+  {
+    "device_id": "DEV0003",
+    "fingerprint_hash": "c99d12ce6ca80227ac35a8c202fa708ddc95933c2c04e44e9eaf42792ce7bceb"
+  },
+  {
+    "device_id": "DEV0004",
+    "fingerprint_hash": "f927513b1a5a6e6e5814b13c49aca9ca4e1eac05e3cdbf5412a32a29cde57d69"
+  },
+  {
+    "device_id": "DEV0005",
+    "fingerprint_hash": "c0ea731f429a0dbfd7eb578e56342e57dd88a032a846ef4bab118927b7288cae"
+  },
+  {
+    "device_id": "DEV0006",
+    "fingerprint_hash": "85a6f257b8128daf5841388e36925a7188d80e60b603d5249c6f2c3384a06bf0"
+  },
+  {
+    "device_id": "DEV0007",
+    "fingerprint_hash": "3f7ad5dfdc47e2f1bdc56a302140cee0f22d497d7d522fb8866a8d01219fe156"
+  },
+  {
+    "device_id": "DEV0008",
+    "fingerprint_hash": "8b986911a10d39a2dfb0c5e5b8c576802a6df1ee945aee55e232926db3f513a1"
+  },
+  {
+    "device_id": "DEV0009",
+    "fingerprint_hash": "24cb30e25cd01403efb4b1b136d0f894cf1a418bc2eae55d085bb678b8a1c347"
+  },
+  {
+    "device_id": "DEV0010",
+    "fingerprint_hash": "de873c3e1a42235d7c544dd2e2f19a43c7f6fde42c7916780f8a89628c3e54dd"
+  },
+  {
+    "device_id": "DEV0011",
+    "fingerprint_hash": "82f5916c950f3834351b58856c880fe4094cf06ce20db6e7f4bf480c2835cf56"
+  },
+  {
+    "device_id": "DEV0012",
+    "fingerprint_hash": "e2f01e9e89b7a4ca0cecf625d781caec16e012be73bb28170b94db3ac5dcc11e"
+  },
+  {
+    "device_id": "DEV0013",
+    "fingerprint_hash": "92b0e7cc6b9af30e56ef6ebb34ecdd7ec2b28b810c4e7ddc196dff166339da14"
+  },
+  {
+    "device_id": "DEV0014",
+    "fingerprint_hash": "a28ac6a502fe2431f72e1f8f0b7d1a081279a5f12c98f2dde7cb6c747f1726e9"
+  },
+  {
+    "device_id": "DEV0015",
+    "fingerprint_hash": "8f745cdeba5ce6854a397da814afd57453b0bff7aa03f47554e8049b112c879a"
+  },
+  {
+    "device_id": "DEV0016",
+    "fingerprint_hash": "e55f152153a5de8d4b958763145e990020c059c770f85e3be729ad324872e429"
+  },
+  {
+    "device_id": "DEV0017",
+    "fingerprint_hash": "90348642cc0b1e13e5b07e17d185255df894eb65ab2dc3d455bc85ec5a1fb1b0"
+  },
+  {
+    "device_id": "DEV0018",
+    "fingerprint_hash": "b0cea960e30dabadbcce56690cc31a9b014e69495d6ed0a3bcc1a8dba1f560c5"
+  },
+  {
+    "device_id": "DEV0019",
+    "fingerprint_hash": "95ac359e7f63862c3b000ff6d7823988535a6d46d63691be7671be8ae5df10a9"
+  },
+  {
+    "device_id": "DEV0020",
+    "fingerprint_hash": "09bd5e6e6fbd47340b2e522f08673748ad4b221bb13acde2f3c852129d58feaa"
+  },
+  {
+    "device_id": "DEV0021",
+    "fingerprint_hash": "ada149be46843f511551113fd3be1c4bcac48adc698a6742dff5642441bae07f"
+  },
+  {
+    "device_id": "DEV0022",
+    "fingerprint_hash": "28ae808824e145ae8d20deebf8379231942fa3a936dbee29a231f483471535c5"
+  },
+  {
+    "device_id": "DEV0023",
+    "fingerprint_hash": "dd93d76752cf7a23c9ffeae68b31852f920a2120867e579a9ec6b8751b13b19d"
+  },
+  {
+    "device_id": "DEV0024",
+    "fingerprint_hash": "66f59f5593724ee82d76af4b65b29bc37b6a5e929acd8c9c4edc7bc1faf0754c"
+  },
+  {
+    "device_id": "DEV0025",
+    "fingerprint_hash": "e247d3458befeaa26ba592d5efb3c6a9d7e931266153015768bd7e43a6dc4f18"
+  },
+  {
+    "device_id": "DEV0026",
+    "fingerprint_hash": "15c786452a55638b8e1ef4741f1919d391e4206766d8db03ecfab29abc08ce9c"
+  },
+  {
+    "device_id": "DEV0027",
+    "fingerprint_hash": "11987d41b7ebe2e0e08d64441810c22dd5406d981e686b9122c6fc75cda8ea51"
+  },
+  {
+    "device_id": "DEV0028",
+    "fingerprint_hash": "ee9b014777a4a7f42658efbaaaeecdb56ae54417649ccba9e315c49f4d029f55"
+  },
+  {
+    "device_id": "DEV0029",
+    "fingerprint_hash": "82c4fcc7343a13da1d8ca1995630bf94e4c48dc82aedaa0d6a940c9b877fe828"
+  },
+  {
+    "device_id": "DEV0030",
+    "fingerprint_hash": "3949979a4ed3302445afa9090c2400e68ce3cf1626f5bd5c56cf8e9beb0c324f"
+  },
+  {
+    "device_id": "DEV0031",
+    "fingerprint_hash": "bf180a97506eddf73f797624d1ca2747fe072a680d1f8feb87522e3a2863ab0d"
+  },
+  {
+    "device_id": "DEV0032",
+    "fingerprint_hash": "35e77b95db18460dcee46976eefdcb656b2538e568caa5c6464107f5b184b43f"
+  },
+  {
+    "device_id": "DEV0033",
+    "fingerprint_hash": "385d83f1e711a6c59b3dae4d4370becdd450d406f5484bdd1d2de4be6ee162dd"
+  },
+  {
+    "device_id": "DEV0034",
+    "fingerprint_hash": "02452a07c7d59ae73938060771f3a3c9e63817115d6c38d1509864019629f455"
+  },
+  {
+    "device_id": "DEV0035",
+    "fingerprint_hash": "070b608bd70d53b152dca52ca12317b6765cd688be2a88d5de4756b956e53231"
+  },
+  {
+    "device_id": "DEV0036",
+    "fingerprint_hash": "1aeae6eda5fb48cb52f612f619fc19dead171885df81e0a24ef8422f4e5a2b3f"
+  },
+  {
+    "device_id": "DEV0037",
+    "fingerprint_hash": "04897f616fa40d3c6f5a183e62e1b4baf5b84a54067fbbc4030f715a5ce85f52"
+  },
+  {
+    "device_id": "DEV0038",
+    "fingerprint_hash": "36347a0ba4c5f8a2b59939b278decd023ae03afad64fd19952d010fdc03e2f93"
+  },
+  {
+    "device_id": "DEV0039",
+    "fingerprint_hash": "19aad1bf58c2ea6d35b23ba6f8c4e8f74098097013e178306f24dee7a5296b0c"
+  },
+  {
+    "device_id": "DEV0040",
+    "fingerprint_hash": "d67ebb698454aaa8428ddbaf5c7c4f0ca27da2a3e62da0ec863ef921ecbb7f34"
+  },
+  {
+    "device_id": "DEV0041",
+    "fingerprint_hash": "cfe8eccabfbe132b62a82adaf7c1c5c9b7566a14e86236361374a3f2429a94d8"
+  },
+  {
+    "device_id": "DEV0042",
+    "fingerprint_hash": "b935a681b5961d6e0fcfaffb5a444c09eebe778ad40335225882706f6ad50f21"
+  },
+  {
+    "device_id": "DEV0043",
+    "fingerprint_hash": "5f530db327760254140d9436404e37d60ca32d8e0158516621e0e447c2292b0e"
+  },
+  {
+    "device_id": "DEV0044",
+    "fingerprint_hash": "c0a6b17ecc53ae4b3ffb43c273f61ef0e32eb9c616a30ed16a213b23b63e322c"
+  },
+  {
+    "device_id": "DEV0045",
+    "fingerprint_hash": "08e48a8732db10b3fcf8c108070736fce2876c5ffbe0f8bade0483c686905aba"
+  },
+  {
+    "device_id": "DEV0046",
+    "fingerprint_hash": "cf2fde6fa6d2405f45eac3d626ee25b406d8a060ea9158439ba188b33a4a3c0f"
+  },
+  {
+    "device_id": "DEV0047",
+    "fingerprint_hash": "c127cdbe3b45ed343cacdd309a58132392becf54e617b3b79aec8a191eaf015d"
+  },
+  {
+    "device_id": "DEV0048",
+    "fingerprint_hash": "61a2d4085c8248d06dd7c68c21e809a1a24bbcc6cc7f2ab2b2296dc98f5127b0"
+  },
+  {
+    "device_id": "DEV0049",
+    "fingerprint_hash": "88eaec76dac2f2438c7bebd81f9dbd0fe4538ecf14b3ff6fe4639f99f02c8a5f"
+  },
+  {
+    "device_id": "DEV0050",
+    "fingerprint_hash": "ca61a32fb5241c5b38f30ef63328eca2816cf15c8b35303c7ae9d7671a599bd8"
+  },
+  {
+    "device_id": "DEV0051",
+    "fingerprint_hash": "3b0151220be1f21df3293068efa7fe45d4076a9c401ed5e63aa73ab66d9fe696"
+  },
+  {
+    "device_id": "DEV0052",
+    "fingerprint_hash": "508c6ceaeb33916ee8b93c51f2b33309df303143351cd680317a0a221b794ab4"
+  },
+  {
+    "device_id": "DEV0053",
+    "fingerprint_hash": "6a5c6fc453aca12e208960d9d81629c889ade125b728ec1e7e099daca2f65163"
+  },
+  {
+    "device_id": "DEV0054",
+    "fingerprint_hash": "aed1fc09f50aabc03df59d27779c2dd5748acc98f1a1d91e22b1a9a5bade0f1b"
+  },
+  {
+    "device_id": "DEV0055",
+    "fingerprint_hash": "7e69d25fcb6a833c665622c788b2f8c4b80acc9f4c9dd9a105d56d415ac352df"
+  },
+  {
+    "device_id": "DEV0056",
+    "fingerprint_hash": "11eaac8509ad85a68ae64ce5c7024f833e44d6f1f7f71bf5a1b8a8f96c694e45"
+  },
+  {
+    "device_id": "DEV0057",
+    "fingerprint_hash": "d25e4e14d7c33840032cd72129ab233e57d854bda2bdb7ffab4cb596e29e6e34"
+  },
+  {
+    "device_id": "DEV0058",
+    "fingerprint_hash": "39f30301a776f5138f910aef0ade88f967829704f9d09ac71968c3a65b0fbcac"
+  },
+  {
+    "device_id": "DEV0059",
+    "fingerprint_hash": "52e183394b013d780ecf97c25a8d93794bbf40c5d589d2c7d32617153e62675f"
+  },
+  {
+    "device_id": "DEV0060",
+    "fingerprint_hash": "24fa85425cafa2a1797f09e9edd43862674ec0808202b573311cdce269abcdc5"
+  },
+  {
+    "device_id": "DEV0061",
+    "fingerprint_hash": "4b3ad38a628d961308e0ac9f295876f6ee9b1754fb95921a2d6eb42306923432"
+  },
+  {
+    "device_id": "DEV0062",
+    "fingerprint_hash": "f78512d228cbcd485d84a5797ad61ac3a48a8c3324ad2cb0bb82a5873dda2d43"
+  },
+  {
+    "device_id": "DEV0063",
+    "fingerprint_hash": "7d6b6550c63d369386428975597f30d0188246cc1b01d993e67853b9df636301"
+  },
+  {
+    "device_id": "DEV0064",
+    "fingerprint_hash": "1701d293a290a967c56fb940b5b1422b732bfd40cb6529afe3b8b8cd4758cffb"
+  },
+  {
+    "device_id": "DEV0065",
+    "fingerprint_hash": "593a3dbea28a2dcb92e04be796fa35ed20b2913dbb2e2af49b549f15d62b01b3"
+  },
+  {
+    "device_id": "DEV0066",
+    "fingerprint_hash": "c6e44d279a50e5762dd8952e0645a6819c7c1029e6bd2e80c6685545357d8b77"
+  },
+  {
+    "device_id": "DEV0067",
+    "fingerprint_hash": "6a0f79364d6c803e9e98ef1ab6d792bab13f500d5ddb2f2fe82e913b306f1b23"
+  },
+  {
+    "device_id": "DEV0068",
+    "fingerprint_hash": "833d50d4829bc0423421094c01bd1db405470b6b01faca640a4ba3e379e43112"
+  },
+  {
+    "device_id": "DEV0069",
+    "fingerprint_hash": "8cb6607072ce80bf74deaf9ebdca568d2c2e103fe549948db981cda1460bd325"
+  },
+  {
+    "device_id": "DEV0070",
+    "fingerprint_hash": "e47b3123946724dda2cab2a22db7da4bca1a1fddd5027c3d295600898f017578"
+  },
+  {
+    "device_id": "DEV0071",
+    "fingerprint_hash": "1fc951170dda341c54c80edb97e05842375a0a3e23568705e85bb9f7dbf81658"
+  },
+  {
+    "device_id": "DEV0072",
+    "fingerprint_hash": "eda05f844426790623c45c1a02431dcdbfbbde54e8c973759529b198760edf5d"
+  },
+  {
+    "device_id": "DEV0073",
+    "fingerprint_hash": "435a3b6bcf4051c62669d3542f57028d4d1ba0137a36f9f38a633332a5303649"
+  },
+  {
+    "device_id": "DEV0074",
+    "fingerprint_hash": "9bee8c9bfb9e65c9c300698d6e4e64e2aa99e5f5c36e7a0ab436d01b2c723d9c"
+  },
+  {
+    "device_id": "DEV0075",
+    "fingerprint_hash": "b617d2dde84b3bf25343094d5809c3a25fb56cd53ab79026249c5ada18b785c4"
+  },
+  {
+    "device_id": "DEV0076",
+    "fingerprint_hash": "8cfc0db2290e2f2bc460b4ec922d4a7276628e252215255b0f53ef1d5710c258"
+  },
+  {
+    "device_id": "DEV0077",
+    "fingerprint_hash": "88e8b456d78265c3342f618b3423eb8abbbccd8eed6937592c78723eccf7411a"
+  },
+  {
+    "device_id": "DEV0078",
+    "fingerprint_hash": "a8b66885ca54d4c85c107e6c27b948c7086bb67244470950347b7c7038470472"
+  },
+  {
+    "device_id": "DEV0079",
+    "fingerprint_hash": "ba15256e56e10ac5743d671438b3e11dc8e4b572a90f49d166f289bcfc47e7ec"
+  },
+  {
+    "device_id": "DEV0080",
+    "fingerprint_hash": "0b9a7c5dd71c9fae593b631c74f562858113bef1919f876b7bfb1995e84bf18a"
+  },
+  {
+    "device_id": "DEV0081",
+    "fingerprint_hash": "b60dce9f2842489e093d7bc9cffec7d253dfe515c2ad5ceeeee55b459daa2cb7"
+  },
+  {
+    "device_id": "DEV0082",
+    "fingerprint_hash": "cc9c8955681c577a3972d4a1004c6c2d2da09e4307958efdb91de32387c4957c"
+  },
+  {
+    "device_id": "DEV0083",
+    "fingerprint_hash": "8741acbe2d0c6b84021c5f1a68d55344a1998f658d41824cbe6686447d61c499"
+  },
+  {
+    "device_id": "DEV0084",
+    "fingerprint_hash": "3a67180bd370329e3d135c153cc84db335eaf25bc782dfe6c8933b74983aa201"
+  },
+  {
+    "device_id": "DEV0085",
+    "fingerprint_hash": "d427efd1ca295210073ae45fbcdabdc077344c7debb3c8efd22fae5697b82322"
+  },
+  {
+    "device_id": "DEV0086",
+    "fingerprint_hash": "bafc63bb7efdabdeecb6a7ec9d74e4fa9c884bc54972f4c1f2f72a260cc48c96"
+  },
+  {
+    "device_id": "DEV0087",
+    "fingerprint_hash": "39ac507dc3ef4d086fa60d8dc9a5be789e5a8bbd28af301d6e5d6265e95d7d6e"
+  },
+  {
+    "device_id": "DEV0088",
+    "fingerprint_hash": "817aa050481c4747e0505cbee4ea854e38ec50289cdfd24058e37959eab5509d"
+  },
+  {
+    "device_id": "DEV0089",
+    "fingerprint_hash": "29327096afa18aae754feccd40ea64aec5abcef342199322e718945f2439e911"
+  },
+  {
+    "device_id": "DEV0090",
+    "fingerprint_hash": "853681c1b26e7425d9b5147565e1ff775de1bfbc1be06161a5a1cd2cb9d113d4"
+  },
+  {
+    "device_id": "DEV0091",
+    "fingerprint_hash": "73e99ec7f1044cbd2318bd1c7ce3648a9d6a83797c2f69aacc7f028fe9942ff8"
+  },
+  {
+    "device_id": "DEV0092",
+    "fingerprint_hash": "41bc4b2975b1a1f96b919175e93bafe6750f3d2075a308ddf5b1754c17308d91"
+  },
+  {
+    "device_id": "DEV0093",
+    "fingerprint_hash": "b73548c18699ce47a63cba0d6632c1b704bf1a77ec74ddafbe0b7247c010b72b"
+  },
+  {
+    "device_id": "DEV0094",
+    "fingerprint_hash": "42213b8e1e6ce68c23aea644889bd79c2c1c116253f98dbea48ace33928c9093"
+  },
+  {
+    "device_id": "DEV0095",
+    "fingerprint_hash": "7a8c66680e4644618e21c8af793aad5fa54c5ecb67b3c0209ce0352982340be5"
+  },
+  {
+    "device_id": "DEV0096",
+    "fingerprint_hash": "f696db148acfdf2419d3f9ab885f62373d7b081e846e452045140839e320f7ae"
+  },
+  {
+    "device_id": "DEV0097",
+    "fingerprint_hash": "54e99a8b6818a9cb2c563c607fbb8c7d57721140b672eca70ddfe50f742cd9bb"
+  },
+  {
+    "device_id": "DEV0098",
+    "fingerprint_hash": "3975220db138f0e5ae60abb0ee590a18c9d5aeaf704dfd3ce8886b3d2dd4a165"
+  },
+  {
+    "device_id": "DEV0099",
+    "fingerprint_hash": "c71abbab99a200f2b44e74fa2db31eb6d7cd36251c6764aaa82fa6153047957c"
+  },
+  {
+    "device_id": "DEV0100",
+    "fingerprint_hash": "91fdbcb265b4e0f06b0dd9ea04f1de208497dd1626e8784057a6b4c5cfaf39e3"
+  },
+  {
+    "device_id": "DEV0101",
+    "fingerprint_hash": "a1947716850a2b2a0ff985aed22b25b3668b2b3cca1270adace18f703f6d25f2"
+  },
+  {
+    "device_id": "DEV0102",
+    "fingerprint_hash": "11df1f2a615dd7ce447ebbec5d6a89f328ed46685f1fd50be02d4f5749c601f3"
+  },
+  {
+    "device_id": "DEV0103",
+    "fingerprint_hash": "3d85970bad02acfdc6e9a6dfab738d26a4102b5b9bc2484008920982303c8d54"
+  },
+  {
+    "device_id": "DEV0104",
+    "fingerprint_hash": "0e3e3040d29b0b93da8cce005cf96febf46e6a7c01c6b4ad56f59d9cd67ee664"
+  },
+  {
+    "device_id": "DEV0105",
+    "fingerprint_hash": "24078f2a3924aae7ae12b4cb88ca8e7631b887d3fb93d3cd9615f18326864dfd"
+  },
+  {
+    "device_id": "DEV0106",
+    "fingerprint_hash": "4d3cf93f98b9ce67bf7753c2c65874b855f80e52cd6340dcebf00ad90d164f01"
+  },
+  {
+    "device_id": "DEV0107",
+    "fingerprint_hash": "b4b5f295d77209a1bc23a06d15fc5dc9d31804844ed47995ea4e1e9cfd148cf8"
+  },
+  {
+    "device_id": "DEV0108",
+    "fingerprint_hash": "83f6ed958b524931453f05f6ac6a8fd585dcf1dbf45ce76be59bc0462318193e"
+  },
+  {
+    "device_id": "DEV0109",
+    "fingerprint_hash": "3f51ed22788be7ce2b8d4cb5a435766cf2725b37061f6fcd27aaec20dbef746e"
+  },
+  {
+    "device_id": "DEV0110",
+    "fingerprint_hash": "0228e1ecc4f4ec1389c84f5e18c3872a3adc1615e2ce472c2c29ac050dd42a29"
+  },
+  {
+    "device_id": "DEV0111",
+    "fingerprint_hash": "e9d645dd83efe4cd562b73710254625d0d27ba8d6f2b896922868146011bf880"
+  },
+  {
+    "device_id": "DEV0112",
+    "fingerprint_hash": "f1b79819a725cf8540e185f7d65c030ae0bc2a889028b2925b138e98db7720f2"
+  },
+  {
+    "device_id": "DEV0113",
+    "fingerprint_hash": "b4d74a90f99ca43f8245cb0e4448785e51b9fd38a1df23c95332a6a3becf3a2a"
+  },
+  {
+    "device_id": "DEV0114",
+    "fingerprint_hash": "8cc68ca33e2d1a114d58215f396ef7f64ad017093e85885ca120ed9b5abe7a85"
+  },
+  {
+    "device_id": "DEV0115",
+    "fingerprint_hash": "f14ef2e9ac6efad90d2146d12f9e938cccd9b99498426caae15595ad71548e11"
+  },
+  {
+    "device_id": "DEV0116",
+    "fingerprint_hash": "0f8e632cf92aee8766989ed8d5673bb1d6677ca631c5adbdb1c06dc319c8140d"
+  },
+  {
+    "device_id": "DEV0117",
+    "fingerprint_hash": "31a317db95f8fa02ba39ec7eda32642c28946ff34d4cfcbd7d8c50846fb27990"
+  },
+  {
+    "device_id": "DEV0118",
+    "fingerprint_hash": "2ea8c20e389601eb119dd62202a3d18a1880ea6aa867da53d309c6fd37f5f61b"
+  },
+  {
+    "device_id": "DEV0119",
+    "fingerprint_hash": "adaf093609582ed99409272e44d76ca6d5ac9c8d076e955b80465197a1bc9f19"
+  },
+  {
+    "device_id": "DEV0120",
+    "fingerprint_hash": "5121f815bfb1639b79245f4b72b3b1f58fb2c91259d514ff713ba3615ac1f52e"
+  },
+  {
+    "device_id": "DEV0121",
+    "fingerprint_hash": "ea247c4bb3cba6cd75617c031f54da431304686119a9f3d7be23639f40ffc099"
+  },
+  {
+    "device_id": "DEV0122",
+    "fingerprint_hash": "26302b55508bdae671399da3870965494a485ba0701d8c01b1f0a9fa579810ab"
+  },
+  {
+    "device_id": "DEV0123",
+    "fingerprint_hash": "3ce8fd49960af87d309113e5b73a42d5e6db899b3689a826aac69b740060c6cc"
+  },
+  {
+    "device_id": "DEV0124",
+    "fingerprint_hash": "5252de879a6ce1b1c25270a458f2dc5473955462ab6a3c2c6ac8d37cac8b3f2f"
+  },
+  {
+    "device_id": "DEV0125",
+    "fingerprint_hash": "dbf386ebfc1134427e0710bcd250e7937cdcd77d06449a727f866d31f865001b"
+  },
+  {
+    "device_id": "DEV0126",
+    "fingerprint_hash": "ffdc1a0ee44a7d2a6066f826c860084d7a39198de7a288042ca249fbca9e6178"
+  },
+  {
+    "device_id": "DEV0127",
+    "fingerprint_hash": "b93c043125e7ac79650e85613a4b21b4e6edcdad28f3300da997cd2792518833"
+  },
+  {
+    "device_id": "DEV0128",
+    "fingerprint_hash": "dba1a97a422446923ddb8e9300700147a16c31479cb18ffdc4f103f022796462"
+  },
+  {
+    "device_id": "DEV0129",
+    "fingerprint_hash": "a52e480fa87f32ff9160a91b94d555dd1438765e36a158e92607cb3def940d03"
+  },
+  {
+    "device_id": "DEV0130",
+    "fingerprint_hash": "a139196ef3ce6f340448120c2e4a622092fc80d7b880227a656d1e957519ddb9"
+  },
+  {
+    "device_id": "DEV0131",
+    "fingerprint_hash": "1b82b311802452415be949f25d3475f4587bbf7ee01732dcce3ab8bf0552e05f"
+  },
+  {
+    "device_id": "DEV0132",
+    "fingerprint_hash": "dbbb1cb9ee3646719e6ae8170fb73ea5fd8406f403a6d32c31802f5033f8043f"
+  },
+  {
+    "device_id": "DEV0133",
+    "fingerprint_hash": "0140442ffe6f2815f974c243450ff57e38670d404bd7aa74072b4cda2b1b282c"
+  },
+  {
+    "device_id": "DEV0134",
+    "fingerprint_hash": "ac8feb34a99e6997a3d556ab2a0e833cc2ad09f336ebc3481787b4ff9448db67"
+  },
+  {
+    "device_id": "DEV0135",
+    "fingerprint_hash": "6138063d33ea88fc5d6f6b8c4249cb47934c8c272018fa2f127ac56003b531ad"
+  },
+  {
+    "device_id": "DEV0136",
+    "fingerprint_hash": "52597e82355f2a4a60f599d68f48416af919c67d60effd2beae178d61f664127"
+  },
+  {
+    "device_id": "DEV0137",
+    "fingerprint_hash": "650301a2c0289b93daf03894c43b20fca874dfa48eb9977e9b50d4497ce61bd1"
+  },
+  {
+    "device_id": "DEV0138",
+    "fingerprint_hash": "c107bbbadad87b75ffa29b367aa6d83b61ebf3c9b8c9e50477039870df596e8b"
+  },
+  {
+    "device_id": "DEV0139",
+    "fingerprint_hash": "e493efd4b5b56aa025cdc6a4db42b70a45d0faaa367792dd8c5788dd92ae2137"
+  },
+  {
+    "device_id": "DEV0140",
+    "fingerprint_hash": "f9e603caa31e2d663a62b9626ce0429c79a5e29e7e586c9703f8a998ffe77d7e"
+  },
+  {
+    "device_id": "DEV0141",
+    "fingerprint_hash": "c5ffe5ebca9062920bd1f5a3288b01216ff26ff7dfd16be8286cdbf16f5452a9"
+  },
+  {
+    "device_id": "DEV0142",
+    "fingerprint_hash": "0a67370ec6ff33ccd8df042e4e2d42ec9f6b73dd65a694eaa88c59627ccb7e55"
+  },
+  {
+    "device_id": "DEV0143",
+    "fingerprint_hash": "cd074ee306428c4c335ebc302c0f9f1c06a31a0c5088f2cb6a06308fb01d7f6e"
+  },
+  {
+    "device_id": "DEV0144",
+    "fingerprint_hash": "a30e6012968ce08aca308c81e0d5297ea387c0becbbdff83e69a518ba8f581c1"
+  },
+  {
+    "device_id": "DEV0145",
+    "fingerprint_hash": "721d0ff76c1bb31718fb7ad3e34c624d235f2a69a388ccac9a900efc1012b763"
+  },
+  {
+    "device_id": "DEV0146",
+    "fingerprint_hash": "735e4f332bcae2d65888e20f78ca33d2cb5a61a6af8b9d639691ba7fa912d30d"
+  },
+  {
+    "device_id": "DEV0147",
+    "fingerprint_hash": "2551ffb8d9624a19de5b38c18e96ee22a3ca589e6b2e81c92b6810b5904de2e6"
+  },
+  {
+    "device_id": "DEV0148",
+    "fingerprint_hash": "28b3d8e8661c5fe68c3adb1d8b39c3ad3cad5583d8d5ee436a5992a97c78e3f1"
+  },
+  {
+    "device_id": "DEV0149",
+    "fingerprint_hash": "c5821128931ca2fe459884b61732177af5e72dd4bfdad39f90c5853570f60084"
+  },
+  {
+    "device_id": "DEV0150",
+    "fingerprint_hash": "f3ac4b7a6b2cbc0ee82f77863ced05418effa7245f7af41a5e9dd4bbf7e5ee2d"
+  },
+  {
+    "device_id": "DEV0151",
+    "fingerprint_hash": "18256de44a5ff0d275facfdf91076d360b28618e09e30789dc61e140f4175b84"
+  },
+  {
+    "device_id": "DEV0152",
+    "fingerprint_hash": "d3a693d2859e142bd62896745814ba2d480488c85d8e19f2ca1f5a5d1193cf9c"
+  },
+  {
+    "device_id": "DEV0153",
+    "fingerprint_hash": "cc68965e9c4b321388dafd285335e1ab7022f4730c32297c825e9ffd406b5f05"
+  },
+  {
+    "device_id": "DEV0154",
+    "fingerprint_hash": "30cb9f08032174e5c6dd523f72e258d6884fcdb73957b1569586388c7ab00180"
+  },
+  {
+    "device_id": "DEV0155",
+    "fingerprint_hash": "b13a743e1569533dd0a0bbe43a26505df8c774167a25bc236794c622196885c7"
+  },
+  {
+    "device_id": "DEV0156",
+    "fingerprint_hash": "ec955baadd25ee1acb240cc18501f5f9b95c922848205dab7e9c52872759dcdc"
+  },
+  {
+    "device_id": "DEV0157",
+    "fingerprint_hash": "c970edd75ce821717e1611fbe03b372b6a38826841a24b5c6984cd86013d3b3b"
+  },
+  {
+    "device_id": "DEV0158",
+    "fingerprint_hash": "73ffa03d52bc3863276f3e295126e6cce5ee0ab35c13cb0406dac8e3226b4fb8"
+  },
+  {
+    "device_id": "DEV0159",
+    "fingerprint_hash": "262d1db466cb3cf4793433f66e35d494bddf2fc59c5b19138a4d21c398a5ff64"
+  },
+  {
+    "device_id": "DEV0160",
+    "fingerprint_hash": "03a317cb2a5f02a977952ba1247f9b038efb3dbd3efe86c6f9c48c7dc1cfc9df"
+  },
+  {
+    "device_id": "DEV0161",
+    "fingerprint_hash": "e62e5ea09c36d9041b6deee7c60dea3b7d05bd0f4e952244fcc2015da6f61859"
+  },
+  {
+    "device_id": "DEV0162",
+    "fingerprint_hash": "63dac30d619826d0b5599ab0d1d9073e0e15181ca13abbc3c31ff1977befb922"
+  },
+  {
+    "device_id": "DEV0163",
+    "fingerprint_hash": "edfc9813f8bee36ca7216bb8981bdbb3a374406c2ab58f5351a062dd415016f8"
+  },
+  {
+    "device_id": "DEV0164",
+    "fingerprint_hash": "19c390de45b31cbb776a6998c0f7fbd79c5e872c79a957aaa869bc93876f4b18"
+  },
+  {
+    "device_id": "DEV0165",
+    "fingerprint_hash": "ba91e9204373aee6955dc387ed84ecc395baf199f0d1ef2ef79ae26c37027e13"
+  },
+  {
+    "device_id": "DEV0166",
+    "fingerprint_hash": "7de60c23c00257130dbc2a5e08f298a247136149f0677d16e2961620806c9d37"
+  },
+  {
+    "device_id": "DEV0167",
+    "fingerprint_hash": "3d1c715cae8a61bf5d23450f090fbec7f78dfde039932cd367bcb5a77aaf5ca5"
+  },
+  {
+    "device_id": "DEV0168",
+    "fingerprint_hash": "fe7e8a3c0f0361b524fb7ca709e30190f80673ccd1062e95e8d36d4c81674430"
+  },
+  {
+    "device_id": "DEV0169",
+    "fingerprint_hash": "10ab263deda4fd612565c0afbe90ac77480cae22c3c294d9831f6744903d954f"
+  },
+  {
+    "device_id": "DEV0170",
+    "fingerprint_hash": "35be714a8ab083931ed33967d7f5a7c9f5d4d7623ac3c5f467d04344156482c8"
+  },
+  {
+    "device_id": "DEV0171",
+    "fingerprint_hash": "5f712ce44385b9425a21f5515cc9b1f2dae60f4966edeaeca67051a615435f35"
+  },
+  {
+    "device_id": "DEV0172",
+    "fingerprint_hash": "6e12751bebcde9f031d952f8ba517b33ecbfff0b35d57c268d6c31f91aae0f47"
+  },
+  {
+    "device_id": "DEV0173",
+    "fingerprint_hash": "eaeae7b62113013a4caad474c4bdb21625805de361025e403186ead5aa8b73c9"
+  },
+  {
+    "device_id": "DEV0174",
+    "fingerprint_hash": "edcbdab7ffa6a2ef975273b8697b64f6a31d1cd3be2403dc711dc047038a3f2f"
+  },
+  {
+    "device_id": "DEV0175",
+    "fingerprint_hash": "08242d9884fa31e09c1674ac77943077b3de52403cb7ae7d81f54294a02209b4"
+  },
+  {
+    "device_id": "DEV0176",
+    "fingerprint_hash": "7427e46e031bd27522540429f46d378b0017ca3a9db0816d4d4fe2a67f1786b2"
+  },
+  {
+    "device_id": "DEV0177",
+    "fingerprint_hash": "dfdef0d703d63aa90cb8f0ee363c42f5fc5116dc0f76937b24e6970b1f11a73a"
+  },
+  {
+    "device_id": "DEV0178",
+    "fingerprint_hash": "80b43b595727d6537c8dddbff9ef04e7d00314376a03d96b0066c01ccf18460e"
+  },
+  {
+    "device_id": "DEV0179",
+    "fingerprint_hash": "d311d35fd114608dd7e6bf5fc398f17451dc46a428cb2c3a8701ceba95e69156"
+  },
+  {
+    "device_id": "DEV0180",
+    "fingerprint_hash": "399e8635b6f3069a22005dff8ed4e5f0d80b5814915cd9c1a5749e7014c7029b"
+  },
+  {
+    "device_id": "DEV0181",
+    "fingerprint_hash": "fb5f2cf6d6f5746debc8e9c02b6ca4aeb94ae9b649440c27b3ba4e1897d1f49c"
+  },
+  {
+    "device_id": "DEV0182",
+    "fingerprint_hash": "acc4bebbdcb38a35a44c119a2fed8169e4b6394c585f28cc3229f28b4966690f"
+  },
+  {
+    "device_id": "DEV0183",
+    "fingerprint_hash": "b5136197ca98f8887336e5be7c4d6eed64fb251f1410ce5315cb9b072c4d2aa5"
+  },
+  {
+    "device_id": "DEV0184",
+    "fingerprint_hash": "c42374f6dca661d5bcaa89688e58d953f128b70efae7b10b09cc770508d46de5"
+  },
+  {
+    "device_id": "DEV0185",
+    "fingerprint_hash": "9e16aef0ecccddbc66e103dff3c2e8b667b26847e0726211e052e02c98e5c9fa"
+  },
+  {
+    "device_id": "DEV0186",
+    "fingerprint_hash": "18048cab487bb65348cd3a6efd795d5e063663bd4112cf6922e480a292ed885e"
+  },
+  {
+    "device_id": "DEV0187",
+    "fingerprint_hash": "3fb4eaca83182b5a815767762aed1f40aea0ba50602a2155a589dc377910ce0f"
+  },
+  {
+    "device_id": "DEV0188",
+    "fingerprint_hash": "55302dccf712906bc73745b0d517f5d77c06710f955b507751e4474278f1d76c"
+  },
+  {
+    "device_id": "DEV0189",
+    "fingerprint_hash": "fbc3969a93fa1c94ad256775bc8435fd679ba369e90950566e918be09c3cc9fa"
+  },
+  {
+    "device_id": "DEV0190",
+    "fingerprint_hash": "f60d71779db65024275c182d8b022869a8548727647525c56e6fe964c59d6e56"
+  },
+  {
+    "device_id": "DEV0191",
+    "fingerprint_hash": "68df74fc2576a202deba87e7cd152ee7dabc1324e2927a8c297faab3d668a1d5"
+  },
+  {
+    "device_id": "DEV0192",
+    "fingerprint_hash": "30d2ea953e10c9903ccf9ec5c90fa8177e2637a5c12b9299a00fd35f57c52f82"
+  },
+  {
+    "device_id": "DEV0193",
+    "fingerprint_hash": "a68aef0c676939714c2435d3d875614469ed7b4d04c4e090ddffdf377b514581"
+  },
+  {
+    "device_id": "DEV0194",
+    "fingerprint_hash": "155549f7554f1c87cb90ba6521ff6469704d5abd37264cbc7f45910bd19aea4c"
+  },
+  {
+    "device_id": "DEV0195",
+    "fingerprint_hash": "8a465d34184ce268b7944c823086408a3b90d264e74fbc1aa81b2c09a3bea06a"
+  },
+  {
+    "device_id": "DEV0196",
+    "fingerprint_hash": "fde7ddc0dc3e9846613438b51af0a9969d34fe11be1ef3b4f83004ca30bcf570"
+  },
+  {
+    "device_id": "DEV0197",
+    "fingerprint_hash": "3bf43de90d07f3b736ed8f19ea917cc07d553a4e03cdbcc213ea50e00d40c5ce"
+  },
+  {
+    "device_id": "DEV0198",
+    "fingerprint_hash": "196717070ceec0f3fb164befeefb465237be2621be068096a1765b6abdb32794"
+  },
+  {
+    "device_id": "DEV0199",
+    "fingerprint_hash": "470b7e115da1bc634a8619f86a6fc3c048462f1017060a8a2c6c2e53958b3bb3"
+  },
+  {
+    "device_id": "DEV0200",
+    "fingerprint_hash": "1aed95af8648d31779ae39ac199fdd1dd8c4165402a688106e4f3d488d10371b"
+  },
+  {
+    "device_id": "DEV0201",
+    "fingerprint_hash": "3fd06f0484a53da6310369cf67e632163c7774a7ea96805b1a81978a98ba8add"
+  },
+  {
+    "device_id": "DEV0202",
+    "fingerprint_hash": "9916255472143fa1306823d2297bfa345acf80f2a8a3f50ca84428549292ddfb"
+  },
+  {
+    "device_id": "DEV0203",
+    "fingerprint_hash": "3bb8027a989fd42a5f2bd3d970e6a99eb37c3783c7559491110a06f303ec184a"
+  },
+  {
+    "device_id": "DEV0204",
+    "fingerprint_hash": "0166621dddb3d2de86c769bafa3819c450986a7e0f53d1f2f1fc4a07186a0b9e"
+  },
+  {
+    "device_id": "DEV0205",
+    "fingerprint_hash": "812934ee92c13fc065e6b96fd019160a13b544c51479e1aa0a16e24be4c67e46"
+  },
+  {
+    "device_id": "DEV0206",
+    "fingerprint_hash": "63f4690e1222a02e0c9ffdad7c6f9dc0573d52728e4af47e05a733718c497da0"
+  },
+  {
+    "device_id": "DEV0207",
+    "fingerprint_hash": "bf4d8a98eb635042702a7b116eb4d54eef1fc80eaffbbb0f7629ffb5d4321c7b"
+  },
+  {
+    "device_id": "DEV0208",
+    "fingerprint_hash": "4b3f01fc311c8f07e6b4fb2c872f2caf3b6062ef177c1ab8b53a23965af2ee0f"
+  },
+  {
+    "device_id": "DEV0209",
+    "fingerprint_hash": "bbcb3a3f4a3af5c569397bbad60189e5999f7b0caefaa342342238d625efccaa"
+  },
+  {
+    "device_id": "DEV0210",
+    "fingerprint_hash": "af497f21beec53d976f1e3ad30874512ad7f7471e4325a6e592eaa78af64ce30"
+  },
+  {
+    "device_id": "DEV0211",
+    "fingerprint_hash": "2243cee127dff765b3e230b7c29e7826d4661ac30e348996371e5d51acf4b3c3"
+  },
+  {
+    "device_id": "DEV0212",
+    "fingerprint_hash": "79cbd127722cf00c7c76832eb89146f2551ff8dab46a82306bed96d62e4b9eb4"
+  },
+  {
+    "device_id": "DEV0213",
+    "fingerprint_hash": "b430c419103350580942a33ab997fa8841de9a072f43a5b61db95bbe79f225ce"
+  },
+  {
+    "device_id": "DEV0214",
+    "fingerprint_hash": "a0dfeda412b8f551552fb8869ebc7fa7b56771969aadb4bc2de7265c2a1dc2b6"
+  },
+  {
+    "device_id": "DEV0215",
+    "fingerprint_hash": "01e54077f641101fb0dc7ce0fad0674132a16b02f1f5f71ecbc8d54b5589f2c4"
+  },
+  {
+    "device_id": "DEV0216",
+    "fingerprint_hash": "3ef25793ccaf095048f58b1364f60a38c1716ebf5c25686d892d4c42b004849e"
+  },
+  {
+    "device_id": "DEV0217",
+    "fingerprint_hash": "72dea3fa1aa9f1986ec5bb8e8e595bad2844428f88c8d9ba0d4cad29ca827c21"
+  },
+  {
+    "device_id": "DEV0218",
+    "fingerprint_hash": "159e132d8a2c24f930254b0677edb3341d2ee86f167dc0ae06c4f22e815206a8"
+  },
+  {
+    "device_id": "DEV0219",
+    "fingerprint_hash": "9457c6d94fb8955bc2c97466f8edb6a7243713bcba553c22bbdb609a0fb89218"
+  },
+  {
+    "device_id": "DEV0220",
+    "fingerprint_hash": "434d62d11cdbc2344233496218fa868b7d7bf9a2bda9fbd47a75b669f6d52b5f"
+  },
+  {
+    "device_id": "DEV0221",
+    "fingerprint_hash": "689569376bbbb0cbcd2b05ff33dde57b7a4d195c1e36d5abd8723dc2044b723b"
+  },
+  {
+    "device_id": "DEV0222",
+    "fingerprint_hash": "de233ffba5a87bfe44c9e7073dfa47b99aaf5d13e478db692e4829d0e3218609"
+  },
+  {
+    "device_id": "DEV0223",
+    "fingerprint_hash": "e75e8776b891663a64004274cd0e4f8df70749289d68c1efb8ba08653c13ba49"
+  },
+  {
+    "device_id": "DEV0224",
+    "fingerprint_hash": "22c70e03cf0c912d7107d1b2a237d6fdee00d40d7c1855109bd5d046198c61f5"
+  },
+  {
+    "device_id": "DEV0225",
+    "fingerprint_hash": "88ad94ba3f7c793aa14b76d04eb7025448db4553394813f9831f167c8db3ba42"
+  },
+  {
+    "device_id": "DEV0226",
+    "fingerprint_hash": "5eda154a3084620278b83268e82b447069a7a71a31cbe9b27d1898a325bba394"
+  },
+  {
+    "device_id": "DEV0227",
+    "fingerprint_hash": "b2e3117cb302b9b9abc4659776045fe4bf0364f24af83ede1529de9a82c49f36"
+  },
+  {
+    "device_id": "DEV0228",
+    "fingerprint_hash": "1a3f2cbe674b27967acd7736658058420ec919821324517c9848e397d615705b"
+  },
+  {
+    "device_id": "DEV0229",
+    "fingerprint_hash": "4a9ee2642b4b61b64de554ac270e7e953ab4aff5cb476e07ef2623d0a9828bf9"
+  },
+  {
+    "device_id": "DEV0230",
+    "fingerprint_hash": "20edef9b43f6c907111bc698aa619f69318aa2e98adacb25ebeaa653b744dea2"
+  },
+  {
+    "device_id": "DEV0231",
+    "fingerprint_hash": "376df9db5dd92f6dbea17173268051ae42842ebca0e932cb67adfd12e6bed64a"
+  },
+  {
+    "device_id": "DEV0232",
+    "fingerprint_hash": "79f0c5218b27ca1db5e4476dd262a4cb4f85b208d500ef805d7edbd5820ff45d"
+  },
+  {
+    "device_id": "DEV0233",
+    "fingerprint_hash": "e09aa73f5c449d5d8575c72ae27a5e5f0dbafa799fa66fca87782d036a9497c1"
+  },
+  {
+    "device_id": "DEV0234",
+    "fingerprint_hash": "7eb3aa19d259892111f9e9b39ff4fd3f5d48e245ec41caf589944e252b63dffc"
+  },
+  {
+    "device_id": "DEV0235",
+    "fingerprint_hash": "7218b5c8dd5d350846f511e479d2e56fbcae4e1db85daa3999bf1c1234d205c8"
+  },
+  {
+    "device_id": "DEV0236",
+    "fingerprint_hash": "d699a6537aea630a34a686ea16b0d22091d625741115fb72411c5acf5d0ff7a7"
+  },
+  {
+    "device_id": "DEV0237",
+    "fingerprint_hash": "0a9852fa27a3bc4038cad803a71f4dbb138e2d24b22d11fbf0315b813b11f989"
+  },
+  {
+    "device_id": "DEV0238",
+    "fingerprint_hash": "cf1d3099b6d156a361bc5fee5468d22d32e65e3a73a35bfd1115ed5ff15f439a"
+  },
+  {
+    "device_id": "DEV0239",
+    "fingerprint_hash": "3fc2edda05b4805b46664b5c2c7f91646ff8241a3d147fd12f80479cb8e34a3a"
+  },
+  {
+    "device_id": "DEV0240",
+    "fingerprint_hash": "e5774e5ee75075147fcd1d974be5a2eb83b824084eac7d861c958558ee6f966d"
+  },
+  {
+    "device_id": "DEV0241",
+    "fingerprint_hash": "9803fba0a77ed4b9482e86af5326239b50d98babb7e54b8ab986b661b77af7b4"
+  },
+  {
+    "device_id": "DEV0242",
+    "fingerprint_hash": "77c5a3d62c14ce322a51bf61eefd89dd6746baa9c1f0fd8d7e44b2b63e2669aa"
+  },
+  {
+    "device_id": "DEV0243",
+    "fingerprint_hash": "f88fe6c35365ace1b3ce78060ad824cf9279f133a932b091bc8cedf61779f46d"
+  },
+  {
+    "device_id": "DEV0244",
+    "fingerprint_hash": "a072252dc49a4f06c01539e34951a966be452f1329d2178c82293a887b54f7d6"
+  },
+  {
+    "device_id": "DEV0245",
+    "fingerprint_hash": "9e8a0c8abce51ce29ebc5cfd6fc68fc6037746d687e82a4bab7c0bd9aaaca636"
+  },
+  {
+    "device_id": "DEV0246",
+    "fingerprint_hash": "eeaf23bc17309e86206849c70e2038796fb56433b97ae8ac139a30a4d062caea"
+  },
+  {
+    "device_id": "DEV0247",
+    "fingerprint_hash": "0130eb1f30cb0a6fa78294d19e7fc7cde8f6f735cf1a5c88a385058c5f9c5bc1"
+  },
+  {
+    "device_id": "DEV0248",
+    "fingerprint_hash": "e08ab35cb947b9eaecab0b3314084b554275838956e394d6b86d91a912767247"
+  },
+  {
+    "device_id": "DEV0249",
+    "fingerprint_hash": "361f3024cae380e9b8061edf1d5d678087f72d69e0fe4d92264a7da854679194"
+  },
+  {
+    "device_id": "DEV0250",
+    "fingerprint_hash": "940e9aed9510c520450126fe9ffc9162fa0f90521d34ae6c82308aca5a3ca28c"
+  },
+  {
+    "device_id": "DEV0251",
+    "fingerprint_hash": "45e76743c14d60d6624a559b24050c2a0da57af5e0dcb778c0ac9ec937158fdc"
+  },
+  {
+    "device_id": "DEV0252",
+    "fingerprint_hash": "b7cb19f546039f6018c5ab57d24c9abc0f6b38beff1d7b8f8e5f7e2388fe4d51"
+  },
+  {
+    "device_id": "DEV0253",
+    "fingerprint_hash": "2011dc6f9723fe4d62a6a83ee2fbb66bff88d69eb50d78df00b5332f47a53b46"
+  },
+  {
+    "device_id": "DEV0254",
+    "fingerprint_hash": "2bc0e4b1573a39910b25c224adf78f5a208b2c9e9ee0c69581984fa979ed8983"
+  },
+  {
+    "device_id": "DEV0255",
+    "fingerprint_hash": "e6bcc7a1ff1504ea43bf0077c22a4bf298f885aac584b064bda4556c2c653db5"
+  },
+  {
+    "device_id": "DEV0256",
+    "fingerprint_hash": "35a434f8fb1e4eda0eba55b969fe9a8c21a62cab301bfaf4f22774812fdd66fc"
+  },
+  {
+    "device_id": "DEV0257",
+    "fingerprint_hash": "a35e497c810174287ddde3c1fa0395d10e4d8c07af3465900ddcec6a28ff4e7d"
+  },
+  {
+    "device_id": "DEV0258",
+    "fingerprint_hash": "e485c5c5989595c09cf28550e0025728090528e05daea7959853a9a58f8ecc99"
+  },
+  {
+    "device_id": "DEV0259",
+    "fingerprint_hash": "97508201bc0fdc707cf88932607f02a86f68af7cd6e379bfbacdc72741e6fb06"
+  },
+  {
+    "device_id": "DEV0260",
+    "fingerprint_hash": "fe727d3d58ebbd52fe18ce7d7a511cbab09394a418fd07ac5159f729934d603d"
+  },
+  {
+    "device_id": "DEV0261",
+    "fingerprint_hash": "aef218c61649a7254d669ca7252acc60859fca2c572b677f675751f55afca419"
+  },
+  {
+    "device_id": "DEV0262",
+    "fingerprint_hash": "40aec91990555870f21788152adf5f62df4862570806062b98472cf2dff565ee"
+  },
+  {
+    "device_id": "DEV0263",
+    "fingerprint_hash": "94596f74aa2c8f4fea21917a2a1602be7a40ad5d0a81157b907ee5243b84f89a"
+  },
+  {
+    "device_id": "DEV0264",
+    "fingerprint_hash": "dd6575da4c3c0d996b18d3a249e3ec4fb6c5ca74e86e7ac91a8c627e3ba36314"
+  },
+  {
+    "device_id": "DEV0265",
+    "fingerprint_hash": "b14c03095bc14d5047ee9288f355d423144797b60b9a5a25e49a545923daace2"
+  },
+  {
+    "device_id": "DEV0266",
+    "fingerprint_hash": "c345b8bc61b558e4d721951dda3ebe120ce794894f369d58a21d3aac8ee801de"
+  },
+  {
+    "device_id": "DEV0267",
+    "fingerprint_hash": "8f079c8bde59314a5abf4b4204f47763111a0201aa45b9844bdf934f1214f9a4"
+  },
+  {
+    "device_id": "DEV0268",
+    "fingerprint_hash": "818f7ed297af84e3fb21d82bf2258e62dcaf8cf4595b6878c9df8de6cf185cd6"
+  },
+  {
+    "device_id": "DEV0269",
+    "fingerprint_hash": "b212ef972f70d9a305f8db4f8194f1895cb1ae8f9757887341c1bf635e32c755"
+  },
+  {
+    "device_id": "DEV0270",
+    "fingerprint_hash": "4a1e9beedf0802026ca15acefdbffa48f6c1c1b2a1d4f5a661d31645f9b3170f"
+  },
+  {
+    "device_id": "DEV0271",
+    "fingerprint_hash": "fa04e38b61bde2015f45df85cab4ef524d95b97e249d9d1d71b33923b0511c2f"
+  },
+  {
+    "device_id": "DEV0272",
+    "fingerprint_hash": "308d0ad5a47f101fca6824f75d4290c9a0605f3433b597466f110607ef85c1c0"
+  },
+  {
+    "device_id": "DEV0273",
+    "fingerprint_hash": "eb9ad067c8fb3774058140f2cce762de45c2369742df9175e5f69cdfc34ad762"
+  },
+  {
+    "device_id": "DEV0274",
+    "fingerprint_hash": "0411946c28629b5c574897d0e741147beb79406a17799172fb8494d2deff0d08"
+  },
+  {
+    "device_id": "DEV0275",
+    "fingerprint_hash": "f1d60cccec882aa844bc4adf031e0128b831f6f3dda5274ffd46160497fe7869"
+  },
+  {
+    "device_id": "DEV0276",
+    "fingerprint_hash": "56b6e4dba423f7a5dea21a7de3369b1beda058f0ba41970989333cabc8c96447"
+  },
+  {
+    "device_id": "DEV0277",
+    "fingerprint_hash": "0ae3d6cdfe244951e0ef6a388bc9c0717f96575255bd85fa4e6a81e8dc08e339"
+  },
+  {
+    "device_id": "DEV0278",
+    "fingerprint_hash": "05f9ec5f1fc4f2627d06046e46e5e2f4339a2b736914cd1ea8bc6f7d9537b6cd"
+  },
+  {
+    "device_id": "DEV0279",
+    "fingerprint_hash": "a02c75ae0aa9bf25d1eff0f6b4a42d166fe0dd51dd85b9656d052717c128ee1e"
+  },
+  {
+    "device_id": "DEV0280",
+    "fingerprint_hash": "ccc25f2c144d3a87db9fb0b0f78beaf568235375b95d2fadd1ba1ce526cb10a1"
+  },
+  {
+    "device_id": "DEV0281",
+    "fingerprint_hash": "f1eae9eb9074a96fbc1d9a85efcd2a62d8c6b26edd81bdd2a047a63612debf56"
+  },
+  {
+    "device_id": "DEV0282",
+    "fingerprint_hash": "0dba34249bf3758faf43772e874e4f570ec74b94a9ce0e08a70afb0b4704ba19"
+  },
+  {
+    "device_id": "DEV0283",
+    "fingerprint_hash": "799b9c978d16417fa96ed40ed5672bcee1baf092decc7d64f7dde926ec6673cd"
+  },
+  {
+    "device_id": "DEV0284",
+    "fingerprint_hash": "4fa0f0e227bcef0b5a418b27dc03cea0fde91aeca44639879e57c849951d3a5c"
+  },
+  {
+    "device_id": "DEV0285",
+    "fingerprint_hash": "17a16524c4c8db6ddca0c95f8484ecd560b1fa917d45533775f0df9643bc41eb"
+  },
+  {
+    "device_id": "DEV0286",
+    "fingerprint_hash": "f255c655bc371ccc72c60855f7dd2eecb3b4d049dd0b7580a760aa1e6917ebca"
+  },
+  {
+    "device_id": "DEV0287",
+    "fingerprint_hash": "42cb6adcc33bd094f6fb7bc27215cfacb12e2da9772e28d778b6d3b213ea4c44"
+  },
+  {
+    "device_id": "DEV0288",
+    "fingerprint_hash": "9884c9f3d31c4520bf51b7b6a86f5295055c48fd7e9dff4c2de429ad19c0c0f3"
+  },
+  {
+    "device_id": "DEV0289",
+    "fingerprint_hash": "63e04749b3acdd7d8b8ee6719e0e55a1200415bdd73d9afbdb217ed12f2d2dfa"
+  },
+  {
+    "device_id": "DEV0290",
+    "fingerprint_hash": "a5cb7af29ebb327d63fc53c95947b68d3b2fde3f42c48efe30f915b2f06f7330"
+  },
+  {
+    "device_id": "DEV0291",
+    "fingerprint_hash": "53db0e6254bfc42930a79acb60f131c7c335fd3ba6410e9e65bfc0767c4e1bb3"
+  },
+  {
+    "device_id": "DEV0292",
+    "fingerprint_hash": "196ce9eff0771823b670fdffd87a86a824180b4c4689e34652647c89ad2b6cd7"
+  },
+  {
+    "device_id": "DEV0293",
+    "fingerprint_hash": "75b4965fa965e6fe282fa84ad83153dfa39c511a927a8c1e1245356644340738"
+  },
+  {
+    "device_id": "DEV0294",
+    "fingerprint_hash": "d7cfd5bcddcc70d778aaf1569f13de81b2fdf861aaf5c8131d2fdce849514c3a"
+  },
+  {
+    "device_id": "DEV0295",
+    "fingerprint_hash": "226a74a548e1b68e061040bd4a15bc18de7200143f53b8411b8a77c02ce2c3cf"
+  },
+  {
+    "device_id": "DEV0296",
+    "fingerprint_hash": "96a97763395eecbfed4d35259f23b8445b5d880bb05e1c52607809102a6a7e8d"
+  },
+  {
+    "device_id": "DEV0297",
+    "fingerprint_hash": "4f94fdc30fb473d365c4f5c22325eba3ef4a3f1124b250d7be5d424f058a9542"
+  },
+  {
+    "device_id": "DEV0298",
+    "fingerprint_hash": "e0ead05768534c144dcb02e403ac009221fa218d46e9879aee2eda58104e6c5a"
+  },
+  {
+    "device_id": "DEV0299",
+    "fingerprint_hash": "7afd340e37cdae754441b2957375828bf78f49e6e1da9386d2ccead93664aa60"
+  },
+  {
+    "device_id": "DEV0300",
+    "fingerprint_hash": "478fa1e0bd46c71f36a41ea86aa124df311596972c889c89f4d6a64361460184"
+  },
+  {
+    "device_id": "DEV0301",
+    "fingerprint_hash": "ae3b48a177e6ea5825410f53698a05f4f125766e2dc0e2b096c6be4581239757"
+  },
+  {
+    "device_id": "DEV0302",
+    "fingerprint_hash": "530353fc4ca2e864505c2d8b41be50f0178117462afc99680b02a4f042b0ceba"
+  },
+  {
+    "device_id": "DEV0303",
+    "fingerprint_hash": "b94dc53c5f996b73a06d1d03384e77401b244bb186b8f4e7290342bb12b1e1bd"
+  },
+  {
+    "device_id": "DEV0304",
+    "fingerprint_hash": "097cd26a06a24db9a55a478fa2af082a469dcf07783b5d57efea758630750678"
+  },
+  {
+    "device_id": "DEV0305",
+    "fingerprint_hash": "f89ed8cea01b6a710988c5c8edb535e3c28e8bd3b34882af0cc21517c7706826"
+  },
+  {
+    "device_id": "DEV0306",
+    "fingerprint_hash": "342d8ebb34825e643d0009e899d584cba16905c113c27c384fe1650320e18877"
+  },
+  {
+    "device_id": "DEV0307",
+    "fingerprint_hash": "14ce23b877155a18113d840441bbde334ff214a75da160d49b5aae672878197a"
+  },
+  {
+    "device_id": "DEV0308",
+    "fingerprint_hash": "ef852f564bd35196718604f93edce4dfcd5f10b1dd00c0f9bce53a1bc77a5d70"
+  },
+  {
+    "device_id": "DEV0309",
+    "fingerprint_hash": "75ef62bad0ba4e4a8596d71bc28cbe9c4cbfc2e81aec002c6fc6a00d83232535"
+  },
+  {
+    "device_id": "DEV0310",
+    "fingerprint_hash": "27d91cab279adf59df1ba2717fa38058ee7641e26ada3ec7bb1c1293b57f38c3"
+  },
+  {
+    "device_id": "DEV0311",
+    "fingerprint_hash": "64ee79d9ea63b60c6c137a1ef834c6e7c4081893be4673cddc3bf5744c150528"
+  },
+  {
+    "device_id": "DEV0312",
+    "fingerprint_hash": "b6dcdf54f6e7dae07baa4dcf061f0d143438a85821e6c1a7c762f5ddef76b443"
+  },
+  {
+    "device_id": "DEV0313",
+    "fingerprint_hash": "73c6a13bd97ad4ddb142e05b3684222d3aa3d3ea492ac98d7a43a3d8de8dec21"
+  },
+  {
+    "device_id": "DEV0314",
+    "fingerprint_hash": "5593c852f7a6a271ed567b69f7977d1242ef28e54c315e443aa5ff73d9488946"
+  },
+  {
+    "device_id": "DEV0315",
+    "fingerprint_hash": "5482861ca7d115e8f2d8d52c63bb5bf354aa99622397c3edcff47df6968a61a9"
+  },
+  {
+    "device_id": "DEV0316",
+    "fingerprint_hash": "44495d2b8c99968fca1d0b42439a9db2ec74eb38d2562b070929ac6832b9c5ac"
+  },
+  {
+    "device_id": "DEV0317",
+    "fingerprint_hash": "6610a743bffff348da753cb730f657601a76f1615e1b326f24d6c41ff3764284"
+  },
+  {
+    "device_id": "DEV0318",
+    "fingerprint_hash": "7ddacc5115a19ed685f78416a488de531c1a31d25d5af726e73a1bc48888720a"
+  },
+  {
+    "device_id": "DEV0319",
+    "fingerprint_hash": "2075803ddee657b4e40349bbface85582ea4e196a6a74318edde22ae1589f971"
+  },
+  {
+    "device_id": "DEV0320",
+    "fingerprint_hash": "028383b21779f887b4485b6d12636a0f85c9702ebc751a092915ce71aa0d6ace"
+  },
+  {
+    "device_id": "DEV0321",
+    "fingerprint_hash": "19c98d16968688b36b2ba0414116092d1b934428165b5c03eef93c497c1eb253"
+  },
+  {
+    "device_id": "DEV0322",
+    "fingerprint_hash": "3baa622d9b536f0c8535c27b004a998cbfa43d5e084b030edd0ccf61d6324367"
+  },
+  {
+    "device_id": "DEV0323",
+    "fingerprint_hash": "1b957e3460a9c0b0b16976409b9899773d55316c876a4bfd54f94a83508b0c6a"
+  },
+  {
+    "device_id": "DEV0324",
+    "fingerprint_hash": "caf6aecc638f095a10c54ba7caec5ce0983d6f2807b9cbb33c0dd39631bbbfe4"
+  },
+  {
+    "device_id": "DEV0325",
+    "fingerprint_hash": "5afb7482e00a7ccc0115e20764dfe23888c834603f16e10c7995e1982b03297a"
+  },
+  {
+    "device_id": "DEV0326",
+    "fingerprint_hash": "1e0c6d91bfd58cf58a033a4b496f07ddc43a2485db53c97f53da65dacd38da7e"
+  },
+  {
+    "device_id": "DEV0327",
+    "fingerprint_hash": "155c3e55d9379fd59e4b8036799da146f8da9f5d959b86a2f597a081e7f99e5b"
+  },
+  {
+    "device_id": "DEV0328",
+    "fingerprint_hash": "963976890db37138d5c74a2010d971f7046a6c6ec4bd83e8e283196a317d13a0"
+  },
+  {
+    "device_id": "DEV0329",
+    "fingerprint_hash": "88c2e53f1ac8d07c4a6d84b17fe40c14cf01d3170fe352d11f35d2715c1e38d6"
+  },
+  {
+    "device_id": "DEV0330",
+    "fingerprint_hash": "4104f7c61aed1b5485f9a790d48f9f744f247d611e2a97726c70c59e71412c40"
+  },
+  {
+    "device_id": "DEV0331",
+    "fingerprint_hash": "3750f10aa2183a663d5bc0dc604a8c7957160f24f6cb759ed3dc119ed76c2ded"
+  },
+  {
+    "device_id": "DEV0332",
+    "fingerprint_hash": "c0287eec8dc85a43611e5eab33179dd0398c8f574c9808a2aca1c623ba730678"
+  },
+  {
+    "device_id": "DEV0333",
+    "fingerprint_hash": "8edcdd8c4b2b95f12224527ae453deda559c28d6bc4bbe3647779be590358f81"
+  },
+  {
+    "device_id": "DEV0334",
+    "fingerprint_hash": "1eb39e1659e45b94fb5edfdb90a251373ffb9cac81a92342e6d19c981beda9d6"
+  },
+  {
+    "device_id": "DEV0335",
+    "fingerprint_hash": "543b663a97bdf6de9c0073710ccb88c24633868e19568f0d5b00c1fb651d5302"
+  },
+  {
+    "device_id": "DEV0336",
+    "fingerprint_hash": "9109c0740248b973f32acb636cb0ec6945005a430ec648f1e95d134914934ce6"
+  },
+  {
+    "device_id": "DEV0337",
+    "fingerprint_hash": "43d63e5a4fd6b18db87da014f647792e182fd842f559cf3e8354363a308c18c3"
+  },
+  {
+    "device_id": "DEV0338",
+    "fingerprint_hash": "f204f3a2e9ecf217124ab56cce15a342946f9a6e3f9ddcfd84fa4dfbbc7c7973"
+  },
+  {
+    "device_id": "DEV0339",
+    "fingerprint_hash": "3e31c4e8b0041b6ea34b3e7617c5a015ac2be9c1a4fab3d318bf419a038e1f9c"
+  },
+  {
+    "device_id": "DEV0340",
+    "fingerprint_hash": "94e61c793b36386518cb7356fe45f9c9b7289d5b146758a0ff7bb2dba5b6b7d2"
+  },
+  {
+    "device_id": "DEV0341",
+    "fingerprint_hash": "43a3a440fe26b7a6e51745b8d7f769b4ed1ca2e93db963386002553adf6b1404"
+  },
+  {
+    "device_id": "DEV0342",
+    "fingerprint_hash": "84902f9d90d5def0a07bb72b4dfaf382ce3febd20ab64afdf36ec77a7ad4fe2f"
+  },
+  {
+    "device_id": "DEV0343",
+    "fingerprint_hash": "aa98a6ceff0130f6b05f87ee4804e9fcf4d6bb66973cede29e53da669094e929"
+  },
+  {
+    "device_id": "DEV0344",
+    "fingerprint_hash": "2b71471008456587b4f61190450ae6e158e1c0cec8dc3da656624c8a66e34321"
+  },
+  {
+    "device_id": "DEV0345",
+    "fingerprint_hash": "78b42c90293a0b12c938260e5ff8fc4ffdfcd6ed12d8ef73f2454265cd23374a"
+  },
+  {
+    "device_id": "DEV0346",
+    "fingerprint_hash": "fc6fc908f8756c581a80b56f98724f82b59beb5ec8a2258557c76a84d35ff415"
+  },
+  {
+    "device_id": "DEV0347",
+    "fingerprint_hash": "4a458da7e90ff787366e0b9aa81ed3cca1e0706320a9e8067318e5636afb8483"
+  },
+  {
+    "device_id": "DEV0348",
+    "fingerprint_hash": "c4cb77dbe60964f26481b11b6d2ff21f1a7cb547ea2ed4eb0f76296cff6c06a8"
+  },
+  {
+    "device_id": "DEV0349",
+    "fingerprint_hash": "e2f8acc7dfafdd1156eb71a5fa39604cf45239acf998eefed3f80c97691f51aa"
+  },
+  {
+    "device_id": "DEV0350",
+    "fingerprint_hash": "5450264019804e6b8ad7817ebb005336cdd1dc8579328461bc4c31ee71cbda0c"
+  },
+  {
+    "device_id": "DEV0351",
+    "fingerprint_hash": "f4e0e8bb056c2357822015c06b3e578f46fe71fd21300b008ab75910359977f9"
+  },
+  {
+    "device_id": "DEV0352",
+    "fingerprint_hash": "40d5cb885d8096b28dbcf5663b4653f6cda18d778ccdcc2249be3ff8ad80e774"
+  },
+  {
+    "device_id": "DEV0353",
+    "fingerprint_hash": "d2872fead2dee3cfbb27636b8ff43a5d680cf79b45c8a86ba9f29ed42d3392f5"
+  },
+  {
+    "device_id": "DEV0354",
+    "fingerprint_hash": "7fa0626b8a6cd632715ea08cbdf7219b87180ae2b0aa5801e7463fa8a357264b"
+  },
+  {
+    "device_id": "DEV0355",
+    "fingerprint_hash": "a67a03b4a7448547ecc7c80f351da5c33ab2579b89a18dd2edf087693bb07660"
+  },
+  {
+    "device_id": "DEV0356",
+    "fingerprint_hash": "2339f6d92bdc987ce2c6b5c8227d45e955f3ff4b432af62a93f523581fadb5b4"
+  },
+  {
+    "device_id": "DEV0357",
+    "fingerprint_hash": "bff83bc52c1e1e2bc655326ba138a45d2938589c687b6e0477ffbd34648348ae"
+  },
+  {
+    "device_id": "DEV0358",
+    "fingerprint_hash": "5845714fbdfb94d51c92cf6cb50ba23d031156e18169bb9a4f587a070daddc3a"
+  },
+  {
+    "device_id": "DEV0359",
+    "fingerprint_hash": "35baa8fbd894f9a68cf3f1b62b1b00fde15af33f5c68f67ae143659ec5791ec8"
+  },
+  {
+    "device_id": "DEV0360",
+    "fingerprint_hash": "4e2c02768131d5c5b5838b6db55d3bd9ddc0baaf38a5d1d39b80cd5d5298b1c6"
+  },
+  {
+    "device_id": "DEV0361",
+    "fingerprint_hash": "7ac2178859e4e6b8e302859df44c3f87124d057cf2356824dffc1b2f1c4f46a0"
+  },
+  {
+    "device_id": "DEV0362",
+    "fingerprint_hash": "7332b2d9887fd9662b3ef0aaba5f91c5e4f92a118f89fe33a22db102b3b064a1"
+  },
+  {
+    "device_id": "DEV0363",
+    "fingerprint_hash": "338b4dac3a52a78e82ed85cb58ed2fa7964028414e6ce59d32cb829f1c642b63"
+  },
+  {
+    "device_id": "DEV0364",
+    "fingerprint_hash": "af9ecf8912bdf2f7d3f5972738bfd7505f4a12fd96be6b51c14b276e711ef73b"
+  },
+  {
+    "device_id": "DEV0365",
+    "fingerprint_hash": "70f63e12b2e021c53fa7344408f82941f8dd75e797997da5f1d45b2c735a1ab2"
+  },
+  {
+    "device_id": "DEV0366",
+    "fingerprint_hash": "7e3414abd8b450355cce3b705aedc419b26e80110536ba090aa4c22807e97f1b"
+  },
+  {
+    "device_id": "DEV0367",
+    "fingerprint_hash": "db98faa388e0f26d2c3ed83ce659b9a9a1913097e87fca0e6bc05798a8e00732"
+  },
+  {
+    "device_id": "DEV0368",
+    "fingerprint_hash": "589ef556ce6deaa39c5736e143c59902e513bafd01dbaac8858a78bb868e3bf1"
+  },
+  {
+    "device_id": "DEV0369",
+    "fingerprint_hash": "c8347ad36369f57975e1ab54654b91b0be097f8f43171fd4b3955af92adfc52a"
+  },
+  {
+    "device_id": "DEV0370",
+    "fingerprint_hash": "3e8717ef1527a4d24e349e35bd0d417368f5f0eed978c059407040681c6d2049"
+  },
+  {
+    "device_id": "DEV0371",
+    "fingerprint_hash": "77af1475dd4f73383fd8b3994f50aee8be397af27e6336ea48d2d710892a4191"
+  },
+  {
+    "device_id": "DEV0372",
+    "fingerprint_hash": "249281fdc14979308cf7f9dfd9e0c0a4d8e645d0662f62c7b45a9862fa872307"
+  },
+  {
+    "device_id": "DEV0373",
+    "fingerprint_hash": "65c5181a99c0005b407171c48cc3f66972194b1e836ad6be9099808e57339cfe"
+  },
+  {
+    "device_id": "DEV0374",
+    "fingerprint_hash": "de67bcdbb80565e48324d8cb0375ce39a5c5321ed69793c2afb50abd92777d1d"
+  },
+  {
+    "device_id": "DEV0375",
+    "fingerprint_hash": "7746acef567ef899529895a1ced139997c755f86eb736a2794da1606be9e4f19"
+  },
+  {
+    "device_id": "DEV0376",
+    "fingerprint_hash": "ca101396c98d27c3a39f8b1eee04cc6be96b633379ee94d47e83ba9c0b2b4b50"
+  },
+  {
+    "device_id": "DEV0377",
+    "fingerprint_hash": "041eb73d9985dade88c032991668490af1423d5b5e7e5fa17527837bd325cb3d"
+  },
+  {
+    "device_id": "DEV0378",
+    "fingerprint_hash": "c4369fe5dbf95cce3a08997f58c30fa7907756719be669419968537de9c25034"
+  },
+  {
+    "device_id": "DEV0379",
+    "fingerprint_hash": "33aa44929db71ba8291c939b6c51714ae6b2fe8a8fbbd70edd0fdaa7f67f950e"
+  },
+  {
+    "device_id": "DEV0380",
+    "fingerprint_hash": "73b65a89e55a261792a5cc2b48a3efee649887dfcf1a5c64816b2d053b43b668"
+  },
+  {
+    "device_id": "DEV0381",
+    "fingerprint_hash": "d4cd3515bc579cd7e2e816214a71771dae33c8a15793fa3423eef304ffc0a390"
+  },
+  {
+    "device_id": "DEV0382",
+    "fingerprint_hash": "591f6f75b2cb13bf674ffec239d8948b291ce59b893594cc0e98c9e6899628e8"
+  },
+  {
+    "device_id": "DEV0383",
+    "fingerprint_hash": "03f4313de40f90ade2782f372440f368f9c324ba7c897716da633a3d6654a01b"
+  },
+  {
+    "device_id": "DEV0384",
+    "fingerprint_hash": "5bade8568d9b9a6a3d05b203c71320c939669d4d4edad8370c9047b6b33a22a3"
+  },
+  {
+    "device_id": "DEV0385",
+    "fingerprint_hash": "de3270a04c5314e5891f4f25580832255a266629e053d60212b4dd1eaf688805"
+  },
+  {
+    "device_id": "DEV0386",
+    "fingerprint_hash": "413e20233009dd16b8386706fb59418f0d9ac9c5b0ebf950cf267d07f8c90262"
+  },
+  {
+    "device_id": "DEV0387",
+    "fingerprint_hash": "8b92abd3377c7807cdca4b7d0e3cac6334e70c59000f4d58734ff1dab8e3f6bc"
+  },
+  {
+    "device_id": "DEV0388",
+    "fingerprint_hash": "10a98c8d2d202659a66009098367be3029eeb4b2863326e13105d4bebd3f85dd"
+  },
+  {
+    "device_id": "DEV0389",
+    "fingerprint_hash": "8b381105a40967c66f6fc5a1116c1547cccb7ce096426bd98e7133b2bbd74d98"
+  },
+  {
+    "device_id": "DEV0390",
+    "fingerprint_hash": "bc050715f180192af101f3b2bbeee265a4389caa222ba4d12492c30071b3a7d7"
+  },
+  {
+    "device_id": "DEV0391",
+    "fingerprint_hash": "eb804ba0eb6894c046b9ebf127509962445ca99e7f1c1fc3b428c9878522c229"
+  },
+  {
+    "device_id": "DEV0392",
+    "fingerprint_hash": "b2c34f0585853455e1a407a438063d959eda2a7798235bbb98250fd3b01975f9"
+  },
+  {
+    "device_id": "DEV0393",
+    "fingerprint_hash": "6a25c786986c447728e40018ec6cddd7d175592b74d9b73b224a1d3b9aa386d9"
+  },
+  {
+    "device_id": "DEV0394",
+    "fingerprint_hash": "37df7aca885d8db022aae3ac94a852453f36349b4422ea16ace306ad322a4f07"
+  },
+  {
+    "device_id": "DEV0395",
+    "fingerprint_hash": "7526569717ebb9132f7304460c53ab0103dc97be83724f75fd34bf020d0914d2"
+  },
+  {
+    "device_id": "DEV0396",
+    "fingerprint_hash": "695c18166bbb89e295302f4fb2456ed01797f322eed553e104f16177105d5879"
+  },
+  {
+    "device_id": "DEV0397",
+    "fingerprint_hash": "59e61fd370f004972f61f8b04291f049181f6d8c6abff2129fb178014569a2b8"
+  },
+  {
+    "device_id": "DEV0398",
+    "fingerprint_hash": "7f4a565f82d58b522a0c1a3f69ff63d43fa623d5a7e51b1de9b0a88daaf4c88d"
+  },
+  {
+    "device_id": "DEV0399",
+    "fingerprint_hash": "f9e9fb1405f2be65698e4dee8f93ef47044f0b62036162203cdce8779057a0fa"
+  },
+  {
+    "device_id": "DEV0400",
+    "fingerprint_hash": "bafc6f592c73bd6b395fcb2c13307d2052db7ba2db2002cf2c070503db0acc3c"
+  },
+  {
+    "device_id": "DEV0401",
+    "fingerprint_hash": "83c011bac645616b7a37e03d837e8b414927d82301adecb82d65dd8357238c41"
+  },
+  {
+    "device_id": "DEV0402",
+    "fingerprint_hash": "3301232dc5d5e575c4ba913a642a50643788c5879005bc4ac4186ac4c923c0f9"
+  },
+  {
+    "device_id": "DEV0403",
+    "fingerprint_hash": "80cd90df12a54dc42a90bb732c572ce24474c839db1a648aaa677645686a6efe"
+  },
+  {
+    "device_id": "DEV0404",
+    "fingerprint_hash": "7f958138868e5dc8bbcc5200af59d09008de909709b51cffdf38d36610be358c"
+  },
+  {
+    "device_id": "DEV0405",
+    "fingerprint_hash": "61d5a1c86a53a9bcd943fd3ae9c3a0ff8d818e7d8d9b0d9541080a1a10c0db51"
+  }
+]

data/synthetic/ground_truth.json

@@ -0,0 +1,165 @@
+{
+  "rings": [
+    {
+      "ring_id": "RING01",
+      "customer_ids": [
+        "CUST0001",
+        "CUST0002",
+        "CUST0003"
+      ],
+      "type": "fraud"
+    },
+    {
+      "ring_id": "RING02",
+      "customer_ids": [
+        "CUST0004",
+        "CUST0005",
+        "CUST0006",
+        "CUST0007",
+        "CUST0008"
+      ],
+      "type": "fraud"
+    },
+    {
+      "ring_id": "RING03",
+      "customer_ids": [
+        "CUST0009",
+        "CUST0010",
+        "CUST0011",
+        "CUST0012",
+        "CUST0013"
+      ],
+      "type": "fraud"
+    },
+    {
+      "ring_id": "RING04",
+      "customer_ids": [
+        "CUST0014",
+        "CUST0015",
+        "CUST0016",
+        "CUST0017",
+        "CUST0018"
+      ],
+      "type": "fraud"
+    },
+    {
+      "ring_id": "RING05",
+      "customer_ids": [
+        "CUST0019",
+        "CUST0020",
+        "CUST0021",
+        "CUST0022",
+        "CUST0023"
+      ],
+      "type": "fraud"
+    },
+    {
+      "ring_id": "RING06",
+      "customer_ids": [
+        "CUST0024",
+        "CUST0025",
+        "CUST0026",
+        "CUST0027",
+        "CUST0028",
+        "CUST0029"
+      ],
+      "type": "fraud"
+    },
+    {
+      "ring_id": "RING07",
+      "customer_ids": [
+        "CUST0030",
+        "CUST0031",
+        "CUST0032"
+      ],
+      "type": "fraud"
+    },
+    {
+      "ring_id": "RING08",
+      "customer_ids": [
+        "CUST0033",
+        "CUST0034",
+        "CUST0035"
+      ],
+      "type": "fraud"
+    },
+    {
+      "ring_id": "RING09",
+      "customer_ids": [
+        "CUST0036",
+        "CUST0037",
+        "CUST0038",
+        "CUST0039",
+        "CUST0040"
+      ],
+      "type": "fraud"
+    },
+    {
+      "ring_id": "RING10",
+      "customer_ids": [
+        "CUST0041",
+        "CUST0042",
+        "CUST0043",
+        "CUST0044"
+      ],
+      "type": "fraud"
+    },
+    {
+      "ring_id": "RING11",
+      "customer_ids": [
+        "CUST0501",
+        "CUST0502",
+        "CUST0503"
+      ],
+      "type": "fraud"
+    }
+  ],
+  "stress_clusters": [
+    {
+      "cluster_id": "STRESS01",
+      "customer_ids": [
+        "CUST0045",
+        "CUST0046",
+        "CUST0047",
+        "CUST0048"
+      ],
+      "type": "legitimate_shared_address"
+    },
+    {
+      "cluster_id": "STRESS02",
+      "customer_ids": [
+        "CUST0049",
+        "CUST0050",
+        "CUST0051",
+        "CUST0052"
+      ],
+      "type": "legitimate_shared_address"
+    },
+    {
+      "cluster_id": "STRESS03",
+      "customer_ids": [
+        "CUST0053",
+        "CUST0054",
+        "CUST0055",
+        "CUST0056"
+      ],
+      "type": "legitimate_shared_address"
+    },
+    {
+      "cluster_id": "STRESS04",
+      "customer_ids": [
+        "CUST0057",
+        "CUST0058"
+      ],
+      "type": "legitimate_shared_address"
+    },
+    {
+      "cluster_id": "STRESS05",
+      "customer_ids": [
+        "CUST0059",
+        "CUST0060"
+      ],
+      "type": "legitimate_shared_address"
+    }
+  ]
+}

data/synthetic/payment_instruments.json

@@ -0,0 +1,1932 @@
+[
+  {
+    "instrument_id": "PAY0001",
+    "type": "upi",
+    "last4_hash": "43854d169fe9af1d004773b3c9822b3f57b140d4703096704f3d795b1581706b"
+  },
+  {
+    "instrument_id": "PAY0002",
+    "type": "card",
+    "last4_hash": "9b3cea3bd8789ba47cc969bff5914da7ebac8e0052e282917299814a53a7dfdb"
+  },
+  {
+    "instrument_id": "PAY0003",
+    "type": "upi",
+    "last4_hash": "72cc443db576409f0e56de30286baede30ae7c2d0f6e3e49f3fc5ac7e9d5b276"
+  },
+  {
+    "instrument_id": "PAY0004",
+    "type": "upi",
+    "last4_hash": "b324329805a238135ed3f429bde8cbe869b2b2190e17edccd3ae9e036565a7ae"
+  },
+  {
+    "instrument_id": "PAY0005",
+    "type": "upi",
+    "last4_hash": "ba2854ea732fe9c5c45bb5532172ea531a8510cd03fde9247f0d1f38368e49f1"
+  },
+  {
+    "instrument_id": "PAY0006",
+    "type": "upi",
+    "last4_hash": "1314e83b4d4e34a1bcdb5665aedc31616d49cef471cca9561c471abd798b0f5a"
+  },
+  {
+    "instrument_id": "PAY0007",
+    "type": "card",
+    "last4_hash": "b39b7ba7e07d710447e89c40277171aa4c568e3de08e4970d7579288af5bf3f4"
+  },
+  {
+    "instrument_id": "PAY0008",
+    "type": "card",
+    "last4_hash": "9199d9e5b275563cd8c867aa6013d32117ad6232cd0d6e0b7e28b9eb54c80b2c"
+  },
+  {
+    "instrument_id": "PAY0009",
+    "type": "card",
+    "last4_hash": "48ccfd09d04c36762f6b0c83bf32f695caa2f2dcf773703acdb4a030a3fa1dc0"
+  },
+  {
+    "instrument_id": "PAY0010",
+    "type": "card",
+    "last4_hash": "c5067bcbc573d35e3788d098a8d58772fc5717e120b4990f8bcbfd634e45853a"
+  },
+  {
+    "instrument_id": "PAY0011",
+    "type": "card",
+    "last4_hash": "39e71545d840848a9af1a669d7e6aabf8ee95536a2f219820c450aa31f796573"
+  },
+  {
+    "instrument_id": "PAY0012",
+    "type": "upi",
+    "last4_hash": "fdb37acb269dac7997162fd98b93c16bc8bdaa2906747082415d809cf1df6be5"
+  },
+  {
+    "instrument_id": "PAY0013",
+    "type": "card",
+    "last4_hash": "44544e0d6770664c4c13ae2ed2438373c2014f112a6f5fcd7c241ba79222bc16"
+  },
+  {
+    "instrument_id": "PAY0014",
+    "type": "upi",
+    "last4_hash": "4765a7b39436675afd5fdff5b56b41db91c05ecb524a2a7def0318b3d5de307c"
+  },
+  {
+    "instrument_id": "PAY0015",
+    "type": "card",
+    "last4_hash": "ed46f81f54d1ac6c2586e0003e659bc0595f7ebfd6fa37dd5b354c56451d4ab2"
+  },
+  {
+    "instrument_id": "PAY0016",
+    "type": "upi",
+    "last4_hash": "2d56bccc2c06064b1cc7bd9a8605692e6c708145159d69b6439d53d5594386b1"
+  },
+  {
+    "instrument_id": "PAY0017",
+    "type": "upi",
+    "last4_hash": "29e3061bd659a3a4b6ac671fd1dc8f7ff9f453aa6caaf15289b9a722abd03b7c"
+  },
+  {
+    "instrument_id": "PAY0018",
+    "type": "upi",
+    "last4_hash": "b2ce2a829d3d50ba8e38eac676598731979f3a8b019e4f247fd85828642ee9f8"
+  },
+  {
+    "instrument_id": "PAY0019",
+    "type": "upi",
+    "last4_hash": "0941be7a3eff242affd46525c1c0468acef945d064fe8c0b553d14e0074303a2"
+  },
+  {
+    "instrument_id": "PAY0020",
+    "type": "upi",
+    "last4_hash": "3769b2ecd28cdf0ed1d25aad16979ce2a3ec7782017c61941e8b161c44baba08"
+  },
+  {
+    "instrument_id": "PAY0021",
+    "type": "card",
+    "last4_hash": "ab86fe44dea15e7a6b438f741aa208f079d55f626f5c471c999b2ed78b37bd6a"
+  },
+  {
+    "instrument_id": "PAY0022",
+    "type": "upi",
+    "last4_hash": "656ec82405d95b39a019289db15f781b4f8be7824060ab938656199fce05961f"
+  },
+  {
+    "instrument_id": "PAY0023",
+    "type": "upi",
+    "last4_hash": "cf5b255ef873089b5efa6efa80f98b50cb5df021246f561771e50fa90ff34c78"
+  },
+  {
+    "instrument_id": "PAY0024",
+    "type": "card",
+    "last4_hash": "faede57cfab74b18e63014bf3295fde357982f05d67e6dac2a0b28b333b5c396"
+  },
+  {
+    "instrument_id": "PAY0025",
+    "type": "upi",
+    "last4_hash": "41e20f4fbbd493f6ee8744ec2ac8143aba5d3cb34eec0338a051e460fed85707"
+  },
+  {
+    "instrument_id": "PAY0026",
+    "type": "card",
+    "last4_hash": "7808b86edb7b1acd97860de5b0014865ec15ea7a2f6e8ba5c0a3c88127a9447b"
+  },
+  {
+    "instrument_id": "PAY0027",
+    "type": "upi",
+    "last4_hash": "ff293a64aa18c6df0364f0a6bea775dcb81d5d5cc4fce958612ecd70eeb64aee"
+  },
+  {
+    "instrument_id": "PAY0028",
+    "type": "card",
+    "last4_hash": "27995993191887fe707d8f4be3a3b3e8190b12dc0ba20cc441ded243f8a304a4"
+  },
+  {
+    "instrument_id": "PAY0029",
+    "type": "upi",
+    "last4_hash": "d8f34a8a519c35fd11b63a97cb68ab840d9993ea4e09f25ba4cbf20fddade301"
+  },
+  {
+    "instrument_id": "PAY0030",
+    "type": "upi",
+    "last4_hash": "ab8ba612699a1b34ab8c4398b833d221f4bfe4e43c188f8c571de33225790419"
+  },
+  {
+    "instrument_id": "PAY0031",
+    "type": "upi",
+    "last4_hash": "382a90367a19bf8ef8631798e4d29596d729604fe59d6c6d439702d5ffe87970"
+  },
+  {
+    "instrument_id": "PAY0032",
+    "type": "upi",
+    "last4_hash": "be53a1caecfd70a4ca27be2289e8c54be3f6af7f9bf67ef45af4ff157e63bf5f"
+  },
+  {
+    "instrument_id": "PAY0033",
+    "type": "card",
+    "last4_hash": "b23d8b6a24247560231f60bd114e288d0ac8707610ff3c53bf82897eace08854"
+  },
+  {
+    "instrument_id": "PAY0034",
+    "type": "upi",
+    "last4_hash": "be08d86973638e7039b4265f4cc42328e0a99933baf066ed7c252a2594c25282"
+  },
+  {
+    "instrument_id": "PAY0035",
+    "type": "card",
+    "last4_hash": "c81b352b270b2e498256b5e24c726a039cd665477a83d71d95633340782782d4"
+  },
+  {
+    "instrument_id": "PAY0036",
+    "type": "upi",
+    "last4_hash": "9881a093db6601fd80ff7825844a119d17b55cde13322cc5b3f8e57a28d9c701"
+  },
+  {
+    "instrument_id": "PAY0037",
+    "type": "upi",
+    "last4_hash": "4537f0f8c9dc0d35adf0338d454e2d324bcd289713b434c7d29708c06a3b5dd7"
+  },
+  {
+    "instrument_id": "PAY0038",
+    "type": "card",
+    "last4_hash": "39d5ce6249548c318a5f0b6bf0a3e2f234750a9741b6ce2f9eff8670bc7cd7af"
+  },
+  {
+    "instrument_id": "PAY0039",
+    "type": "upi",
+    "last4_hash": "9f7599e78fd8ce9803dd88d94f75372f3d02c861b7d4b8b6a14144a3eb7d15d0"
+  },
+  {
+    "instrument_id": "PAY0040",
+    "type": "card",
+    "last4_hash": "cf7dee3443a859943d9ecf558fe5cf260f9acb4150b5d18d8b2e04be1cc3ca54"
+  },
+  {
+    "instrument_id": "PAY0041",
+    "type": "card",
+    "last4_hash": "55fdec963805de594b61b2c1692cadc2c1dfb844d6ac10c5bfed33c842087b2e"
+  },
+  {
+    "instrument_id": "PAY0042",
+    "type": "card",
+    "last4_hash": "03afa3478417ff5086b7560e07ccafaeb73f2772ce4e05fecde39fcd32a5437e"
+  },
+  {
+    "instrument_id": "PAY0043",
+    "type": "upi",
+    "last4_hash": "d934819258879e09373979e6d1fb0d5da3404039738bd5d806df0eca450a3429"
+  },
+  {
+    "instrument_id": "PAY0044",
+    "type": "card",
+    "last4_hash": "48f84ea6043f5313382811d79ec2ba3d163c943b0e33ca3fe2ba2e9a4f18e935"
+  },
+  {
+    "instrument_id": "PAY0045",
+    "type": "upi",
+    "last4_hash": "37a5eba54aa89b51ef3f6bb7615f043ac56587d4d67e10fb85e1ec42495a41ab"
+  },
+  {
+    "instrument_id": "PAY0046",
+    "type": "card",
+    "last4_hash": "a11f793d5f24841552b2befc8d946d9f7e5652edfb58bd49c3a8423a6031c952"
+  },
+  {
+    "instrument_id": "PAY0047",
+    "type": "upi",
+    "last4_hash": "140a85d1fd4c514a914643ec2922ce6818ae675e73e2eef6e641fee467438ed2"
+  },
+  {
+    "instrument_id": "PAY0048",
+    "type": "upi",
+    "last4_hash": "4b734774fc6f30af3315238e013de6354eeff142ce27efdbc72705e3a13f3d5f"
+  },
+  {
+    "instrument_id": "PAY0049",
+    "type": "card",
+    "last4_hash": "0bb96d87216f5159d997d566969931579a74f4fcd81e9e4280e5354a4ba0c2fd"
+  },
+  {
+    "instrument_id": "PAY0050",
+    "type": "card",
+    "last4_hash": "5e27752e101ec179e101f6016d7b015f2a408795a3135953ef6066f73753fadd"
+  },
+  {
+    "instrument_id": "PAY0051",
+    "type": "upi",
+    "last4_hash": "89c55097f7602585a95ada5ee3a306ecb4966744a2586946b15e978fb46f6832"
+  },
+  {
+    "instrument_id": "PAY0052",
+    "type": "card",
+    "last4_hash": "741f514bbddf7310b4d0fe0dea7f062941dd67b8090b66d2bcf6bc621bbd2021"
+  },
+  {
+    "instrument_id": "PAY0053",
+    "type": "upi",
+    "last4_hash": "eaf52ea64d24cb70e3cb3e2790c8cead55c6bd523b0b0de6cd0dbad4cae51c6e"
+  },
+  {
+    "instrument_id": "PAY0054",
+    "type": "card",
+    "last4_hash": "cc55b4c2e7e6d5ef1642141e8a25944cfa6d7ed2b8e8385adcc92c6bb3aee571"
+  },
+  {
+    "instrument_id": "PAY0055",
+    "type": "card",
+    "last4_hash": "1f5e12a1dcc43dfd45e6709479e6cc62b1a6426a6ff78926166808e13a813c9d"
+  },
+  {
+    "instrument_id": "PAY0056",
+    "type": "card",
+    "last4_hash": "900cb8e36b717b81b9dbbb51cfae997cf5236c36afea5ba7100d005ad52d1cbc"
+  },
+  {
+    "instrument_id": "PAY0057",
+    "type": "upi",
+    "last4_hash": "a6781ca9dc05eb6d4a0cd863aa22a558554373cfcd2be1242f8df005f98416d6"
+  },
+  {
+    "instrument_id": "PAY0058",
+    "type": "upi",
+    "last4_hash": "3a14d640f42889d144e71bda4425b340e0b1f62d87f6d9bffd464598655d43c1"
+  },
+  {
+    "instrument_id": "PAY0059",
+    "type": "upi",
+    "last4_hash": "cec0db12ddf9f2d2e372eca85188235f332465f82e371202f3950d0b77fb5b6c"
+  },
+  {
+    "instrument_id": "PAY0060",
+    "type": "upi",
+    "last4_hash": "c779fc361251ad4c3db1fd8655b96bfd565eff7f13a414ace30eda2f89c9cd45"
+  },
+  {
+    "instrument_id": "PAY0061",
+    "type": "card",
+    "last4_hash": "07007ba87ea78903e0ec57160cac3ff7dfc1109f2d84cc6eed9ffbd8358a62d1"
+  },
+  {
+    "instrument_id": "PAY0062",
+    "type": "upi",
+    "last4_hash": "f1980fc5343d10c28baa1ce65de484c08733c23d7e43977d53aec202775ca81e"
+  },
+  {
+    "instrument_id": "PAY0063",
+    "type": "upi",
+    "last4_hash": "4ef51169834cb29778aa564dd2dd9ebc6017656e2d47f5a21c689a44798c1605"
+  },
+  {
+    "instrument_id": "PAY0064",
+    "type": "card",
+    "last4_hash": "9302a967cce443780477212553ae3d74625dd165a096a4ee8299dcf0c8079866"
+  },
+  {
+    "instrument_id": "PAY0065",
+    "type": "upi",
+    "last4_hash": "86dfdb71e782a8cec375d222d5383cc95e042aad436f9fa3a47f1d7c8c3864f1"
+  },
+  {
+    "instrument_id": "PAY0066",
+    "type": "card",
+    "last4_hash": "ef3eaba74dcc1c3e6536e12d4a95d94fe3d32b29f1d217ce6b377049efe60511"
+  },
+  {
+    "instrument_id": "PAY0067",
+    "type": "upi",
+    "last4_hash": "7f58b9425094d6f95c8f177eb93b12c1701a63978f9bb291335345d852425c90"
+  },
+  {
+    "instrument_id": "PAY0068",
+    "type": "card",
+    "last4_hash": "45b120bb08854e18123188f37c3d0b52d06781d415fb5b56a684870a779e09bc"
+  },
+  {
+    "instrument_id": "PAY0069",
+    "type": "card",
+    "last4_hash": "43a70ba6b21894cd24165d9d7ee59b012e698cf97e22fd275ccdd05a556e7061"
+  },
+  {
+    "instrument_id": "PAY0070",
+    "type": "upi",
+    "last4_hash": "22e8922403c6ca57a6a7eb282337498418fa5f89ac9110534278aef3736aa28c"
+  },
+  {
+    "instrument_id": "PAY0071",
+    "type": "upi",
+    "last4_hash": "7a9e521df4c6ba4d71bb4bad29f1f771918ffa0b507d9a6381f642b6dfa4f6d3"
+  },
+  {
+    "instrument_id": "PAY0072",
+    "type": "card",
+    "last4_hash": "f35bdfbc3b7e089ee8a84c2684deb1f5c96732895d79086eb8fa0458742a469d"
+  },
+  {
+    "instrument_id": "PAY0073",
+    "type": "card",
+    "last4_hash": "e26f3716adbe1a42e8b784b9d6670f922151428235cd146c3cc7b80653b2f9d7"
+  },
+  {
+    "instrument_id": "PAY0074",
+    "type": "upi",
+    "last4_hash": "dff532682c09d2c12af80c41a395f9b853e8edae6ae0bb129fd9dc7274d52b63"
+  },
+  {
+    "instrument_id": "PAY0075",
+    "type": "card",
+    "last4_hash": "4817ddb54d7a527a5cb1e069434fdfb5350bfdc910cc6a3aa35b54a4036b9809"
+  },
+  {
+    "instrument_id": "PAY0076",
+    "type": "card",
+    "last4_hash": "58e47bf2ec5f0d27dab6b42787a65ea8cb3a2c3c88816dbc2eee8bc20d600a1a"
+  },
+  {
+    "instrument_id": "PAY0077",
+    "type": "upi",
+    "last4_hash": "e7635b6f96d7470ea28f65a202cd9e0c7abf789ee12f1ef5b9b80fe5f1bae7d2"
+  },
+  {
+    "instrument_id": "PAY0078",
+    "type": "upi",
+    "last4_hash": "6692ae362de80b8769f3a4b190b025dfdf3ba7b184d2928d35c9c80061ff5348"
+  },
+  {
+    "instrument_id": "PAY0079",
+    "type": "upi",
+    "last4_hash": "399bd15402e3c3ea9a728f19361e02c634c9a46b40051212832d8817d830aa5e"
+  },
+  {
+    "instrument_id": "PAY0080",
+    "type": "upi",
+    "last4_hash": "3ee5cc0469c7e67513b98291b35b41cbd00f5c5e20973b783b2b3b44b225a81c"
+  },
+  {
+    "instrument_id": "PAY0081",
+    "type": "card",
+    "last4_hash": "e753559bf52404ed6877db546a80e189e22c6ff1ca965d7115a882d4335c9be7"
+  },
+  {
+    "instrument_id": "PAY0082",
+    "type": "upi",
+    "last4_hash": "ab3c34aaa228b75bdfa29314f3c134dacadede2deee1dfdb3fda46cf3628e178"
+  },
+  {
+    "instrument_id": "PAY0083",
+    "type": "upi",
+    "last4_hash": "d0fc42ab00f8e3e251e0418898559ee941b0de10f08f170bd11bf51757b270bd"
+  },
+  {
+    "instrument_id": "PAY0084",
+    "type": "card",
+    "last4_hash": "e18f50079034b56f8d86d0e2fcaf15a6685ef6b8771711ff7bda1f937f4b10f4"
+  },
+  {
+    "instrument_id": "PAY0085",
+    "type": "upi",
+    "last4_hash": "e8df2d2fe1fbb87bcf45505fb1dea35d09a702ef3af8625ebf5a39830c45d1a9"
+  },
+  {
+    "instrument_id": "PAY0086",
+    "type": "card",
+    "last4_hash": "05e5ddbcc59eb669a2e5987e36a40537fcdb275a87b38023f9da6d41f0a5c37b"
+  },
+  {
+    "instrument_id": "PAY0087",
+    "type": "upi",
+    "last4_hash": "a5613be89b56ff92cc0d12ca5b2900ff0a503c9a61bc731d9a0f3c00f7bdeaa0"
+  },
+  {
+    "instrument_id": "PAY0088",
+    "type": "upi",
+    "last4_hash": "1c313ae9b416b8c468f7079bca03b08c0273987a41024976176bd1992399a4fa"
+  },
+  {
+    "instrument_id": "PAY0089",
+    "type": "upi",
+    "last4_hash": "aa0363dfc3a149e6665dff0b628546f766d3ea2910ec17078c4968c97102a9a2"
+  },
+  {
+    "instrument_id": "PAY0090",
+    "type": "card",
+    "last4_hash": "35a6c93873ac10cb4386cd7e3b071daa8b3002025a30284d26b3d5a0950f7900"
+  },
+  {
+    "instrument_id": "PAY0091",
+    "type": "card",
+    "last4_hash": "d308ddc557e774187ad6c7da81b39db990def9a39e6c88e79cb163b6bdd6e8e5"
+  },
+  {
+    "instrument_id": "PAY0092",
+    "type": "upi",
+    "last4_hash": "f6a7b1da20dc935df8ad86e6d517983f6f08a9937ebedd49046d209630cf5307"
+  },
+  {
+    "instrument_id": "PAY0093",
+    "type": "upi",
+    "last4_hash": "48ba0d3f602388d95de0e7a95c21a34daa9438ed5eb6617619728482e2da6c30"
+  },
+  {
+    "instrument_id": "PAY0094",
+    "type": "card",
+    "last4_hash": "97d2244a7046f7586a09c23c5c270591dfc00dbadcbe8a279d045825bd669d29"
+  },
+  {
+    "instrument_id": "PAY0095",
+    "type": "upi",
+    "last4_hash": "8dc3b1593caf9cd87427a1ecebe50a2c9e8a52759900cfe3358f3a55c5400bcf"
+  },
+  {
+    "instrument_id": "PAY0096",
+    "type": "upi",
+    "last4_hash": "899c2819e671e9241ca393e13a3b08a70d6e6c0ecad7faf5cf22999e648577aa"
+  },
+  {
+    "instrument_id": "PAY0097",
+    "type": "card",
+    "last4_hash": "c9916ca5880d11008999dc266d47657e927fde740bb899caa52ca5528da796b2"
+  },
+  {
+    "instrument_id": "PAY0098",
+    "type": "card",
+    "last4_hash": "5f9e3d4a0fa6f44a726c146ead2a301eb4026ae67b06cc5971210e5d749762b0"
+  },
+  {
+    "instrument_id": "PAY0099",
+    "type": "card",
+    "last4_hash": "99eeb44d3313dca5755437fca4cb7f50ccb1e9f7279abd334045d05a001fd8b8"
+  },
+  {
+    "instrument_id": "PAY0100",
+    "type": "card",
+    "last4_hash": "ecf8103662b93ccc41bc42ba074379fedce3688c49093102b9200f0cec473559"
+  },
+  {
+    "instrument_id": "PAY0101",
+    "type": "upi",
+    "last4_hash": "624ea67219372de3ff819cad5c6feb061f1f719eda1f44f94d154bf758cfd97e"
+  },
+  {
+    "instrument_id": "PAY0102",
+    "type": "card",
+    "last4_hash": "8c8699214133b6332adc864763533eb8880c1e6f6a7787a962cd3f5ed3e72026"
+  },
+  {
+    "instrument_id": "PAY0103",
+    "type": "card",
+    "last4_hash": "2e1943902be93c87e9d31812ef887d1f91326e8e8e3779feb59c48c9ac32f58f"
+  },
+  {
+    "instrument_id": "PAY0104",
+    "type": "upi",
+    "last4_hash": "facf1b4df937f097837b3f532e2b5dae513e76a448557385a979faa12bebef01"
+  },
+  {
+    "instrument_id": "PAY0105",
+    "type": "upi",
+    "last4_hash": "36d4f0701702ed6e66256c81f32544275aec9155741661d2cdad2b83525b1129"
+  },
+  {
+    "instrument_id": "PAY0106",
+    "type": "upi",
+    "last4_hash": "b861be6fb870d16d203b8521a1805a974d4e0503c67120b44c4c9df5778f8f12"
+  },
+  {
+    "instrument_id": "PAY0107",
+    "type": "card",
+    "last4_hash": "fef8ef07b0461c95476dd7b995f3333578ae82031501d300354f2263cf1a73cb"
+  },
+  {
+    "instrument_id": "PAY0108",
+    "type": "upi",
+    "last4_hash": "1f8509908e8e7d16fe0c0de8ea7cc0fbadb2d3c24f831ab4ac7b35ee85f09f49"
+  },
+  {
+    "instrument_id": "PAY0109",
+    "type": "upi",
+    "last4_hash": "7baf56a8601326f4d3b8fb1a73b013a596f37175d60075a2ba2d0cd45735ad2f"
+  },
+  {
+    "instrument_id": "PAY0110",
+    "type": "upi",
+    "last4_hash": "09509b09bdfa083ce661063988b5fd1fb9cb92a37c341a67355eecd6c6a778dc"
+  },
+  {
+    "instrument_id": "PAY0111",
+    "type": "upi",
+    "last4_hash": "ab9a89784cfae1151df7fa3057f7c20b90374c0b338fefabda9fb5954db70060"
+  },
+  {
+    "instrument_id": "PAY0112",
+    "type": "upi",
+    "last4_hash": "ba7b98b0c987d50c24b7851e96ba961b0f6561ba32e669918f2203ac26035829"
+  },
+  {
+    "instrument_id": "PAY0113",
+    "type": "upi",
+    "last4_hash": "b0e75b9a16227b9d288a465d0de38135405a7e4f0529545d7e26a4fa4c366728"
+  },
+  {
+    "instrument_id": "PAY0114",
+    "type": "card",
+    "last4_hash": "e68259a977d847c9ed26fc47c6855021c9a51eac28a63f9a0c012f7bbac0b556"
+  },
+  {
+    "instrument_id": "PAY0115",
+    "type": "card",
+    "last4_hash": "b789f4e82d98e206cf0f6cc237eff16955db29bee865372ac12af8e0226b7607"
+  },
+  {
+    "instrument_id": "PAY0116",
+    "type": "card",
+    "last4_hash": "b8b2994c7b8b89680041f24fbb0d10dc76e072cbefb1a342fb3d7cf4b002bec6"
+  },
+  {
+    "instrument_id": "PAY0117",
+    "type": "card",
+    "last4_hash": "70e8d52a9b4616e728112408ceca7abd9507e93f30e958ecb09f6ea607d99d06"
+  },
+  {
+    "instrument_id": "PAY0118",
+    "type": "card",
+    "last4_hash": "5b33003a928495b97792ac286d477b54dd20eb773c74ae2fb3653bc5950ad6dd"
+  },
+  {
+    "instrument_id": "PAY0119",
+    "type": "card",
+    "last4_hash": "173deaaf2f52bf79f028106fdbb995142cd7d77fc8b15cfa2a82d93c13f76f3c"
+  },
+  {
+    "instrument_id": "PAY0120",
+    "type": "card",
+    "last4_hash": "80ae95e050f70cddb9725d71e8fa265f27ecb95db2613461fd6d1c085f2dab0c"
+  },
+  {
+    "instrument_id": "PAY0121",
+    "type": "card",
+    "last4_hash": "e12868d1f3963e0e1129b3a6442b6d3d7d8e4eb7e27d34348b29523f0ae37748"
+  },
+  {
+    "instrument_id": "PAY0122",
+    "type": "upi",
+    "last4_hash": "d21c586a53e9762ebe64b80b124b6b6f00e3549ec304f6d4aa8d0c65c02e8704"
+  },
+  {
+    "instrument_id": "PAY0123",
+    "type": "card",
+    "last4_hash": "2c6b7bf2a4ef2fc87ce92b7097c6c477825b898877f3c7406c9609f7e5ddb8e6"
+  },
+  {
+    "instrument_id": "PAY0124",
+    "type": "upi",
+    "last4_hash": "204674eb811a35ba00447c090c783777da3a020da4f14a3ee0f335e1cfb15481"
+  },
+  {
+    "instrument_id": "PAY0125",
+    "type": "upi",
+    "last4_hash": "63892e272bb1f0641fd09ad8fe6a1e8192c901dabe381ae4406f9864d6ae33af"
+  },
+  {
+    "instrument_id": "PAY0126",
+    "type": "upi",
+    "last4_hash": "5aed8261495824595ecc62f78b62a0a2d4db3cdee22034652d7da89ca7c3e5f5"
+  },
+  {
+    "instrument_id": "PAY0127",
+    "type": "card",
+    "last4_hash": "bc805497f86694cef2e1cf6df62989cb818f69a80f4fb170e5b7cd66aef5d64b"
+  },
+  {
+    "instrument_id": "PAY0128",
+    "type": "card",
+    "last4_hash": "3b040bdd21634e7a8d08619c89e37e4c8430e955fc88e3888b1c13bda6ec32f3"
+  },
+  {
+    "instrument_id": "PAY0129",
+    "type": "card",
+    "last4_hash": "3141d9c749e88ee5550192703c3e025a9b3446ec543f17a923cb24e7b61dfece"
+  },
+  {
+    "instrument_id": "PAY0130",
+    "type": "upi",
+    "last4_hash": "cf14794b7a9b6f825719bdda892b9045b829a28de7b72f97b5cff93acd8d1433"
+  },
+  {
+    "instrument_id": "PAY0131",
+    "type": "upi",
+    "last4_hash": "ce048a27aa4bec0b76ec7ae0641f3922a3aa58625051690fedf4994419c320db"
+  },
+  {
+    "instrument_id": "PAY0132",
+    "type": "upi",
+    "last4_hash": "0c9bd7be38dee8ce8a5b02f67b9d656793177d9849808032ad412708b018159c"
+  },
+  {
+    "instrument_id": "PAY0133",
+    "type": "card",
+    "last4_hash": "e0b372a86dbfe71bdb7bf19790cacabd2ad2c4d145087a94d842fe6ffe2eef8b"
+  },
+  {
+    "instrument_id": "PAY0134",
+    "type": "card",
+    "last4_hash": "154d72ced7915c97e73258a22c08e202e8974629b1847c87d516fa6e692672a4"
+  },
+  {
+    "instrument_id": "PAY0135",
+    "type": "upi",
+    "last4_hash": "a8abbe1f7b05c096d4875eb9a7bdc9d8b539bd230ee8a1e02d2346bb91c2c080"
+  },
+  {
+    "instrument_id": "PAY0136",
+    "type": "card",
+    "last4_hash": "aeba47c17ca09557808c333b49af210a0209aff6b4334fa7192aaed69c5f36dd"
+  },
+  {
+    "instrument_id": "PAY0137",
+    "type": "upi",
+    "last4_hash": "708cdcc0590a516387b0a516be7719fc8ad25d5653c6d22cf3ae6d690cf9d126"
+  },
+  {
+    "instrument_id": "PAY0138",
+    "type": "upi",
+    "last4_hash": "2013ea8b947491e0829514c81c52326ae69a14c1cc664630584252db2c63de82"
+  },
+  {
+    "instrument_id": "PAY0139",
+    "type": "upi",
+    "last4_hash": "817856dcff013f1e9f1a8edb78c4e86baaba6fcf7d80b6d941a1c4cf888cc32b"
+  },
+  {
+    "instrument_id": "PAY0140",
+    "type": "card",
+    "last4_hash": "0287240ca63d3c73cb1231b69fce2315e999f3fce091140acf6ec1c6e5ba094f"
+  },
+  {
+    "instrument_id": "PAY0141",
+    "type": "upi",
+    "last4_hash": "6de4e9cef3c5ae3c7fffbac619b70bfd7a22202d2e509456cbcad3b2fed89b67"
+  },
+  {
+    "instrument_id": "PAY0142",
+    "type": "card",
+    "last4_hash": "fc2cd6cb85339b6c15848a08e7713f73cf55148b21cb32fb78cdf5ff44c90634"
+  },
+  {
+    "instrument_id": "PAY0143",
+    "type": "upi",
+    "last4_hash": "771e6ad04e8013ba5c223e96eb0bc38e3204a962bceac2aa4b84103c40472d8a"
+  },
+  {
+    "instrument_id": "PAY0144",
+    "type": "card",
+    "last4_hash": "e209f7b7bfe0f9ccd9c6b5d8aa497b55b76812f56ff0cddd6fc9f703f17192b3"
+  },
+  {
+    "instrument_id": "PAY0145",
+    "type": "upi",
+    "last4_hash": "21ed781581839a96745dd321ed9348facf34331e26259f530953741a22503ece"
+  },
+  {
+    "instrument_id": "PAY0146",
+    "type": "upi",
+    "last4_hash": "d4f4ef122bb30e6818329d5fda158711c5b9384bc86a8c7bcfe7cd85435c8a6d"
+  },
+  {
+    "instrument_id": "PAY0147",
+    "type": "card",
+    "last4_hash": "79c56d57697ddb331d1e1a1b62fa940d1579447b35221a7f6b68924dfce14244"
+  },
+  {
+    "instrument_id": "PAY0148",
+    "type": "card",
+    "last4_hash": "0c921bbabc0503bc23abfa92b5a4dab3c14f362b16096238fb4a71c279bc0eeb"
+  },
+  {
+    "instrument_id": "PAY0149",
+    "type": "upi",
+    "last4_hash": "0405a13200aa02e63862cbfd66a5beb1129f206eab06fe05ec0b8dc0217825b0"
+  },
+  {
+    "instrument_id": "PAY0150",
+    "type": "upi",
+    "last4_hash": "2be4f967ce834c65b10d2f91f45ab8fbfe083b468a6247363056ff29091d34bc"
+  },
+  {
+    "instrument_id": "PAY0151",
+    "type": "upi",
+    "last4_hash": "a5382c5d33b1be5721daf7b45e98810db32b230b67f5ffe23287be1480a6068c"
+  },
+  {
+    "instrument_id": "PAY0152",
+    "type": "upi",
+    "last4_hash": "09a9221450d87d97c48b18fde1c53ca2214c426f90d2207a3d66c7e65b6d3da8"
+  },
+  {
+    "instrument_id": "PAY0153",
+    "type": "upi",
+    "last4_hash": "03438f89353aee02887a151bd0d762f4bd10f7ec9d9d0e4536df399c92ae4103"
+  },
+  {
+    "instrument_id": "PAY0154",
+    "type": "upi",
+    "last4_hash": "f802afc52a957e1ed75ee48fbd23ee909a57f3253aaa28ee9080b1a6ae11e48e"
+  },
+  {
+    "instrument_id": "PAY0155",
+    "type": "card",
+    "last4_hash": "5078264a209e1ee0e3a8a46a11d840dfd1879500f7fd02aef1f8b6a2276a5be9"
+  },
+  {
+    "instrument_id": "PAY0156",
+    "type": "upi",
+    "last4_hash": "b7b90855c2dce7710c8e11ac3de1bbaea0656042dae41ec753f843a8ff372c9f"
+  },
+  {
+    "instrument_id": "PAY0157",
+    "type": "upi",
+    "last4_hash": "313b93b4b6589d56dcc1baaaccca223a337628b43d2ef51af234b03c242b9895"
+  },
+  {
+    "instrument_id": "PAY0158",
+    "type": "card",
+    "last4_hash": "02be6761f02f2291c1fe54c103440f7b9ba65da74dc84f1e08736d7a67824643"
+  },
+  {
+    "instrument_id": "PAY0159",
+    "type": "upi",
+    "last4_hash": "11e62f15fe7bc8ee83afaf873048a419a8aea5ef444c49e73bc86bd3968b9d01"
+  },
+  {
+    "instrument_id": "PAY0160",
+    "type": "upi",
+    "last4_hash": "541823e83415b95589e462d2508c4f67104b90322e542373205b7c62d1ed4468"
+  },
+  {
+    "instrument_id": "PAY0161",
+    "type": "card",
+    "last4_hash": "b4eed30197a01d2d80fb165bd90e21ebe54ec513216a8f03a3ed263d98f00f03"
+  },
+  {
+    "instrument_id": "PAY0162",
+    "type": "upi",
+    "last4_hash": "ead77f513269818b47d13226728ff3c031ead406abb8d0d70c247da00aa9eb75"
+  },
+  {
+    "instrument_id": "PAY0163",
+    "type": "card",
+    "last4_hash": "deb4666a75564b9fceea3b10aab44ec24089d914d7bb262b51145392856e3714"
+  },
+  {
+    "instrument_id": "PAY0164",
+    "type": "card",
+    "last4_hash": "b5a711fe7e92051816302057edc97e9e4a471802b96e379e8d382b4e9f248478"
+  },
+  {
+    "instrument_id": "PAY0165",
+    "type": "upi",
+    "last4_hash": "0248b9e54ef1e41897a3198b509166905ab9d941758886bbfb03bd72449bf63e"
+  },
+  {
+    "instrument_id": "PAY0166",
+    "type": "upi",
+    "last4_hash": "22d3ae0abdf720e6adb3f0e4b81366e3fbeee65658825bf2f10f43fa92c1f298"
+  },
+  {
+    "instrument_id": "PAY0167",
+    "type": "upi",
+    "last4_hash": "7ceb89878579e1cb86a8fd6fb4a5cc8abad8660793354ee9b9d0371c3fb4f924"
+  },
+  {
+    "instrument_id": "PAY0168",
+    "type": "upi",
+    "last4_hash": "de7217c8927ffa8f8f97c5871b9c2c1d5f543b115cc71aa30150439c42e5fb71"
+  },
+  {
+    "instrument_id": "PAY0169",
+    "type": "card",
+    "last4_hash": "28af0d12785efe04f62544cbd49382ebbce76256e800fecaeb4e4b7871d402bb"
+  },
+  {
+    "instrument_id": "PAY0170",
+    "type": "card",
+    "last4_hash": "4708f9c61a5e58b72774193f69bf37442423a4d2673e1f2cdc3f07f2614aba5d"
+  },
+  {
+    "instrument_id": "PAY0171",
+    "type": "card",
+    "last4_hash": "fa2fd37dfe0f812d3dc55fb445a1a4d04b95032b6753f0d4943313655bd3f8e8"
+  },
+  {
+    "instrument_id": "PAY0172",
+    "type": "upi",
+    "last4_hash": "01f955f2f2122036a9cc5ed2a780461307978fddff83f39da70b6d63a4d2b2e8"
+  },
+  {
+    "instrument_id": "PAY0173",
+    "type": "card",
+    "last4_hash": "747edaa66dec70dc1256e51b15be0fe658902a7af7a427db024df3431678152c"
+  },
+  {
+    "instrument_id": "PAY0174",
+    "type": "upi",
+    "last4_hash": "a81e1a8dde5c3b93cdb41a6b6cc6648a5131af5cd3ebe6b79ebb6fe97c3b418e"
+  },
+  {
+    "instrument_id": "PAY0175",
+    "type": "upi",
+    "last4_hash": "b9447cbd1457eb1c2caded2e63e3a8fd05d507415f6ad4afcf49ccc128dc5697"
+  },
+  {
+    "instrument_id": "PAY0176",
+    "type": "card",
+    "last4_hash": "228971770cc3111e6ce32b17a5ab6ae7c0ba993634ac8d83fd32ec6e91615995"
+  },
+  {
+    "instrument_id": "PAY0177",
+    "type": "upi",
+    "last4_hash": "6f7cef78841fc154d6bfb8be9bf2a47e83608cdac4a926d63dc80ffd83a66378"
+  },
+  {
+    "instrument_id": "PAY0178",
+    "type": "card",
+    "last4_hash": "80e107ee3ba12129c178396a4b3a4c502210b378e8b85d3b14087d7becbcfde7"
+  },
+  {
+    "instrument_id": "PAY0179",
+    "type": "card",
+    "last4_hash": "0c064516a0ec5302a2d2c2c270b509ca7a100b012445086f8900891310bdb9f3"
+  },
+  {
+    "instrument_id": "PAY0180",
+    "type": "upi",
+    "last4_hash": "f4fcc2f7dccbda17aa92ab8605e0fdd2f9861670fb8c5738241a029c8fc617ae"
+  },
+  {
+    "instrument_id": "PAY0181",
+    "type": "upi",
+    "last4_hash": "cb898ba3b47402b2495eb12f3902cb3e82775f59b3a8d9f726c5c9ebb8016352"
+  },
+  {
+    "instrument_id": "PAY0182",
+    "type": "upi",
+    "last4_hash": "7c27c01a452ad2badbecfd322f506a6548830e9a8e35c981c14b670cd7f70c95"
+  },
+  {
+    "instrument_id": "PAY0183",
+    "type": "upi",
+    "last4_hash": "9cc2c4fccfd99215e979acb8954a605c5c78b62013aa22c501bb8b8a214114c6"
+  },
+  {
+    "instrument_id": "PAY0184",
+    "type": "upi",
+    "last4_hash": "8dcedc831b40346fadc5ca491a72ba1dad3f3b68ac3296ef7ef13bdb9082b92b"
+  },
+  {
+    "instrument_id": "PAY0185",
+    "type": "upi",
+    "last4_hash": "a724557fb7d3f2c6838beccb4ad12f8be4e5d8383529b8c7f53ab4544c085ea9"
+  },
+  {
+    "instrument_id": "PAY0186",
+    "type": "upi",
+    "last4_hash": "49d223c477447c4bdb7061d8b342eb725eed77cf8e8a0c2e99e19328fd7ca39d"
+  },
+  {
+    "instrument_id": "PAY0187",
+    "type": "upi",
+    "last4_hash": "bf3433d6d20b35fe62cb7b46f00c64ab984686f7a708ce00277ee3c26fae51f2"
+  },
+  {
+    "instrument_id": "PAY0188",
+    "type": "upi",
+    "last4_hash": "a992b986fc27b23a80c1c673ecfbbbc8957647865f50c63d1c30be78311f0cce"
+  },
+  {
+    "instrument_id": "PAY0189",
+    "type": "upi",
+    "last4_hash": "04002abb101a19fcabea0f4b87b8b79fe73dee47233d463fcbd9e65d17a6ef8d"
+  },
+  {
+    "instrument_id": "PAY0190",
+    "type": "card",
+    "last4_hash": "a24a6dd3d30dc6b15f3e9ea581805d199270323842004e03ddf3c932478b1f11"
+  },
+  {
+    "instrument_id": "PAY0191",
+    "type": "card",
+    "last4_hash": "35c7a070e74b4380b5cb8cb0b68f3a06b331e4477b720c43df7af5c4c36fd2c0"
+  },
+  {
+    "instrument_id": "PAY0192",
+    "type": "card",
+    "last4_hash": "5c6ab6a10221871a18b25558a77a99d1324732e4d5ac403e0bed5d85acba24fd"
+  },
+  {
+    "instrument_id": "PAY0193",
+    "type": "card",
+    "last4_hash": "1f61c05e5a94a1995a4dbaf9e3b9280190246279d7d9a0f229b16627fe15845a"
+  },
+  {
+    "instrument_id": "PAY0194",
+    "type": "card",
+    "last4_hash": "1628e107773c273d5509e2051386207d1b75f1dc3a97e6e4844e132837fae396"
+  },
+  {
+    "instrument_id": "PAY0195",
+    "type": "upi",
+    "last4_hash": "246b0ab60280d23005e7aaf3941fced7b84e933dd5e05be3b404e90dea26368d"
+  },
+  {
+    "instrument_id": "PAY0196",
+    "type": "upi",
+    "last4_hash": "8aa885ab588858db215e6b851dab0bc5ea5c2f058a0cc193501e56d30e5c6143"
+  },
+  {
+    "instrument_id": "PAY0197",
+    "type": "card",
+    "last4_hash": "3b96b9d2dc7cd7d338d73571d7c46eefe383955e4427d8a2079f4c41b374d988"
+  },
+  {
+    "instrument_id": "PAY0198",
+    "type": "upi",
+    "last4_hash": "847cf2f668d88b1db87b3e8a2e991510040236da249d12e2beaa36c15e34dc09"
+  },
+  {
+    "instrument_id": "PAY0199",
+    "type": "card",
+    "last4_hash": "153de641ea0557acad28b609f949ec7ad909c0168b9af1d810fcf26e45e7e428"
+  },
+  {
+    "instrument_id": "PAY0200",
+    "type": "card",
+    "last4_hash": "1adb67461a4c3739283e6b97ba7984e0d4b024a4dcbf1f341c389f0c581fde4a"
+  },
+  {
+    "instrument_id": "PAY0201",
+    "type": "card",
+    "last4_hash": "5f9e3d4a0fa6f44a726c146ead2a301eb4026ae67b06cc5971210e5d749762b0"
+  },
+  {
+    "instrument_id": "PAY0202",
+    "type": "card",
+    "last4_hash": "647011fb41f77ae398ca6e97e8ee589270760023219a9a1f6d6eaf0055c75c8d"
+  },
+  {
+    "instrument_id": "PAY0203",
+    "type": "card",
+    "last4_hash": "baa8a6ecbabbd5ef02222983b23a4fbd1d794cca68c767e07d67af2d9f46e737"
+  },
+  {
+    "instrument_id": "PAY0204",
+    "type": "upi",
+    "last4_hash": "c6e1a3b42f1021a8897367a8ecc5906409856ad96b3429d9209f56abb44a1773"
+  },
+  {
+    "instrument_id": "PAY0205",
+    "type": "card",
+    "last4_hash": "ee4f8781fe28ccc751937330fc03ce427460b7ad4a0cf741019863eba4e53a6c"
+  },
+  {
+    "instrument_id": "PAY0206",
+    "type": "card",
+    "last4_hash": "15197cf7214b58e67cae565e573ffd9aa44bcb81f8ee5d7185dfe8da0a16ef43"
+  },
+  {
+    "instrument_id": "PAY0207",
+    "type": "upi",
+    "last4_hash": "46f292108babd9c0500ef511728ac5099eb586c255eebc72ab2499c6df520384"
+  },
+  {
+    "instrument_id": "PAY0208",
+    "type": "upi",
+    "last4_hash": "331160121aa00fa6b8954bf8b72f9dcfead4093d79b1ed0e2b50379b9e53479c"
+  },
+  {
+    "instrument_id": "PAY0209",
+    "type": "upi",
+    "last4_hash": "8f3cd17919924b85057eace5ba32fda81fff13141c996693288487601ea4a717"
+  },
+  {
+    "instrument_id": "PAY0210",
+    "type": "upi",
+    "last4_hash": "d7defea67028f03ef163788c9c2d89392f5b26588771452c170625109f2f3406"
+  },
+  {
+    "instrument_id": "PAY0211",
+    "type": "upi",
+    "last4_hash": "6588bc3e00cce5a692f811a8c1e96c9c8d22168939143dc6713258fdd4da54e2"
+  },
+  {
+    "instrument_id": "PAY0212",
+    "type": "card",
+    "last4_hash": "9e645831962e43f534afd60532328ba5ab4eefea6e73765fb9690e4d395112f8"
+  },
+  {
+    "instrument_id": "PAY0213",
+    "type": "upi",
+    "last4_hash": "fcbaac549e7b8ade5d0f9826413f5bd2c0beb47df11d76c889c9b417d2446295"
+  },
+  {
+    "instrument_id": "PAY0214",
+    "type": "upi",
+    "last4_hash": "9fd6e8b29087ba21fabbc6c46979750b8008ed05ba77b9e3ff032b3bd3d5992e"
+  },
+  {
+    "instrument_id": "PAY0215",
+    "type": "card",
+    "last4_hash": "06c973e49be9fdbfdaa84a397b35b05baceec93d5ea35e4cba2d6eaecf97700f"
+  },
+  {
+    "instrument_id": "PAY0216",
+    "type": "upi",
+    "last4_hash": "c4af926b851b86d5aeb3264a3f62db4b4944d664be2b9570512764afd8223b18"
+  },
+  {
+    "instrument_id": "PAY0217",
+    "type": "upi",
+    "last4_hash": "a276a4e96e2b4b8f8ce5ed09536cb5cd0df9b7c12351e00aa839c0d209b8a175"
+  },
+  {
+    "instrument_id": "PAY0218",
+    "type": "card",
+    "last4_hash": "8a80efe71cf1850186e908f82e61d9ab6c95889d927cfec59d69f8aeb0673ff7"
+  },
+  {
+    "instrument_id": "PAY0219",
+    "type": "card",
+    "last4_hash": "2a0400a50c04a5c64e7818f4920059acbedcc2a193e715b0f0310923405b36e8"
+  },
+  {
+    "instrument_id": "PAY0220",
+    "type": "upi",
+    "last4_hash": "29a217fcc45f734ac6d51f8c61d010cb40b8fb2b73e3d626b6ff3772c113a7a0"
+  },
+  {
+    "instrument_id": "PAY0221",
+    "type": "card",
+    "last4_hash": "b6cfb9e281f64066b6cb04d5d9108276f4e3ecbc68c860831c481ba932591673"
+  },
+  {
+    "instrument_id": "PAY0222",
+    "type": "card",
+    "last4_hash": "2a2ea325451e51e5f810010975fad2f66ad56342d85586d74bc0a2b3e31c8356"
+  },
+  {
+    "instrument_id": "PAY0223",
+    "type": "card",
+    "last4_hash": "8d3498f1ebcbd5f1bd973e184059a6f0e06812c898d4375da115bdedff492d0f"
+  },
+  {
+    "instrument_id": "PAY0224",
+    "type": "upi",
+    "last4_hash": "f36a35c022224e38de6430182e0a0fa554ce384997ce1086599f6317e832d9be"
+  },
+  {
+    "instrument_id": "PAY0225",
+    "type": "upi",
+    "last4_hash": "1a18b0a3dde64484d80eddbf00483e3681e94109324ae8368917e5ec79d0b634"
+  },
+  {
+    "instrument_id": "PAY0226",
+    "type": "upi",
+    "last4_hash": "1991e7b38912c102c0207ee57f8fa9a5ef96f1e93ae277d7e6bb0e35d8a23483"
+  },
+  {
+    "instrument_id": "PAY0227",
+    "type": "upi",
+    "last4_hash": "841bed395f6f9ce6dc0d396f472fc2e31dab0dcd0def14798686d8518e6c34fb"
+  },
+  {
+    "instrument_id": "PAY0228",
+    "type": "upi",
+    "last4_hash": "96f227129556c6f3080ebfe4619c520c6ee640fec486c406dc28cd95da33b3fd"
+  },
+  {
+    "instrument_id": "PAY0229",
+    "type": "upi",
+    "last4_hash": "cc4a428f3be823f76b73d2e772096feb84244db682b479d1038d58d0fa278b30"
+  },
+  {
+    "instrument_id": "PAY0230",
+    "type": "card",
+    "last4_hash": "63727ee4cff30e597899af01a4a355ea91e54c9a055e1da18e25b89dfa90352c"
+  },
+  {
+    "instrument_id": "PAY0231",
+    "type": "card",
+    "last4_hash": "3d19e945ed5a25a399da3036bc862275296544218de8260d47a5ad3d9b853af4"
+  },
+  {
+    "instrument_id": "PAY0232",
+    "type": "upi",
+    "last4_hash": "5ad4c5cccee58bdc87eda85a48e67559432b3ebb4092d49cb142d8ed206bfc7c"
+  },
+  {
+    "instrument_id": "PAY0233",
+    "type": "card",
+    "last4_hash": "a948b46c0f1890667de7b60fb490c573d60c647158f91193dd915afde693529c"
+  },
+  {
+    "instrument_id": "PAY0234",
+    "type": "card",
+    "last4_hash": "f23784f17468ea72b2dd3fd4e4faecb40ccdd24f39584fcd283c391d4faf87c8"
+  },
+  {
+    "instrument_id": "PAY0235",
+    "type": "card",
+    "last4_hash": "739dbef65f3b1e4117d7bdb2a76efa5b91943f8f26ab74a7c1cad9feb0324129"
+  },
+  {
+    "instrument_id": "PAY0236",
+    "type": "card",
+    "last4_hash": "93352a86094a964ef548e72c79a9f6187b7f561773c31f3371e86ecffe9a4dd9"
+  },
+  {
+    "instrument_id": "PAY0237",
+    "type": "upi",
+    "last4_hash": "9569757a6fce9018ec528b20fc71028be89e24df343779ff25a281d38a330a3d"
+  },
+  {
+    "instrument_id": "PAY0238",
+    "type": "upi",
+    "last4_hash": "28ef70984a8eb80ae8606716d8824302fa6e616bb525846f7088e774ce66684c"
+  },
+  {
+    "instrument_id": "PAY0239",
+    "type": "upi",
+    "last4_hash": "b44cb9e82dd50599a74e865bd98505ff40dd1f873ff236bbac53b7982a678678"
+  },
+  {
+    "instrument_id": "PAY0240",
+    "type": "card",
+    "last4_hash": "7b4c3b903da50f31cacf58554525088f6ca687144330569605c8de3a404d2ce6"
+  },
+  {
+    "instrument_id": "PAY0241",
+    "type": "upi",
+    "last4_hash": "0bf788aaa8956e3115a533477b96a937c64a7cd80f318540fb2ce36ebd4fcf32"
+  },
+  {
+    "instrument_id": "PAY0242",
+    "type": "upi",
+    "last4_hash": "4e29a008959a9ddb23bd5d117d6a71f1e75f3b17df6ca4dd313a16711ea6ec45"
+  },
+  {
+    "instrument_id": "PAY0243",
+    "type": "upi",
+    "last4_hash": "3206fba387884735f6ec58259d62fe0943b0ee14fe9dc5137cc676428cb07bb9"
+  },
+  {
+    "instrument_id": "PAY0244",
+    "type": "card",
+    "last4_hash": "8f10833e3ef270a248b7001d1f7b30b7d05e67d648ce3e87f000215e38451b7b"
+  },
+  {
+    "instrument_id": "PAY0245",
+    "type": "upi",
+    "last4_hash": "a10b8d76c071f9e86d3906e6e15cf75709a4c6b2d7439ec2cc44e97f08119936"
+  },
+  {
+    "instrument_id": "PAY0246",
+    "type": "card",
+    "last4_hash": "9da8780a381915da91ad2693249937745283266704661097a162c48c38976205"
+  },
+  {
+    "instrument_id": "PAY0247",
+    "type": "card",
+    "last4_hash": "c44066cadb3f91efc94e8a80b0ed370d9e9775d66fad1dac4e546bf9f1916df9"
+  },
+  {
+    "instrument_id": "PAY0248",
+    "type": "card",
+    "last4_hash": "6b724a6630ccac9ac84a7af1031414c6b6b6a16b59aafe635d5b93b12d44c7bf"
+  },
+  {
+    "instrument_id": "PAY0249",
+    "type": "upi",
+    "last4_hash": "6ac7bd0865c7d5d212b9a9b1c92bb4b85a3a06e03e9bbc458782ad4c9382345f"
+  },
+  {
+    "instrument_id": "PAY0250",
+    "type": "card",
+    "last4_hash": "e2135c76ca7863e51154829de3d9cb0a8f6398d31518c7581d370ed9cc521811"
+  },
+  {
+    "instrument_id": "PAY0251",
+    "type": "card",
+    "last4_hash": "45d1bf544943b021e561523370fcb4b73bf8c67de1b68a3ec825fb28144ff434"
+  },
+  {
+    "instrument_id": "PAY0252",
+    "type": "upi",
+    "last4_hash": "22ea6cd8dc9b0b8247357f4c1736437b7cf6d5e886f8a386948476e7f72b0fc4"
+  },
+  {
+    "instrument_id": "PAY0253",
+    "type": "upi",
+    "last4_hash": "5d6076ec8f36560a68a9a3d910e459f6250c5acb00237a9558102f863a4955ab"
+  },
+  {
+    "instrument_id": "PAY0254",
+    "type": "upi",
+    "last4_hash": "89f7bcb6f95d9f29063738e097b0255e0bcc57a39caeff49ddf6925789120fc5"
+  },
+  {
+    "instrument_id": "PAY0255",
+    "type": "card",
+    "last4_hash": "1b457162acd3bc249096a759943769e27b27ecbd1fb20a870b70b82e9c0264d6"
+  },
+  {
+    "instrument_id": "PAY0256",
+    "type": "upi",
+    "last4_hash": "4e52a13115d05bc553e5de70cba00a6ff3ec75172ab1015133999f4907b06739"
+  },
+  {
+    "instrument_id": "PAY0257",
+    "type": "upi",
+    "last4_hash": "ce9020dc8536f77c86e0179914a12928e8171560a9f865f50a905f4b175f2fe2"
+  },
+  {
+    "instrument_id": "PAY0258",
+    "type": "upi",
+    "last4_hash": "c11930573da1982285db8a450fbe77496b12ff78b3585ba26f70d4c9fa325dac"
+  },
+  {
+    "instrument_id": "PAY0259",
+    "type": "upi",
+    "last4_hash": "e346f2da1da79fb8f2e685ba08e28022f4f11c15169f74e7fc17d8bfce87865f"
+  },
+  {
+    "instrument_id": "PAY0260",
+    "type": "upi",
+    "last4_hash": "f766d02d1d0991fdde7d1ac9e3e1a5164f68f3bd79328a1a42c5c9484237606b"
+  },
+  {
+    "instrument_id": "PAY0261",
+    "type": "card",
+    "last4_hash": "ef0d2e96b08cee59a3557ca2ddba84bd84b4dabd4e2f882122a291bffee0acdd"
+  },
+  {
+    "instrument_id": "PAY0262",
+    "type": "upi",
+    "last4_hash": "3760c7faec9d9f32031e20781e1123f31b2085d3407995ea8f9c195a90fbe0b0"
+  },
+  {
+    "instrument_id": "PAY0263",
+    "type": "upi",
+    "last4_hash": "7c5fb304ddb31dad9639fe0566854024009252dd1edc43cee8d3b2fefe63c2cc"
+  },
+  {
+    "instrument_id": "PAY0264",
+    "type": "upi",
+    "last4_hash": "694c4311e4d3d0aa2d05a079ec09c9bffbbbf38ccee37912e59fcfa13d0cf36b"
+  },
+  {
+    "instrument_id": "PAY0265",
+    "type": "card",
+    "last4_hash": "94f136bbfce57e5c6f1c86bbf6bf8a4e1d48ef230373eda2859170e158b1c185"
+  },
+  {
+    "instrument_id": "PAY0266",
+    "type": "card",
+    "last4_hash": "462b01670e50875a9df97c18e87cd367dbe7fbd6f32783be20883747e37d5af6"
+  },
+  {
+    "instrument_id": "PAY0267",
+    "type": "upi",
+    "last4_hash": "b5c6ea4587d7483b25440b9af696a3cf45757a33dc540449fff4d084d5669179"
+  },
+  {
+    "instrument_id": "PAY0268",
+    "type": "card",
+    "last4_hash": "0228374d12ee995cdeff8e25d819990f80d40c8773321fafe4ce1572c7df29af"
+  },
+  {
+    "instrument_id": "PAY0269",
+    "type": "upi",
+    "last4_hash": "dbcbd844304e9fc9edd4bead28de1ce924bb41b2a4a602591ac1a4051c5f426f"
+  },
+  {
+    "instrument_id": "PAY0270",
+    "type": "upi",
+    "last4_hash": "cccbf5f0cdc758831aec3f19afd46cfa357cc9b667b1eeb3c1a93776e824df10"
+  },
+  {
+    "instrument_id": "PAY0271",
+    "type": "upi",
+    "last4_hash": "d34e5be7cf681f8f8f5e2aca5d0b120463f3d560a2b827ce0d12a8534600d179"
+  },
+  {
+    "instrument_id": "PAY0272",
+    "type": "card",
+    "last4_hash": "e1d07bb0923258ae0e244dcf81f4c87614b833fc15ca7dbb77740d1860f915d2"
+  },
+  {
+    "instrument_id": "PAY0273",
+    "type": "upi",
+    "last4_hash": "8da78c744ccbc95e9980bb28010978baaae42fdc0341cdc6c52e49df0bec5337"
+  },
+  {
+    "instrument_id": "PAY0274",
+    "type": "card",
+    "last4_hash": "b1e9300609f0a4556b09d64f62814c5e0e839e4353f262869bbd78e745f458b1"
+  },
+  {
+    "instrument_id": "PAY0275",
+    "type": "upi",
+    "last4_hash": "9dd391923f6fcb022c671681f977b1c8220bd55d67bbf4cf92c7134765ac1110"
+  },
+  {
+    "instrument_id": "PAY0276",
+    "type": "upi",
+    "last4_hash": "666a8b82b2fa22b380990b945a7253edb81077025e894f9a240f79e34e12073f"
+  },
+  {
+    "instrument_id": "PAY0277",
+    "type": "upi",
+    "last4_hash": "a44c062db40c7f0dfa7759a203df799c3ec1481692f1f83b0f0ac0982949cebf"
+  },
+  {
+    "instrument_id": "PAY0278",
+    "type": "upi",
+    "last4_hash": "6ea839f917ae2c0a0f39dee4625f232a1d6437678c541ecd3c94f1954e4c3159"
+  },
+  {
+    "instrument_id": "PAY0279",
+    "type": "card",
+    "last4_hash": "dd6103b473273ecdbf0446eb7e22d5e3397e03f2e275379e4427a36cfc6e87e6"
+  },
+  {
+    "instrument_id": "PAY0280",
+    "type": "upi",
+    "last4_hash": "cec4f851cb0a168f0193b99eddefa1a34121054b40755447ea1a3366d843242a"
+  },
+  {
+    "instrument_id": "PAY0281",
+    "type": "upi",
+    "last4_hash": "69fc261f5af1673a971aa89d9ed10b17c59df9ef4297d175768431999fbd880e"
+  },
+  {
+    "instrument_id": "PAY0282",
+    "type": "upi",
+    "last4_hash": "cad2316eaa81e19e7bf2c3f0033243acf6cef44d18c5fa5e8ad54d490d68bde7"
+  },
+  {
+    "instrument_id": "PAY0283",
+    "type": "upi",
+    "last4_hash": "c2c26a9990dd4dfa95ad0087849ddb39141a04454ad2377c1a2c35b3a388b57b"
+  },
+  {
+    "instrument_id": "PAY0284",
+    "type": "upi",
+    "last4_hash": "beb62c82d72fd3a83113c930c0c3139e9418636d9e03ab36cec1956eed69524c"
+  },
+  {
+    "instrument_id": "PAY0285",
+    "type": "card",
+    "last4_hash": "94bcfe105ad50803f9a5dc7f60974e21d47f608826675250c699467f116608dc"
+  },
+  {
+    "instrument_id": "PAY0286",
+    "type": "card",
+    "last4_hash": "16d5164c6ff6bf7201a90e9ec406fe9d122c9894cda28854a05e68ff1f2e804c"
+  },
+  {
+    "instrument_id": "PAY0287",
+    "type": "upi",
+    "last4_hash": "cae39f6224161df60057c04084845eda4dfdb9649900ec17915a42986b8cff36"
+  },
+  {
+    "instrument_id": "PAY0288",
+    "type": "upi",
+    "last4_hash": "75a62033cfdee7f67ff55d062301bc80b3e62c727b18ddc9f0a44902a33afd86"
+  },
+  {
+    "instrument_id": "PAY0289",
+    "type": "card",
+    "last4_hash": "f24e819d0999cd9c97741467925b406532813857723c7bf4971a13a938b1e9ec"
+  },
+  {
+    "instrument_id": "PAY0290",
+    "type": "card",
+    "last4_hash": "56d5f315cf3312f1e33f250993515e272210f45eab5f60d84c892b70abfbfcfd"
+  },
+  {
+    "instrument_id": "PAY0291",
+    "type": "upi",
+    "last4_hash": "a5772758cde632e399e9a2b7724d9501a27aac225a129665910593a12c0c2278"
+  },
+  {
+    "instrument_id": "PAY0292",
+    "type": "upi",
+    "last4_hash": "c4ed339846ea9d2738e34a5b070bbbee4233f162a9bbde017ecf0c45e090628c"
+  },
+  {
+    "instrument_id": "PAY0293",
+    "type": "card",
+    "last4_hash": "2478b5cdde898af6ad0d268fed2485ec42b63add9b159770b287e7396b71088c"
+  },
+  {
+    "instrument_id": "PAY0294",
+    "type": "upi",
+    "last4_hash": "15996fbb0ec304c8c73f5553cd5197ce660af069ad8bbb9a8e65056e9595390d"
+  },
+  {
+    "instrument_id": "PAY0295",
+    "type": "upi",
+    "last4_hash": "3294898f52ec90ff6f3ec3ba983335fb585c8d190d66add034f6e7a27f063b82"
+  },
+  {
+    "instrument_id": "PAY0296",
+    "type": "upi",
+    "last4_hash": "d2325dfbe73e62fb166e0ad38da68236e73bca72bfa5fffb881f9a12f708cc5b"
+  },
+  {
+    "instrument_id": "PAY0297",
+    "type": "card",
+    "last4_hash": "d03e59748526d73618c46171c55d8d067777013738ef99f13b529687bf45e453"
+  },
+  {
+    "instrument_id": "PAY0298",
+    "type": "upi",
+    "last4_hash": "1acc803411f7af2580d5f0c757d4cb78f524e0ae3745a7f384909be50ab22f2b"
+  },
+  {
+    "instrument_id": "PAY0299",
+    "type": "card",
+    "last4_hash": "8d7daebcedc86c90dd1d3e3b90bbd2eaf71f37073bd1ee57efcdd4040316b26e"
+  },
+  {
+    "instrument_id": "PAY0300",
+    "type": "upi",
+    "last4_hash": "0a4508eafdae8d6a2bd1733dd666b2d5c59bfc151c5f32c04adac18f9b638f88"
+  },
+  {
+    "instrument_id": "PAY0301",
+    "type": "upi",
+    "last4_hash": "d70428d6bb194ce315ecab17251724d537377a4dfdbe25c0789891064785bd66"
+  },
+  {
+    "instrument_id": "PAY0302",
+    "type": "card",
+    "last4_hash": "e56ae9ee21661d3febabe8d1e03ee82d02466a5d405e89f2acf449cd6a6240b8"
+  },
+  {
+    "instrument_id": "PAY0303",
+    "type": "card",
+    "last4_hash": "c6b4c26e25da2dadec4164bd0a517057cfd5a1e873700a843ab37f4ef270d8ec"
+  },
+  {
+    "instrument_id": "PAY0304",
+    "type": "upi",
+    "last4_hash": "dbb5f75c348639638a2aab13c65b11dd7b0e0600d76ebd2750cf87ba9df98c36"
+  },
+  {
+    "instrument_id": "PAY0305",
+    "type": "card",
+    "last4_hash": "3e1884b53656e0d408c17c114fdd85744c1c61bd05262f049d2da235b84332de"
+  },
+  {
+    "instrument_id": "PAY0306",
+    "type": "card",
+    "last4_hash": "061b4fe85b47a13e1887f386477e0e0d5c9f3003f53e4b20472bbaa8b46ce766"
+  },
+  {
+    "instrument_id": "PAY0307",
+    "type": "card",
+    "last4_hash": "2d13ab427ffdaa125d6436319549bdaddd8670d8146a0a4f351d92e8c68adbf3"
+  },
+  {
+    "instrument_id": "PAY0308",
+    "type": "card",
+    "last4_hash": "3b37ef6a1be31e06efd2f3cfa7e70d2b25d95efec97e5b71503a95e249d13250"
+  },
+  {
+    "instrument_id": "PAY0309",
+    "type": "upi",
+    "last4_hash": "266dff09c92592d24ea4075cd51306a16180b0f3ebbdeb1ca9d82825ebc336d5"
+  },
+  {
+    "instrument_id": "PAY0310",
+    "type": "upi",
+    "last4_hash": "354b6eb4abca5abe614439fb0b8b2b99c4c2355038a2ad3258e85cfc77d02875"
+  },
+  {
+    "instrument_id": "PAY0311",
+    "type": "upi",
+    "last4_hash": "10f99f24ed74bf5c316688025910683a18ed067997d370489e303a575838f5ea"
+  },
+  {
+    "instrument_id": "PAY0312",
+    "type": "upi",
+    "last4_hash": "b2e598cf7f4d1db437d7d03a427053d34a26d051cab739b4ae2a47d16ddf9881"
+  },
+  {
+    "instrument_id": "PAY0313",
+    "type": "upi",
+    "last4_hash": "e64f311e1a8becec62c892f8b2867161e823a37e4ae01f747195d18973b1b3ca"
+  },
+  {
+    "instrument_id": "PAY0314",
+    "type": "upi",
+    "last4_hash": "1a80e1c99d7f19f80d33f52ba6f325c5d9161f4024a5a343806c52d697aa6751"
+  },
+  {
+    "instrument_id": "PAY0315",
+    "type": "card",
+    "last4_hash": "ee25805dac49adcdff3916bbd8bd1e8d5ee8f6d230f0e23a2614917da628ecf2"
+  },
+  {
+    "instrument_id": "PAY0316",
+    "type": "card",
+    "last4_hash": "ecdac8c096f7fb91ae3bcefb2f0cfaba82e07a674645c53e9f497c1afe73175e"
+  },
+  {
+    "instrument_id": "PAY0317",
+    "type": "card",
+    "last4_hash": "63adeada6605b9c8dd505f7f351e0c3dc03c77c19a53b41879a3123a087477b1"
+  },
+  {
+    "instrument_id": "PAY0318",
+    "type": "card",
+    "last4_hash": "21113a3f71a6ecda7311c6e8539d1c9bd9949d3e6f23fbf4bcc781f96b30083a"
+  },
+  {
+    "instrument_id": "PAY0319",
+    "type": "upi",
+    "last4_hash": "bbdc5fcb2b45ff7d3cd16a87f0671aaee83912eb9b32a35c19395e203768b4a3"
+  },
+  {
+    "instrument_id": "PAY0320",
+    "type": "upi",
+    "last4_hash": "9a2999f3ed28c49fcc116d94d091d0ff78986843cf1d9b4394e9539a439127ae"
+  },
+  {
+    "instrument_id": "PAY0321",
+    "type": "upi",
+    "last4_hash": "bf32228a9690f9fb114d8aa365c6526bd8d3e92647b96664dc9625f66f45dd12"
+  },
+  {
+    "instrument_id": "PAY0322",
+    "type": "upi",
+    "last4_hash": "b7f71f7714700248db2c27fcdc5ce58eecbbe6eaaeb6e4b7200acfe54ae96a25"
+  },
+  {
+    "instrument_id": "PAY0323",
+    "type": "card",
+    "last4_hash": "a5f643933b06668a9f74e6a6d378d41da76b4a7048fbabfd493c005eb2547a2e"
+  },
+  {
+    "instrument_id": "PAY0324",
+    "type": "upi",
+    "last4_hash": "c898fdb8c7646d10a32c1957f562db1c4ff16d2e9283dc065b4238363d6a5129"
+  },
+  {
+    "instrument_id": "PAY0325",
+    "type": "upi",
+    "last4_hash": "5e36561eadf5d0e282bdc2c52526cc239f4b438b7acc5e76f27be7762f2ca435"
+  },
+  {
+    "instrument_id": "PAY0326",
+    "type": "card",
+    "last4_hash": "41cf091084d6f30cf45e97b16bf165283dc5ab27f7999f7d31bd405e082e3fdd"
+  },
+  {
+    "instrument_id": "PAY0327",
+    "type": "upi",
+    "last4_hash": "3ac36e235d07e875c8de80ac99489e4563f34f7f18d4341619ebdb240519b524"
+  },
+  {
+    "instrument_id": "PAY0328",
+    "type": "upi",
+    "last4_hash": "c2132a1f935492df0fb57f1891781663046815e5cc48c89814fac4b72c6a7559"
+  },
+  {
+    "instrument_id": "PAY0329",
+    "type": "upi",
+    "last4_hash": "0ee1d8cda501a0cb3f2d5018074bbe666418af01d4447fca40c9a89ff8dcfbf0"
+  },
+  {
+    "instrument_id": "PAY0330",
+    "type": "upi",
+    "last4_hash": "aa59688cda0c3346742fa30c47cd9e560ad74edc205db781ed47b1709f00af8f"
+  },
+  {
+    "instrument_id": "PAY0331",
+    "type": "upi",
+    "last4_hash": "b3cb76dfa3ad98bf3afbf973b480efc0a1c6a3a07079582468753cfad76c8213"
+  },
+  {
+    "instrument_id": "PAY0332",
+    "type": "upi",
+    "last4_hash": "d0702d17dcf77b735276f8927f4b839e98bdcc4a04f3e4e5935cd564fc5c17fa"
+  },
+  {
+    "instrument_id": "PAY0333",
+    "type": "card",
+    "last4_hash": "1eab3dfd35a341f4b06d518d53f3f4701e69fedff5c65230df8fe475ac85b50c"
+  },
+  {
+    "instrument_id": "PAY0334",
+    "type": "upi",
+    "last4_hash": "7ef788d37b374936a437a4dfd5b4e6a020edc2e51d2f96132a213a8ff1d48b7a"
+  },
+  {
+    "instrument_id": "PAY0335",
+    "type": "card",
+    "last4_hash": "212375a537417e324d381e18e45023efd4f801ef3b35f0fe3983a2b58bdca4f6"
+  },
+  {
+    "instrument_id": "PAY0336",
+    "type": "card",
+    "last4_hash": "73bdbf955c7575599cb8c4618d79874c1b8e6c2888cbb3fbd3009d4744e2d218"
+  },
+  {
+    "instrument_id": "PAY0337",
+    "type": "upi",
+    "last4_hash": "99f2f36d5c419aa2134b9e3b20e343480c31ae6ff043e4385c720ab982407394"
+  },
+  {
+    "instrument_id": "PAY0338",
+    "type": "upi",
+    "last4_hash": "7b595dc28809c3d97bf4d9e3eda45ab903506341d09bff25f5746fe43ddab60a"
+  },
+  {
+    "instrument_id": "PAY0339",
+    "type": "upi",
+    "last4_hash": "cccdcedf3abed8d74731a23a8915a398027aa5b12ca03404f5d7e099db30f5ff"
+  },
+  {
+    "instrument_id": "PAY0340",
+    "type": "card",
+    "last4_hash": "77e359e4b1cb258f9f9d2d1dad6db4b74e3d0c5abdc1fcefa7ae126747386c27"
+  },
+  {
+    "instrument_id": "PAY0341",
+    "type": "card",
+    "last4_hash": "e12868d1f3963e0e1129b3a6442b6d3d7d8e4eb7e27d34348b29523f0ae37748"
+  },
+  {
+    "instrument_id": "PAY0342",
+    "type": "upi",
+    "last4_hash": "34a380a592a4c2a07b75793b430ee351affa2ca1ad546c9480142f48e6327955"
+  },
+  {
+    "instrument_id": "PAY0343",
+    "type": "card",
+    "last4_hash": "213f367b1af015a79e0b0cf92a3d4084e6bd5fbc3fbb4364881b05854bf86924"
+  },
+  {
+    "instrument_id": "PAY0344",
+    "type": "card",
+    "last4_hash": "fc42f716c296b127850fbca1e103af089ecf2fb1f7c152d8d5440c6130bb6018"
+  },
+  {
+    "instrument_id": "PAY0345",
+    "type": "upi",
+    "last4_hash": "3b399b28cde16ac76a67359ad2a450043f1d3d2878409fdc5ea1602f4d3a54d7"
+  },
+  {
+    "instrument_id": "PAY0346",
+    "type": "card",
+    "last4_hash": "e8202f39fad74acd9b6d38a6038da9e8682e857bbdb977cf7b913963bfa2b393"
+  },
+  {
+    "instrument_id": "PAY0347",
+    "type": "upi",
+    "last4_hash": "06688d0561c37484aa6c49d1f0339ec2f2072c281147aadd727de8ac463a1d95"
+  },
+  {
+    "instrument_id": "PAY0348",
+    "type": "card",
+    "last4_hash": "90d441ee3fce238cd433ff208078e8d4351fb2969f9ed6b595e588ed7ead5d4e"
+  },
+  {
+    "instrument_id": "PAY0349",
+    "type": "upi",
+    "last4_hash": "ee1ece82ffecfdc2200e5e48dcd8cc066370deb5e1f40869819d811479513a3a"
+  },
+  {
+    "instrument_id": "PAY0350",
+    "type": "upi",
+    "last4_hash": "c25d57a3151ee5284a8f037f6336a93978a496859309bdabd4c56a7c2a7b5a2e"
+  },
+  {
+    "instrument_id": "PAY0351",
+    "type": "upi",
+    "last4_hash": "7d980fadbd3bd55b3e1b798fb4b4b03382fa7707bb5ce37eb1bbbda580f83a46"
+  },
+  {
+    "instrument_id": "PAY0352",
+    "type": "upi",
+    "last4_hash": "6ef67c4f417b6f6c74d139a18b6050097ae4bbbc08320e83acb5a16490d20fdf"
+  },
+  {
+    "instrument_id": "PAY0353",
+    "type": "card",
+    "last4_hash": "cc56038328ffa43e94516e446dfde7b43c94f18e80e59a79476598a1f394da23"
+  },
+  {
+    "instrument_id": "PAY0354",
+    "type": "upi",
+    "last4_hash": "178cb40bf3dfbaac93fd5ae16d819c1758a5f2d0e655bb0ac5afadf8b42df466"
+  },
+  {
+    "instrument_id": "PAY0355",
+    "type": "upi",
+    "last4_hash": "cb57ca8d48ec0507d179908b7b9bfac3cd3fdebae39a78685baaf71588d1c7d7"
+  },
+  {
+    "instrument_id": "PAY0356",
+    "type": "upi",
+    "last4_hash": "e087a3643dd4fb96b2332a2bdbbf293149b3b0367345a13a11f2d19266199d4f"
+  },
+  {
+    "instrument_id": "PAY0357",
+    "type": "card",
+    "last4_hash": "b8dc2c143be8994682b08461f46487e05874e59dd9ab65cf973e3a3c67a763aa"
+  },
+  {
+    "instrument_id": "PAY0358",
+    "type": "card",
+    "last4_hash": "adb736c83f5cec1d6a275694c25b544137ab4d4c760c4dadf89523c4167d619a"
+  },
+  {
+    "instrument_id": "PAY0359",
+    "type": "card",
+    "last4_hash": "18167da210996cf3525e400870f7d4955d6b983a7b7d237586e242e59888ad86"
+  },
+  {
+    "instrument_id": "PAY0360",
+    "type": "upi",
+    "last4_hash": "5103fb8e14eb2dfc189a9906bc9919ba997350fb031790387759ad9cf7e08266"
+  },
+  {
+    "instrument_id": "PAY0361",
+    "type": "card",
+    "last4_hash": "a0869c71d5d3edb240ae3d132a0ff1d6be1c6468831710c4ce156d55e95c0190"
+  },
+  {
+    "instrument_id": "PAY0362",
+    "type": "upi",
+    "last4_hash": "a5d5e96c52a28d8c9d546c9d35fa722d7078965c111e628b452bf97881c7b5e9"
+  },
+  {
+    "instrument_id": "PAY0363",
+    "type": "card",
+    "last4_hash": "ecdd7308565e93e4912e8f9353031766e18d531cedfc452835951fdba61f695f"
+  },
+  {
+    "instrument_id": "PAY0364",
+    "type": "card",
+    "last4_hash": "6815fd0989fc6aa86c07ab321564405cedf401875868f38d40a4631cac92e599"
+  },
+  {
+    "instrument_id": "PAY0365",
+    "type": "upi",
+    "last4_hash": "3d6a4c17b72c189aeed3615836662368f2a3905437c216f67701a97d3e3743fa"
+  },
+  {
+    "instrument_id": "PAY0366",
+    "type": "card",
+    "last4_hash": "addcb3d9599cca858acab111858cfc4df9cd6cdf515a5cfdeba5c12e15ba11b3"
+  },
+  {
+    "instrument_id": "PAY0367",
+    "type": "upi",
+    "last4_hash": "923d83d13047ef02d57106faf2a8cfba184b6c037055ab3c231f365565ee7606"
+  },
+  {
+    "instrument_id": "PAY0368",
+    "type": "upi",
+    "last4_hash": "d98cbbd5a443ecf65f57ad3fb285624a2e1056c51ca2f6cb208af2b2af2e029e"
+  },
+  {
+    "instrument_id": "PAY0369",
+    "type": "upi",
+    "last4_hash": "06c860802560f4929f8c4b9a8aaeb4993a9585da5581afe8b9cbfd86e48e53db"
+  },
+  {
+    "instrument_id": "PAY0370",
+    "type": "card",
+    "last4_hash": "0f8eb4b72b6e0c9e88b388eb967b49e067ef1004bf07bffc22c3acb13b43580a"
+  },
+  {
+    "instrument_id": "PAY0371",
+    "type": "upi",
+    "last4_hash": "737762cbae1acba8015e18595a847ffeb0c9f47897944f1cef319d4c2e0445cf"
+  },
+  {
+    "instrument_id": "PAY0372",
+    "type": "card",
+    "last4_hash": "c69f0baaa1af7cebe9998a386323206afb0bc08a19eea74253299e75dc5556e6"
+  },
+  {
+    "instrument_id": "PAY0373",
+    "type": "card",
+    "last4_hash": "83d7ce01060ca9e9dc54af6c215800d77b11bad522966f80acd79deddb36f73b"
+  },
+  {
+    "instrument_id": "PAY0374",
+    "type": "upi",
+    "last4_hash": "ce9609640075f8b7cd58aa69194024559a1a1efbea6234486fac17f6c5476d6d"
+  },
+  {
+    "instrument_id": "PAY0375",
+    "type": "card",
+    "last4_hash": "ba733a2bc3efb152c8d435ef1d7640160090d996a866e1b4970f233d19371edb"
+  },
+  {
+    "instrument_id": "PAY0376",
+    "type": "card",
+    "last4_hash": "dcbff8f66de95d7c6148b3fbb3a9934d226ffb6dfd405c8394ae5454dc87d348"
+  },
+  {
+    "instrument_id": "PAY0377",
+    "type": "upi",
+    "last4_hash": "f8061506e4c8bc3fab9a2eebb988cd6f52f14a4ee0b3f5c1d739fb88750d4036"
+  },
+  {
+    "instrument_id": "PAY0378",
+    "type": "upi",
+    "last4_hash": "c4210996b202fac099a1603ed8c708528a08bf3fda87d3754b552653b717a5cf"
+  },
+  {
+    "instrument_id": "PAY0379",
+    "type": "upi",
+    "last4_hash": "ce9cb7fd6836d4eb0cc662b98193a5d99e1d4cd611143b5f00c5c4062471aa91"
+  },
+  {
+    "instrument_id": "PAY0380",
+    "type": "upi",
+    "last4_hash": "8c2962f6101b265bf3420df868fd8a06c225af08e9121dc5f6b14675a5143430"
+  },
+  {
+    "instrument_id": "PAY0381",
+    "type": "upi",
+    "last4_hash": "90bdbc041092b051ce60a757fbbeacce2413d02e9aa8394341f26a69a5df36fa"
+  },
+  {
+    "instrument_id": "PAY0382",
+    "type": "upi",
+    "last4_hash": "f05bf7a9ac195b334d23d13b00567122f67958d56bd92ac804f9b17a757f9c74"
+  },
+  {
+    "instrument_id": "PAY0383",
+    "type": "upi",
+    "last4_hash": "c5fb52368f1b210a540e1cffb745ec44c44c268cecb39c2096c3553fa0e93231"
+  },
+  {
+    "instrument_id": "PAY0384",
+    "type": "upi",
+    "last4_hash": "00eb14b679e5b17abc30ee049f64e324c36f2f279d29d725ce4785fdd73a340e"
+  },
+  {
+    "instrument_id": "PAY0385",
+    "type": "upi",
+    "last4_hash": "4ade1343ca03e40d753469311da59756151e8757e013b54af5c65c13d3b224c8"
+  },
+  {
+    "instrument_id": "PAY0386",
+    "type": "upi",
+    "last4_hash": "689d840a91eed6a0aa9ebb341fe2dde3405981143f991f52234d2c31b4519875"
+  }
+]

data/synthetic/scored_clusters.json

@@ -0,0 +1,812 @@
+[
+  {
+    "cluster_id": "CAND01",
+    "customer_ids": [
+      "CUST0041",
+      "CUST0042",
+      "CUST0043",
+      "CUST0044"
+    ],
+    "size": 4,
+    "strong_link_count": 6,
+    "weak_link_count": 0,
+    "formation_type": "strong_anchored",
+    "method": "connected_components",
+    "risk_probability": 0.9958773824170732
+  },
+  {
+    "cluster_id": "CAND02",
+    "customer_ids": [
+      "CUST0009",
+      "CUST0010",
+      "CUST0011",
+      "CUST0012",
+      "CUST0013"
+    ],
+    "size": 5,
+    "strong_link_count": 10,
+    "weak_link_count": 0,
+    "formation_type": "strong_anchored",
+    "method": "connected_components",
+    "risk_probability": 0.9968053865328341
+  },
+  {
+    "cluster_id": "CAND03",
+    "customer_ids": [
+      "CUST0001",
+      "CUST0002",
+      "CUST0003"
+    ],
+    "size": 3,
+    "strong_link_count": 3,
+    "weak_link_count": 0,
+    "formation_type": "strong_anchored",
+    "method": "connected_components",
+    "risk_probability": 0.9950028638127228
+  },
+  {
+    "cluster_id": "CAND04",
+    "customer_ids": [
+      "CUST0014",
+      "CUST0015",
+      "CUST0016",
+      "CUST0017",
+      "CUST0018"
+    ],
+    "size": 5,
+    "strong_link_count": 10,
+    "weak_link_count": 0,
+    "formation_type": "strong_anchored",
+    "method": "connected_components",
+    "risk_probability": 0.9955003588432927
+  },
+  {
+    "cluster_id": "CAND05",
+    "customer_ids": [
+      "CUST0033",
+      "CUST0034",
+      "CUST0035"
+    ],
+    "size": 3,
+    "strong_link_count": 3,
+    "weak_link_count": 0,
+    "formation_type": "strong_anchored",
+    "method": "connected_components",
+    "risk_probability": 0.9969893801631614
+  },
+  {
+    "cluster_id": "CAND06",
+    "customer_ids": [
+      "CUST0036",
+      "CUST0037",
+      "CUST0038",
+      "CUST0039",
+      "CUST0040"
+    ],
+    "size": 5,
+    "strong_link_count": 10,
+    "weak_link_count": 0,
+    "formation_type": "strong_anchored",
+    "method": "connected_components",
+    "risk_probability": 0.9969377389203431
+  },
+  {
+    "cluster_id": "CAND07",
+    "customer_ids": [
+      "CUST0019",
+      "CUST0020",
+      "CUST0021",
+      "CUST0022",
+      "CUST0023"
+    ],
+    "size": 5,
+    "strong_link_count": 10,
+    "weak_link_count": 0,
+    "formation_type": "strong_anchored",
+    "method": "connected_components",
+    "risk_probability": 0.9940850384979173
+  },
+  {
+    "cluster_id": "CAND08",
+    "customer_ids": [
+      "CUST0501",
+      "CUST0502",
+      "CUST0503"
+    ],
+    "size": 3,
+    "strong_link_count": 3,
+    "weak_link_count": 0,
+    "formation_type": "strong_anchored",
+    "method": "connected_components",
+    "risk_probability": 0.9867466290228035
+  },
+  {
+    "cluster_id": "CAND09",
+    "customer_ids": [
+      "CUST0024",
+      "CUST0025",
+      "CUST0026",
+      "CUST0027",
+      "CUST0028",
+      "CUST0029"
+    ],
+    "size": 6,
+    "strong_link_count": 15,
+    "weak_link_count": 0,
+    "formation_type": "strong_anchored",
+    "method": "connected_components",
+    "risk_probability": 0.9913203009684713
+  },
+  {
+    "cluster_id": "CAND10",
+    "customer_ids": [
+      "CUST0004",
+      "CUST0005",
+      "CUST0006",
+      "CUST0007",
+      "CUST0008"
+    ],
+    "size": 5,
+    "strong_link_count": 7,
+    "weak_link_count": 0,
+    "formation_type": "strong_anchored",
+    "method": "connected_components",
+    "risk_probability": 0.9959231873282086
+  },
+  {
+    "cluster_id": "CAND11",
+    "customer_ids": [
+      "CUST0030",
+      "CUST0031",
+      "CUST0032"
+    ],
+    "size": 3,
+    "strong_link_count": 2,
+    "weak_link_count": 0,
+    "formation_type": "strong_anchored",
+    "method": "connected_components",
+    "risk_probability": 0.9927212653417073
+  },
+  {
+    "cluster_id": "CAND12",
+    "customer_ids": [
+      "CUST0001",
+      "CUST0019",
+      "CUST0040",
+      "CUST0126",
+      "CUST0138",
+      "CUST0168",
+      "CUST0211",
+      "CUST0219",
+      "CUST0258",
+      "CUST0281",
+      "CUST0292",
+      "CUST0322",
+      "CUST0325",
+      "CUST0371",
+      "CUST0373",
+      "CUST0400",
+      "CUST0419",
+      "CUST0465"
+    ],
+    "size": 18,
+    "strong_link_count": 0,
+    "weak_link_count": 41,
+    "formation_type": "weak_only",
+    "method": "louvain",
+    "risk_probability": 0.00720344453548541
+  },
+  {
+    "cluster_id": "CAND13",
+    "customer_ids": [
+      "CUST0002",
+      "CUST0009",
+      "CUST0020",
+      "CUST0041",
+      "CUST0064",
+      "CUST0073",
+      "CUST0074",
+      "CUST0080",
+      "CUST0081",
+      "CUST0086",
+      "CUST0089",
+      "CUST0103",
+      "CUST0162",
+      "CUST0167",
+      "CUST0173",
+      "CUST0181",
+      "CUST0226",
+      "CUST0240",
+      "CUST0253",
+      "CUST0262",
+      "CUST0272",
+      "CUST0286",
+      "CUST0287",
+      "CUST0324",
+      "CUST0331",
+      "CUST0333",
+      "CUST0369",
+      "CUST0392",
+      "CUST0393",
+      "CUST0398",
+      "CUST0406",
+      "CUST0410",
+      "CUST0411",
+      "CUST0413",
+      "CUST0415",
+      "CUST0423",
+      "CUST0446",
+      "CUST0468",
+      "CUST0476",
+      "CUST0478",
+      "CUST0485"
+    ],
+    "size": 41,
+    "strong_link_count": 0,
+    "weak_link_count": 146,
+    "formation_type": "weak_only",
+    "method": "louvain",
+    "risk_probability": 0.007559966037372132
+  },
+  {
+    "cluster_id": "CAND14",
+    "customer_ids": [
+      "CUST0003",
+      "CUST0012",
+      "CUST0043",
+      "CUST0077",
+      "CUST0088",
+      "CUST0098",
+      "CUST0109",
+      "CUST0125",
+      "CUST0139",
+      "CUST0153",
+      "CUST0159",
+      "CUST0166",
+      "CUST0180",
+      "CUST0201",
+      "CUST0209",
+      "CUST0245",
+      "CUST0261",
+      "CUST0265",
+      "CUST0273",
+      "CUST0277",
+      "CUST0297",
+      "CUST0315",
+      "CUST0334",
+      "CUST0347",
+      "CUST0381",
+      "CUST0405",
+      "CUST0412",
+      "CUST0428",
+      "CUST0429",
+      "CUST0462",
+      "CUST0471"
+    ],
+    "size": 31,
+    "strong_link_count": 0,
+    "weak_link_count": 84,
+    "formation_type": "weak_only",
+    "method": "louvain",
+    "risk_probability": 0.0033586500336385537
+  },
+  {
+    "cluster_id": "CAND15",
+    "customer_ids": [
+      "CUST0004",
+      "CUST0008",
+      "CUST0027",
+      "CUST0030",
+      "CUST0042",
+      "CUST0063",
+      "CUST0075",
+      "CUST0087",
+      "CUST0110",
+      "CUST0117",
+      "CUST0120",
+      "CUST0133",
+      "CUST0135",
+      "CUST0137",
+      "CUST0144",
+      "CUST0152",
+      "CUST0154",
+      "CUST0163",
+      "CUST0170",
+      "CUST0178",
+      "CUST0179",
+      "CUST0195",
+      "CUST0198",
+      "CUST0208",
+      "CUST0210",
+      "CUST0221",
+      "CUST0228",
+      "CUST0231",
+      "CUST0237",
+      "CUST0246",
+      "CUST0251",
+      "CUST0254",
+      "CUST0266",
+      "CUST0282",
+      "CUST0295",
+      "CUST0300",
+      "CUST0303",
+      "CUST0304",
+      "CUST0335",
+      "CUST0355",
+      "CUST0365",
+      "CUST0399",
+      "CUST0420",
+      "CUST0425",
+      "CUST0470",
+      "CUST0483",
+      "CUST0486",
+      "CUST0487",
+      "CUST0491",
+      "CUST0492",
+      "CUST0498"
+    ],
+    "size": 51,
+    "strong_link_count": 1,
+    "weak_link_count": 167,
+    "formation_type": "weak_only",
+    "method": "louvain",
+    "risk_probability": 0.007646314104742993
+  },
+  {
+    "cluster_id": "CAND16",
+    "customer_ids": [
+      "CUST0005",
+      "CUST0022",
+      "CUST0028",
+      "CUST0038",
+      "CUST0102",
+      "CUST0111",
+      "CUST0113",
+      "CUST0140",
+      "CUST0161",
+      "CUST0171",
+      "CUST0177",
+      "CUST0202",
+      "CUST0222",
+      "CUST0235",
+      "CUST0252",
+      "CUST0257",
+      "CUST0263",
+      "CUST0288",
+      "CUST0293",
+      "CUST0299",
+      "CUST0308",
+      "CUST0311",
+      "CUST0353",
+      "CUST0357",
+      "CUST0364",
+      "CUST0370",
+      "CUST0383",
+      "CUST0386",
+      "CUST0387",
+      "CUST0391",
+      "CUST0396",
+      "CUST0438",
+      "CUST0442",
+      "CUST0456",
+      "CUST0460",
+      "CUST0488",
+      "CUST0490",
+      "CUST0495",
+      "CUST0496",
+      "CUST0500"
+    ],
+    "size": 40,
+    "strong_link_count": 0,
+    "weak_link_count": 112,
+    "formation_type": "weak_only",
+    "method": "louvain",
+    "risk_probability": 0.0035456513474297945
+  },
+  {
+    "cluster_id": "CAND17",
+    "customer_ids": [
+      "CUST0006",
+      "CUST0011",
+      "CUST0024",
+      "CUST0025",
+      "CUST0033",
+      "CUST0044",
+      "CUST0079",
+      "CUST0084",
+      "CUST0101",
+      "CUST0107",
+      "CUST0108",
+      "CUST0114",
+      "CUST0132",
+      "CUST0145",
+      "CUST0157",
+      "CUST0172",
+      "CUST0227",
+      "CUST0244",
+      "CUST0255",
+      "CUST0256",
+      "CUST0267",
+      "CUST0271",
+      "CUST0284",
+      "CUST0291",
+      "CUST0296",
+      "CUST0316",
+      "CUST0319",
+      "CUST0329",
+      "CUST0336",
+      "CUST0342",
+      "CUST0348",
+      "CUST0349",
+      "CUST0367",
+      "CUST0375",
+      "CUST0389",
+      "CUST0407",
+      "CUST0455",
+      "CUST0489",
+      "CUST0499"
+    ],
+    "size": 39,
+    "strong_link_count": 1,
+    "weak_link_count": 95,
+    "formation_type": "weak_only",
+    "method": "louvain",
+    "risk_probability": 0.006896318028648148
+  },
+  {
+    "cluster_id": "CAND18",
+    "customer_ids": [
+      "CUST0007",
+      "CUST0014",
+      "CUST0015",
+      "CUST0016",
+      "CUST0032",
+      "CUST0037",
+      "CUST0065",
+      "CUST0069",
+      "CUST0070",
+      "CUST0095",
+      "CUST0106",
+      "CUST0115",
+      "CUST0124",
+      "CUST0131",
+      "CUST0149",
+      "CUST0191",
+      "CUST0200",
+      "CUST0204",
+      "CUST0215",
+      "CUST0218",
+      "CUST0223",
+      "CUST0238",
+      "CUST0239",
+      "CUST0241",
+      "CUST0242",
+      "CUST0264",
+      "CUST0276",
+      "CUST0280",
+      "CUST0305",
+      "CUST0314",
+      "CUST0323",
+      "CUST0327",
+      "CUST0339",
+      "CUST0350",
+      "CUST0352",
+      "CUST0356",
+      "CUST0358",
+      "CUST0362",
+      "CUST0379",
+      "CUST0401",
+      "CUST0418",
+      "CUST0440",
+      "CUST0443",
+      "CUST0449",
+      "CUST0479"
+    ],
+    "size": 45,
+    "strong_link_count": 3,
+    "weak_link_count": 106,
+    "formation_type": "weak_only",
+    "method": "louvain",
+    "risk_probability": 0.0034784417949967547
+  },
+  {
+    "cluster_id": "CAND19",
+    "customer_ids": [
+      "CUST0010",
+      "CUST0023",
+      "CUST0029",
+      "CUST0090",
+      "CUST0119",
+      "CUST0121",
+      "CUST0129",
+      "CUST0141",
+      "CUST0143",
+      "CUST0155",
+      "CUST0196",
+      "CUST0207",
+      "CUST0230",
+      "CUST0233",
+      "CUST0247",
+      "CUST0330",
+      "CUST0338",
+      "CUST0372",
+      "CUST0374",
+      "CUST0380",
+      "CUST0461",
+      "CUST0467",
+      "CUST0480",
+      "CUST0493"
+    ],
+    "size": 24,
+    "strong_link_count": 0,
+    "weak_link_count": 47,
+    "formation_type": "weak_only",
+    "method": "louvain",
+    "risk_probability": 0.007559966037372132
+  },
+  {
+    "cluster_id": "CAND20",
+    "customer_ids": [
+      "CUST0013",
+      "CUST0026",
+      "CUST0031",
+      "CUST0035",
+      "CUST0039",
+      "CUST0067",
+      "CUST0068",
+      "CUST0094",
+      "CUST0116",
+      "CUST0122",
+      "CUST0128",
+      "CUST0184",
+      "CUST0193",
+      "CUST0203",
+      "CUST0212",
+      "CUST0217",
+      "CUST0248",
+      "CUST0268",
+      "CUST0283",
+      "CUST0312",
+      "CUST0340",
+      "CUST0351",
+      "CUST0384",
+      "CUST0402",
+      "CUST0417",
+      "CUST0427",
+      "CUST0430",
+      "CUST0434",
+      "CUST0451",
+      "CUST0463",
+      "CUST0466",
+      "CUST0469",
+      "CUST0473",
+      "CUST0482",
+      "CUST0497"
+    ],
+    "size": 35,
+    "strong_link_count": 0,
+    "weak_link_count": 84,
+    "formation_type": "weak_only",
+    "method": "louvain",
+    "risk_probability": 0.007561351403107364
+  },
+  {
+    "cluster_id": "CAND21",
+    "customer_ids": [
+      "CUST0017",
+      "CUST0036",
+      "CUST0062",
+      "CUST0071",
+      "CUST0082",
+      "CUST0083",
+      "CUST0085",
+      "CUST0096",
+      "CUST0123",
+      "CUST0142",
+      "CUST0147",
+      "CUST0175",
+      "CUST0189",
+      "CUST0216",
+      "CUST0220",
+      "CUST0224",
+      "CUST0232",
+      "CUST0234",
+      "CUST0306",
+      "CUST0328",
+      "CUST0337",
+      "CUST0341",
+      "CUST0345",
+      "CUST0346",
+      "CUST0361",
+      "CUST0388",
+      "CUST0421",
+      "CUST0426",
+      "CUST0433",
+      "CUST0435",
+      "CUST0444",
+      "CUST0450",
+      "CUST0454",
+      "CUST0477"
+    ],
+    "size": 34,
+    "strong_link_count": 0,
+    "weak_link_count": 79,
+    "formation_type": "weak_only",
+    "method": "louvain",
+    "risk_probability": 0.0035456513474297945
+  },
+  {
+    "cluster_id": "CAND22",
+    "customer_ids": [
+      "CUST0018",
+      "CUST0021",
+      "CUST0091",
+      "CUST0092",
+      "CUST0112",
+      "CUST0160",
+      "CUST0174",
+      "CUST0187",
+      "CUST0270",
+      "CUST0298",
+      "CUST0307",
+      "CUST0310",
+      "CUST0332",
+      "CUST0343",
+      "CUST0431",
+      "CUST0474",
+      "CUST0475",
+      "CUST0494"
+    ],
+    "size": 18,
+    "strong_link_count": 0,
+    "weak_link_count": 35,
+    "formation_type": "weak_only",
+    "method": "louvain",
+    "risk_probability": 0.03305941612108949
+  },
+  {
+    "cluster_id": "CAND23",
+    "customer_ids": [
+      "CUST0034",
+      "CUST0066",
+      "CUST0078",
+      "CUST0100",
+      "CUST0104",
+      "CUST0134",
+      "CUST0146",
+      "CUST0150",
+      "CUST0151",
+      "CUST0158",
+      "CUST0182",
+      "CUST0183",
+      "CUST0185",
+      "CUST0188",
+      "CUST0190",
+      "CUST0259",
+      "CUST0269",
+      "CUST0279",
+      "CUST0285",
+      "CUST0289",
+      "CUST0301",
+      "CUST0302",
+      "CUST0344",
+      "CUST0354",
+      "CUST0366",
+      "CUST0368",
+      "CUST0378",
+      "CUST0404",
+      "CUST0408",
+      "CUST0432",
+      "CUST0436",
+      "CUST0437",
+      "CUST0439",
+      "CUST0441",
+      "CUST0447",
+      "CUST0448",
+      "CUST0457",
+      "CUST0481"
+    ],
+    "size": 38,
+    "strong_link_count": 0,
+    "weak_link_count": 90,
+    "formation_type": "weak_only",
+    "method": "louvain",
+    "risk_probability": 0.0035456513474297945
+  },
+  {
+    "cluster_id": "CAND24",
+    "customer_ids": [
+      "CUST0061",
+      "CUST0093",
+      "CUST0105",
+      "CUST0130",
+      "CUST0136",
+      "CUST0148",
+      "CUST0164",
+      "CUST0186",
+      "CUST0197",
+      "CUST0205",
+      "CUST0206",
+      "CUST0214",
+      "CUST0260",
+      "CUST0309",
+      "CUST0320",
+      "CUST0321",
+      "CUST0359",
+      "CUST0363",
+      "CUST0376",
+      "CUST0377",
+      "CUST0385",
+      "CUST0394",
+      "CUST0403",
+      "CUST0416",
+      "CUST0424",
+      "CUST0453"
+    ],
+    "size": 26,
+    "strong_link_count": 0,
+    "weak_link_count": 71,
+    "formation_type": "weak_only",
+    "method": "louvain",
+    "risk_probability": 0.0035456513474297945
+  },
+  {
+    "cluster_id": "CAND25",
+    "customer_ids": [
+      "CUST0045",
+      "CUST0046",
+      "CUST0047",
+      "CUST0048"
+    ],
+    "size": 4,
+    "strong_link_count": 0,
+    "weak_link_count": 6,
+    "formation_type": "weak_only",
+    "method": "connected_components",
+    "risk_probability": 0.0035484719416527992
+  },
+  {
+    "cluster_id": "CAND26",
+    "customer_ids": [
+      "CUST0049",
+      "CUST0050",
+      "CUST0051",
+      "CUST0052"
+    ],
+    "size": 4,
+    "strong_link_count": 0,
+    "weak_link_count": 6,
+    "formation_type": "weak_only",
+    "method": "connected_components",
+    "risk_probability": 0.0031715868230902835
+  },
+  {
+    "cluster_id": "CAND27",
+    "customer_ids": [
+      "CUST0053",
+      "CUST0054",
+      "CUST0055",
+      "CUST0056"
+    ],
+    "size": 4,
+    "strong_link_count": 0,
+    "weak_link_count": 6,
+    "formation_type": "weak_only",
+    "method": "connected_components",
+    "risk_probability": 0.003058997865465553
+  },
+  {
+    "cluster_id": "CAND28",
+    "customer_ids": [
+      "CUST0504",
+      "CUST0505",
+      "CUST0506"
+    ],
+    "size": 3,
+    "strong_link_count": 0,
+    "weak_link_count": 3,
+    "formation_type": "weak_only",
+    "method": "connected_components",
+    "risk_probability": 0.003546939654202544
+  }
+]

eval/.gitkeep

@@ -0,0 +1 @@
+

eval/eval_results.json

@@ -0,0 +1,24 @@
+{
+  "precision": 1.0,
+  "recall": 1.0,
+  "f1": 1.0,
+  "classification_threshold": 0.5,
+  "holdout_size": 22,
+  "holdout_positives": 12,
+  "holdout_negatives": 10,
+  "false_positive_cost_inr": 0.0,
+  "false_positive_clusters": [],
+  "latency_total_ms": 0.9449999997741543,
+  "latency_per_cluster_ms": 0.04295454544427974,
+  "policy_action_breakdown": {
+    "auto_hold": 0,
+    "escalate": 12,
+    "log_only": 10
+  },
+  "confusion_matrix": {
+    "true_positive": 12,
+    "false_positive": 0,
+    "true_negative": 10,
+    "false_negative": 0
+  }
+}

eval/evaluate.py

@@ -0,0 +1,153 @@
+"""
+evaluate.py
+-----------
+Evaluate the trained LightGBM model on the held-out split.
+
+Outputs:
+  eval/eval_results.json — precision, recall, F1, false-positive cost,
+                            latency, and policy-action breakdown.
+
+Usage:
+    python eval/evaluate.py
+"""
+
+import json
+import os
+import sys
+import time
+
+import lightgbm as lgb
+import numpy as np
+from sklearn.metrics import precision_score, recall_score, f1_score
+
+BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+sys.path.insert(0, BASE_DIR)
+
+SCORING_DIR = os.path.join(BASE_DIR, "scoring")
+EVAL_DIR    = os.path.join(BASE_DIR, "eval")
+
+MODEL_PATH   = os.path.join(SCORING_DIR, "model.lgb")
+HOLDOUT_PATH = os.path.join(SCORING_DIR, "holdout_split.json")
+RESULTS_PATH = os.path.join(EVAL_DIR, "eval_results.json")
+
+CLASSIFICATION_THRESHOLD = 0.5
+
+# Policy thresholds (must match agent/policy.py exactly)
+POLICY_HIGH_THRESHOLD    = 0.85
+POLICY_MEDIUM_THRESHOLD  = 0.50
+POLICY_LOW_EXPOSURE      = 5000.0
+
+
+def policy_decide(risk_probability: float, exposure_rupees: float) -> str:
+    """Deterministic policy gate — mirrors agent/policy.py exactly."""
+    if risk_probability >= POLICY_HIGH_THRESHOLD and exposure_rupees <= POLICY_LOW_EXPOSURE:
+        return "auto_hold"
+    if risk_probability >= POLICY_HIGH_THRESHOLD and exposure_rupees > POLICY_LOW_EXPOSURE:
+        return "escalate"
+    if POLICY_MEDIUM_THRESHOLD <= risk_probability < POLICY_HIGH_THRESHOLD:
+        return "escalate"
+    return "log_only"
+
+
+def evaluate():
+    print(f"Loading model from {MODEL_PATH}...")
+    model = lgb.Booster(model_file=MODEL_PATH)
+
+    print(f"Loading holdout split from {HOLDOUT_PATH}...")
+    with open(HOLDOUT_PATH, encoding="utf-8") as f:
+        holdout = json.load(f)
+
+    print(f"  {len(holdout)} held-out clusters")
+
+    # 6.1 — reconstruct feature matrix
+    from scoring.features import FEATURE_COLUMNS
+
+    feature_rows = [row["features"] for row in holdout]
+    y_true = np.array([row["label"] for row in holdout])
+
+    # 6.2 — time the feature extraction + inference pass
+    start = time.perf_counter()
+    X = np.array([[fr[f] for f in FEATURE_COLUMNS] for fr in feature_rows])
+    probs = model.predict(X)
+    elapsed = time.perf_counter() - start
+
+    y_pred = (probs >= CLASSIFICATION_THRESHOLD).astype(int)
+
+    precision = float(precision_score(y_true, y_pred, zero_division=0))
+    recall    = float(recall_score(y_true, y_pred, zero_division=0))
+    f1        = float(f1_score(y_true, y_pred, zero_division=0))
+    latency_total_ms   = elapsed * 1000
+    latency_per_cluster_ms = latency_total_ms / len(holdout)
+
+    print(f"\n--- Classification at threshold {CLASSIFICATION_THRESHOLD} ---")
+    print(f"  Precision : {precision:.4f}")
+    print(f"  Recall    : {recall:.4f}")
+    print(f"  F1        : {f1:.4f}")
+    print(f"  Latency   : {latency_total_ms:.2f} ms total, "
+          f"{latency_per_cluster_ms:.3f} ms/cluster")
+
+    # 6.3 — policy simulation on holdout
+    policy_counts = {"auto_hold": 0, "escalate": 0, "log_only": 0}
+    for i, row in enumerate(holdout):
+        prob = float(probs[i])
+        exposure = row["features"].get("total_amount", 0.0)
+        decision = policy_decide(prob, exposure)
+        policy_counts[decision] += 1
+
+    print(f"\n--- Policy simulation on holdout ---")
+    for action, count in policy_counts.items():
+        print(f"  {action:12s}: {count}")
+
+    # 6.4 — false-positive cost
+    fp_cost = 0.0
+    fp_clusters = []
+    for i, row in enumerate(holdout):
+        if y_pred[i] == 1 and y_true[i] == 0:
+            amt = row["features"].get("total_amount", 0.0)
+            fp_cost += amt
+            fp_clusters.append({
+                "cluster_id": row["cluster_id"],
+                "total_amount": amt,
+                "risk_probability": float(probs[i]),
+            })
+
+    print(f"\n--- False-positive cost ---")
+    print(f"  FP clusters: {len(fp_clusters)}")
+    print(f"  FP cost (total_amount): Rs.{fp_cost:.2f}")
+
+    # 6.5 — latency already computed above
+    print(f"\n--- Latency ---")
+    print(f"  Feature extraction + inference: {latency_per_cluster_ms:.3f} ms/cluster")
+
+    # 6.6 — write eval_results.json
+    os.makedirs(EVAL_DIR, exist_ok=True)
+    results = {
+        "precision": precision,
+        "recall": recall,
+        "f1": f1,
+        "classification_threshold": CLASSIFICATION_THRESHOLD,
+        "holdout_size": len(holdout),
+        "holdout_positives": int(y_true.sum()),
+        "holdout_negatives": int((len(y_true) - y_true.sum())),
+        "false_positive_cost_inr": fp_cost,
+        "false_positive_clusters": fp_clusters,
+        "latency_total_ms": latency_total_ms,
+        "latency_per_cluster_ms": latency_per_cluster_ms,
+        "policy_action_breakdown": policy_counts,
+        "confusion_matrix": {
+            "true_positive": int(((y_pred == 1) & (y_true == 1)).sum()),
+            "false_positive": int(((y_pred == 1) & (y_true == 0)).sum()),
+            "true_negative": int(((y_pred == 0) & (y_true == 0)).sum()),
+            "false_negative": int(((y_pred == 0) & (y_true == 1)).sum()),
+        },
+    }
+
+    with open(RESULTS_PATH, "w", encoding="utf-8") as f:
+        json.dump(results, f, indent=2, ensure_ascii=False)
+    print(f"\nEvaluation results saved to {RESULTS_PATH}")
+
+    return results
+
+
+if __name__ == "__main__":
+    evaluate()

graph/.gitkeep

@@ -0,0 +1 @@
+

graph/build_graph.py

@@ -0,0 +1,299 @@
+"""
+build_graph.py
+--------------
+Loads synthetic data from a data directory, applies entity resolution,
+builds a heterogeneous NetworkX graph, and saves it to graph.gpickle
+in the same directory.
+
+Node types : customer, device, address, payment_instrument
+Edge types  : uses_device (strong), pays_with (strong),
+              ships_to (weak), same_contact (strong)
+
+Each edge carries: relation, strength, first_seen, last_seen, count, order_ids
+
+Usage:
+    python graph/build_graph.py [--data-dir PATH]
+    DATA_DIR defaults to data/synthetic/
+"""
+
+import argparse
+import json
+import os
+import pickle
+import re
+from collections import defaultdict
+
+import networkx as nx
+
+# ---------------------------------------------------------------------------
+# Paths
+# ---------------------------------------------------------------------------
+BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+
+
+def _load(filename, data_dir):
+    with open(os.path.join(data_dir, filename), encoding="utf-8") as f:
+        return json.load(f)
+
+
+# ---------------------------------------------------------------------------
+# Entity resolution helpers (3.1 – 3.4)
+# ---------------------------------------------------------------------------
+
+def normalize_phone(raw: str) -> str:
+    """Strip non-digits; drop a leading country code (91 for India, 1 for US)."""
+    digits = re.sub(r"\D", "", raw)
+    # Drop leading 91 (India) or 1 (US) if the result would be 10 digits
+    for prefix in ("91", "1"):
+        if digits.startswith(prefix) and len(digits) - len(prefix) == 10:
+            digits = digits[len(prefix):]
+            break
+    return digits
+
+
+def normalize_email(raw: str) -> str:
+    """Lowercase and strip whitespace."""
+    return raw.lower().strip()
+
+
+def normalize_address(raw: str) -> str:
+    """Lowercase, strip whitespace, collapse multiple spaces."""
+    return re.sub(r"\s+", " ", raw.lower().strip())
+
+
+# ---------------------------------------------------------------------------
+# Graph construction
+# ---------------------------------------------------------------------------
+
+def build_graph(data_dir: str = None) -> nx.Graph:
+    if data_dir is None:
+        data_dir = os.path.join(BASE_DIR, "data", "synthetic")
+    # Load data
+    customers         = _load("customers.json", data_dir)
+    devices           = _load("devices.json", data_dir)
+    addresses         = _load("addresses.json", data_dir)
+    payment_instr     = _load("payment_instruments.json", data_dir)
+    orders            = _load("orders.json", data_dir)
+
+    # Apply entity resolution (normalizations run on load — see ground rules 3.4)
+    # For the synthetic data these are effectively no-ops since fields are already
+    # hashed/normalized, but the hooks are live for real-data use.
+    for c in customers:
+        # phone_hash and email_hash are already SHA256 digests — we normalize
+        # the raw strings before hashing in a real pipeline; here we just ensure
+        # the lookup maps are keyed consistently.
+        pass  # hash fields are already canonical
+
+    # Build lookup maps
+    device_map    = {d["device_id"]: d for d in devices}
+    addr_map      = {a["address_id"]: a for a in addresses}
+    instr_map     = {p["instrument_id"]: p for p in payment_instr}
+    cust_map      = {c["customer_id"]: c for c in customers}
+
+    G = nx.Graph()
+
+    # ---------------------------------------------------------------------------
+    # Add all nodes with their type attribute (3.5)
+    # ---------------------------------------------------------------------------
+    for c in customers:
+        G.add_node(c["customer_id"], node_type="customer", **{
+            k: v for k, v in c.items() if k != "customer_id"
+        })
+
+    for d in devices:
+        G.add_node(d["device_id"], node_type="device", **{
+            k: v for k, v in d.items() if k != "device_id"
+        })
+
+    for a in addresses:
+        G.add_node(a["address_id"], node_type="address", **{
+            k: v for k, v in a.items() if k != "address_id"
+        })
+
+    for p in payment_instr:
+        G.add_node(p["instrument_id"], node_type="payment_instrument", **{
+            k: v for k, v in p.items() if k != "instrument_id"
+        })
+
+    # ---------------------------------------------------------------------------
+    # Accumulate edge data from orders before writing to graph
+    # edge_accum[(u, v)] = {"relation":..., "strength":..., "timestamps":[...], "order_ids":[...]}
+    # ---------------------------------------------------------------------------
+    EdgeKey = tuple  # (node_a, node_b) always sorted so undirected
+    edge_accum: dict[EdgeKey, dict] = defaultdict(lambda: {
+        "relation": None, "strength": None, "timestamps": [], "order_ids": []
+    })
+
+    def _add_edge_data(u, v, relation, strength, timestamp, order_id):
+        key = (min(u, v), max(u, v))
+        ea = edge_accum[key]
+        ea["relation"] = relation
+        ea["strength"] = strength
+        ea["timestamps"].append(timestamp)
+        ea["order_ids"].append(order_id)
+
+    for order in orders:
+        cid  = order["customer_id"]
+        did  = order["device_id"]
+        iid  = order["payment_instrument_id"]
+        aid  = order["address_id"]
+        ts   = order["timestamp"]
+        oid  = order["order_id"]
+
+        # customer –uses– device (strong)
+        _add_edge_data(cid, did, "uses_device", "strong", ts, oid)
+
+        # customer –pays_with– payment_instrument (strong)
+        _add_edge_data(cid, iid, "pays_with", "strong", ts, oid)
+
+        # customer –ships_to– address (weak)
+        _add_edge_data(cid, aid, "ships_to", "weak", ts, oid)
+
+    # ---------------------------------------------------------------------------
+    # customer –same_contact– customer edges (3.6 last bullet)
+    # Two customers share the same email_hash OR phone_hash → strong link
+    # ---------------------------------------------------------------------------
+    email_to_custs: dict[str, list[str]] = defaultdict(list)
+    phone_to_custs: dict[str, list[str]] = defaultdict(list)
+
+    for c in customers:
+        email_to_custs[c["email_hash"]].append(c["customer_id"])
+        phone_to_custs[c["phone_hash"]].append(c["customer_id"])
+
+    same_contact_pairs: set[tuple] = set()
+    for group in list(email_to_custs.values()) + list(phone_to_custs.values()):
+        if len(group) > 1:
+            for i in range(len(group)):
+                for j in range(i + 1, len(group)):
+                    pair = (min(group[i], group[j]), max(group[i], group[j]))
+                    same_contact_pairs.add(pair)
+
+    # Use a synthetic timestamp for same_contact edges (first order timestamp of either cust)
+    cust_first_ts: dict[str, str] = {}
+    cust_orders: dict[str, list] = defaultdict(list)
+    for o in orders:
+        cust_orders[o["customer_id"]].append(o)
+    for cid, ords in cust_orders.items():
+        sorted_ts = sorted(ords, key=lambda x: x["timestamp"])
+        cust_first_ts[cid] = sorted_ts[0]["timestamp"]
+
+    for (ca, cb) in same_contact_pairs:
+        ts = min(cust_first_ts.get(ca, "2026-01-01T00:00:00"),
+                 cust_first_ts.get(cb, "2026-01-01T00:00:00"))
+        key = (ca, cb)
+        edge_accum[key]["relation"] = "same_contact"
+        edge_accum[key]["strength"] = "strong"
+        edge_accum[key]["timestamps"].append(ts)
+        edge_accum[key]["order_ids"].append("CONTACT_MATCH")
+
+    # ---------------------------------------------------------------------------
+    # Write all accumulated edges to the graph (3.7)
+    # ---------------------------------------------------------------------------
+    for (u, v), data in edge_accum.items():
+        sorted_ts = sorted(data["timestamps"])
+        G.add_edge(
+            u, v,
+            relation=data["relation"],
+            strength=data["strength"],
+            first_seen=sorted_ts[0] if sorted_ts else None,
+            last_seen=sorted_ts[-1] if sorted_ts else None,
+            count=len(data["order_ids"]),
+            order_ids=data["order_ids"],
+        )
+
+    return G
+
+
+# ---------------------------------------------------------------------------
+# Save graph (3.8) — try write_gpickle; fall back to pickle.dump if unavailable
+# ---------------------------------------------------------------------------
+def save_graph(G: nx.Graph, path: str):
+    try:
+        nx.write_gpickle(G, path)
+        return "nx.write_gpickle"
+    except AttributeError:
+        # networkx >= 3.0 removed write_gpickle — use pickle directly
+        with open(path, "wb") as f:
+            pickle.dump(G, f, protocol=pickle.HIGHEST_PROTOCOL)
+        return "pickle.dump (nx.write_gpickle unavailable in this nx version)"
+
+
+def load_graph(path: str) -> nx.Graph:
+    try:
+        return nx.read_gpickle(path)
+    except AttributeError:
+        with open(path, "rb") as f:
+            return pickle.load(f)
+
+
+# ---------------------------------------------------------------------------
+# Console stats (3.9)
+# ---------------------------------------------------------------------------
+def print_stats(G: nx.Graph):
+    from collections import Counter
+
+    node_type_counts = Counter(
+        data.get("node_type", "unknown")
+        for _, data in G.nodes(data=True)
+    )
+
+    edge_rel_counts: dict[str, dict] = defaultdict(lambda: {"strong": 0, "weak": 0, "total": 0})
+    for u, v, data in G.edges(data=True):
+        rel = data.get("relation", "unknown")
+        strength = data.get("strength", "unknown")
+        edge_rel_counts[rel][strength] = edge_rel_counts[rel].get(strength, 0) + 1
+        edge_rel_counts[rel]["total"] += 1
+
+    cust_degrees = [
+        d for n, d in G.degree()
+        if G.nodes[n].get("node_type") == "customer"
+    ]
+
+    print("\n========== Graph Statistics ==========")
+    print("Nodes by type:")
+    for ntype, cnt in sorted(node_type_counts.items()):
+        print(f"  {ntype:25s}: {cnt}")
+    print(f"  {'TOTAL':25s}: {G.number_of_nodes()}")
+
+    print("\nEdges by relation / strength:")
+    for rel, counts in sorted(edge_rel_counts.items()):
+        strong = counts.get("strong", 0)
+        weak   = counts.get("weak", 0)
+        total  = counts["total"]
+        print(f"  {rel:25s}: total={total:5d}  strong={strong:5d}  weak={weak:5d}")
+    print(f"  {'TOTAL':25s}: {G.number_of_edges()}")
+
+    if cust_degrees:
+        print(f"\nCustomer node degree — min={min(cust_degrees)}  "
+              f"max={max(cust_degrees)}  "
+              f"avg={sum(cust_degrees)/len(cust_degrees):.2f}")
+    print("======================================")
+
+
+# ---------------------------------------------------------------------------
+# Entry point
+# ---------------------------------------------------------------------------
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--data-dir",
+        default=None,
+        help="Path to data directory (containing JSON files). Defaults to data/synthetic/",
+    )
+    args = parser.parse_args()
+
+    data_dir = args.data_dir or os.path.join(BASE_DIR, "data", "synthetic")
+    print(f"Building graph from {data_dir} ...")
+    G = build_graph(data_dir)
+
+    out_path = os.path.join(data_dir, "graph.gpickle")
+    method = save_graph(G, out_path)
+    print(f"Graph saved to {out_path}  (method: {method})")
+
+    # Verify reload
+    G2 = load_graph(out_path)
+    assert G2.number_of_nodes() == G.number_of_nodes(), "Reload node count mismatch!"
+    assert G2.number_of_edges() == G.number_of_edges(), "Reload edge count mismatch!"
+    print("Reload verified OK.")
+
+    print_stats(G)

graph/detect_communities.py

@@ -0,0 +1,466 @@
+"""
+detect_communities.py
+---------------------
+Loads graph.gpickle, builds two separate customer-customer projections
+(strong-anchored and weak-only), runs community detection on each,
+filters candidate clusters, validates against ground truth, and writes
+candidate_clusters.json.
+
+Phase 2 fix: uses two separate connectivity passes instead of one blended
+weight projection, preventing incidental address-sharing from merging
+otherwise-unrelated rings into a single blob.
+
+Usage:
+    python graph/detect_communities.py [--data-dir DATA_DIR]
+
+    DATA_DIR defaults to data/synthetic/
+"""
+
+import argparse
+import json
+import os
+import pickle
+import sys
+from collections import defaultdict
+
+import community as community_louvain  # python-louvain
+import networkx as nx
+
+# ---------------------------------------------------------------------------
+# Paths (overridable via CLI so the same script serves both datasets)
+# ---------------------------------------------------------------------------
+BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+
+LOUVAIN_THRESHOLD = 12
+LOUVAIN_RESOLUTION = 1.0
+LOUVAIN_SEED = 42
+
+
+# ---------------------------------------------------------------------------
+# Load graph (handles both nx.read_gpickle and pickle.load fallback)
+# ---------------------------------------------------------------------------
+def load_graph(path: str) -> nx.Graph:
+    try:
+        return nx.read_gpickle(path)
+    except AttributeError:
+        with open(path, "rb") as f:
+            return pickle.load(f)
+
+
+# ---------------------------------------------------------------------------
+# 1.2  Build two customer-customer projections
+# ---------------------------------------------------------------------------
+
+# Strong edge formation threshold: a customer-entity-customer edge only counts
+# as a strong link if the customer has used that shared entity in >= 2 orders.
+# This filters out accidental single-order pool collisions (which would create
+# cross-ring contamination) while preserving intentional ring device sharing
+# (rings place 2-4 orders per member on their shared device).
+# same_contact edges are always strong regardless of order count.
+STRONG_EDGE_MIN_ORDERS = 2
+
+
+def _build_entity_customer_index(G: nx.Graph):
+    """
+    Returns three items:
+        strong_entity_to_custs: entity_id → list of (customer_id, order_count)
+                                for STRONG edges (uses_device / pays_with).
+        weak_entity_to_custs:   entity_id → list of customer_ids for WEAK edges.
+        same_contact_pairs:     list of (customer_a, customer_b) direct strong edges.
+    """
+    strong_entity_to_custs: dict[str, list[tuple[str, int]]] = defaultdict(list)
+    weak_entity_to_custs:   dict[str, list[str]] = defaultdict(list)
+    same_contact_pairs: list[tuple[str, str]] = []
+
+    for u, v, data in G.edges(data=True):
+        u_type = G.nodes[u].get("node_type", "")
+        v_type = G.nodes[v].get("node_type", "")
+
+        # same_contact is a direct customer-customer strong edge
+        if data.get("relation") == "same_contact":
+            if u_type == "customer" and v_type == "customer":
+                same_contact_pairs.append((u, v))
+            continue
+
+        # For customer-entity edges determine which customer and which entity
+        if u_type == "customer" and v_type != "customer":
+            cust, entity = u, v
+        elif v_type == "customer" and u_type != "customer":
+            cust, entity = v, u
+        else:
+            continue
+
+        strength = data.get("strength", "weak")
+        order_count = data.get("count", 1)
+        if strength == "strong":
+            strong_entity_to_custs[entity].append((cust, order_count))
+        else:
+            weak_entity_to_custs[entity].append(cust)
+
+    return strong_entity_to_custs, weak_entity_to_custs, same_contact_pairs
+
+
+def build_strong_subgraph(G: nx.Graph) -> nx.Graph:
+    """
+    Edge between two customers exists only if they share >= 1 STRONG link
+    (shared device used >= STRONG_EDGE_MIN_ORDERS times, shared payment instrument
+    used >= STRONG_EDGE_MIN_ORDERS times, or same_contact).
+    Edge weight = count of shared strong-link attributes.
+
+    The order-count threshold eliminates accidental single-order pool collisions
+    that would create false cross-ring bridges in the data.
+    """
+    strong_entity_to_custs, _, same_contact_pairs = _build_entity_customer_index(G)
+
+    all_customers = [n for n, d in G.nodes(data=True) if d.get("node_type") == "customer"]
+    SG = nx.Graph()
+    SG.add_nodes_from(all_customers)
+
+    # Shared entity edges — only for customers meeting the order-count threshold
+    for entity, custs_and_counts in strong_entity_to_custs.items():
+        eligible = [(c, cnt) for c, cnt in custs_and_counts if cnt >= STRONG_EDGE_MIN_ORDERS]
+        if len(eligible) < 2:
+            continue
+        for i in range(len(eligible)):
+            for j in range(i + 1, len(eligible)):
+                ca, _ = eligible[i]
+                cb, _ = eligible[j]
+                if ca == cb:
+                    continue
+                if SG.has_edge(ca, cb):
+                    SG[ca][cb]["weight"] += 1
+                    SG[ca][cb]["strong_share_count"] += 1
+                else:
+                    SG.add_edge(ca, cb, weight=1, strong_share_count=1, weak_share_count=0)
+
+    # same_contact direct edges (no order-count threshold — contact match is always strong)
+    for (ca, cb) in same_contact_pairs:
+        if SG.has_edge(ca, cb):
+            SG[ca][cb]["weight"] += 1
+            SG[ca][cb]["strong_share_count"] += 1
+        else:
+            SG.add_edge(ca, cb, weight=1, strong_share_count=1, weak_share_count=0)
+
+    return SG
+
+
+def build_weak_only_subgraph(G: nx.Graph, strong_subgraph: nx.Graph) -> nx.Graph:
+    """
+    Edge between two customers exists only if they share >= 1 WEAK link (address)
+    AND they share ZERO strong links (i.e., not already connected in strong_subgraph).
+    Edge weight = count of shared weak-link attributes.
+    """
+    _, weak_entity_to_custs, _ = _build_entity_customer_index(G)
+
+    all_customers = [n for n, d in G.nodes(data=True) if d.get("node_type") == "customer"]
+    WG = nx.Graph()
+    WG.add_nodes_from(all_customers)
+
+    for entity, custs in weak_entity_to_custs.items():
+        if len(custs) < 2:
+            continue
+        for i in range(len(custs)):
+            for j in range(i + 1, len(custs)):
+                ca, cb = custs[i], custs[j]
+                if ca == cb:
+                    continue
+                # Skip if the pair has any strong link
+                if strong_subgraph.has_edge(ca, cb):
+                    continue
+                if WG.has_edge(ca, cb):
+                    WG[ca][cb]["weight"] += 1
+                    WG[ca][cb]["weak_share_count"] += 1
+                else:
+                    WG.add_edge(ca, cb, weight=1, strong_share_count=0, weak_share_count=1)
+
+    return WG
+
+
+# ---------------------------------------------------------------------------
+# 1.3 + 1.4  Community detection on each subgraph
+# ---------------------------------------------------------------------------
+
+def detect_on_subgraph(subgraph: nx.Graph, formation_type: str) -> list[tuple[frozenset, str, str]]:
+    """
+    Returns list of (frozenset_of_customer_ids, method_string, formation_type).
+    """
+    connected_nodes = [n for n in subgraph.nodes() if subgraph.degree(n) > 0]
+    if not connected_nodes:
+        return []
+    CG_connected = subgraph.subgraph(connected_nodes)
+    components = list(nx.connected_components(CG_connected))
+    communities = []
+
+    for comp in components:
+        if len(comp) <= LOUVAIN_THRESHOLD:
+            communities.append((frozenset(comp), "connected_components", formation_type))
+        else:
+            sub = subgraph.subgraph(comp).copy()
+            partition = community_louvain.best_partition(
+                sub,
+                weight="weight",
+                resolution=LOUVAIN_RESOLUTION,
+                random_state=LOUVAIN_SEED,
+            )
+            sub_communities: dict[int, list[str]] = defaultdict(list)
+            for node, label in partition.items():
+                sub_communities[label].append(node)
+            for members in sub_communities.values():
+                communities.append((frozenset(members), "louvain", formation_type))
+
+    return communities
+
+
+# ---------------------------------------------------------------------------
+# Compute cluster metadata (strong_link_count + weak_link_count)
+# These are computed from BOTH subgraphs regardless of formation_type (1.7)
+# ---------------------------------------------------------------------------
+
+def compute_cluster_metadata(
+    cluster_nodes: frozenset,
+    strong_subgraph: nx.Graph,
+    weak_only_subgraph: nx.Graph,
+) -> dict:
+    """
+    strong_link_count: total strong-link edges within the cluster
+    weak_link_count:   total weak-link edges within the cluster
+    Both are computed regardless of formation_type (per spec 1.7).
+    """
+    strong = 0
+    weak = 0
+    members = list(cluster_nodes)
+    for i in range(len(members)):
+        for j in range(i + 1, len(members)):
+            ca, cb = members[i], members[j]
+            if strong_subgraph.has_edge(ca, cb):
+                strong += strong_subgraph[ca][cb].get("strong_share_count", 0)
+            # weak_only edges: only pairs not in strong_subgraph
+            if weak_only_subgraph.has_edge(ca, cb):
+                weak += weak_only_subgraph[ca][cb].get("weak_share_count", 0)
+
+    return {"strong_link_count": strong, "weak_link_count": weak}
+
+
+# ---------------------------------------------------------------------------
+# Write candidate_clusters.json
+# ---------------------------------------------------------------------------
+def write_clusters(clusters: list[dict], path: str):
+    os.makedirs(os.path.dirname(path), exist_ok=True)
+    with open(path, "w", encoding="utf-8") as f:
+        json.dump(clusters, f, indent=2, ensure_ascii=False)
+    print(f"Wrote {path} ({len(clusters)} clusters)")
+
+
+# ---------------------------------------------------------------------------
+# Ground-truth validation (works for both Phase 1 and Phase 2 datasets)
+# Phase 1 ground truth key: "stress_clusters"
+# Phase 2 ground truth key: "benign_clusters" (falls back to "stress_clusters")
+# ---------------------------------------------------------------------------
+def validate(clusters: list[dict], ground_truth_path: str):
+    with open(ground_truth_path, encoding="utf-8") as f:
+        gt = json.load(f)
+
+    rings = gt["rings"]
+    # Support both Phase 1 ("stress_clusters") and Phase 2 ("benign_clusters") naming
+    benign_clusters = gt.get("benign_clusters", gt.get("stress_clusters", []))
+
+    cand_sets = [set(c["customer_ids"]) for c in clusters]
+
+    # Ring recovery: each ring should appear in its OWN separate candidate cluster
+    rings_recovered = 0
+    ring_detail = []
+    for ring in rings:
+        ring_members = set(ring["customer_ids"])
+        ring_id = ring.get("ring_id", ring.get("cluster_id", "?"))
+        threshold = len(ring_members) * 0.5
+        best_overlap = 0
+        best_cand = None
+        for i, cs in enumerate(cand_sets):
+            overlap = len(ring_members & cs)
+            if overlap > best_overlap:
+                best_overlap = overlap
+                best_cand = clusters[i]["cluster_id"]
+        recovered = best_overlap >= threshold
+        if recovered:
+            rings_recovered += 1
+        ring_detail.append({
+            "ring_id": ring_id,
+            "ring_size": len(ring_members),
+            "best_overlap": best_overlap,
+            "recovered": recovered,
+            "matched_cluster": best_cand,
+        })
+
+    # Separation check: rings should be in SEPARATE candidates
+    # Count how many rings map to the same best candidate
+    matched_by_cand: dict[str, int] = defaultdict(int)
+    for d in ring_detail:
+        if d["matched_cluster"]:
+            matched_by_cand[d["matched_cluster"]] += 1
+    rings_sharing_cluster = sum(1 for cnt in matched_by_cand.values() if cnt > 1)
+
+    # Benign/stress cluster check: should appear as weak_only candidates
+    # with strong_link_count == 0
+    benign_detail = []
+    benign_correctly_weak = 0
+    for sc in benign_clusters:
+        sc_members = set(sc["customer_ids"])
+        sc_id = sc.get("cluster_id", "?")
+        threshold = len(sc_members) * 0.5
+        best_overlap = 0
+        best_cand_id = None
+        best_strong = 0
+        best_formation = None
+        for cand in clusters:
+            overlap = len(sc_members & set(cand["customer_ids"]))
+            if overlap > best_overlap:
+                best_overlap = overlap
+                best_cand_id = cand["cluster_id"]
+                best_strong = cand["strong_link_count"]
+                best_formation = cand.get("formation_type", "unknown")
+        appears = best_overlap >= threshold
+        correctly_weak = appears and best_strong == 0 and best_formation == "weak_only"
+        if correctly_weak:
+            benign_correctly_weak += 1
+        benign_detail.append({
+            "cluster_id": sc_id,
+            "sc_size": len(sc_members),
+            "appears_as_candidate": appears,
+            "best_overlap": best_overlap,
+            "best_cand_id": best_cand_id,
+            "max_strong_in_match": best_strong,
+            "formation_type": best_formation,
+            "correctly_weak": correctly_weak,
+        })
+
+    # Print
+    print("\n========== Ground-Truth Validation ==========")
+    print(f"Rings recovered (>=50% members in a candidate): {rings_recovered} / {len(rings)}")
+
+    # Separation check
+    separate_count = sum(1 for cnt in matched_by_cand.values() if cnt == 1)
+    multi_count = sum(1 for cnt in matched_by_cand.values() if cnt > 1)
+    print(f"Rings in separate candidates: {separate_count}  (rings sharing a candidate: {rings_sharing_cluster})")
+
+    for d in ring_detail:
+        flag = "[OK]" if d["recovered"] else "[--]"
+        share_flag = ""
+        if d["matched_cluster"] and matched_by_cand.get(d["matched_cluster"], 0) > 1:
+            share_flag = "  [SHARES WITH OTHER RING]"
+        print(f"  {flag} {d['ring_id']} (size={d['ring_size']})  "
+              f"best_overlap={d['best_overlap']}  "
+              f"candidate={d['matched_cluster']}{share_flag}")
+
+    below_filter = [d for d in benign_detail if not d["appears_as_candidate"]]
+    appeared = [d for d in benign_detail if d["appears_as_candidate"]]
+    print(f"\nBenign/stress clusters (size>=3) appearing as candidates: {len(appeared)} / {len(benign_clusters)}")
+    print(f"  Correctly as weak_only with strong_link_count=0: {benign_correctly_weak} / {len(appeared)}")
+    for d in benign_detail:
+        if not d["appears_as_candidate"]:
+            print(f"  [SKIP] {d['cluster_id']} (size={d['sc_size']}) -- below size>=3 filter or no match")
+        else:
+            flag = "[OK]" if d["correctly_weak"] else "[FP]"
+            print(f"  {flag}  {d['cluster_id']} (size={d['sc_size']})  "
+                  f"formation={d['formation_type']}  "
+                  f"strong={d['max_strong_in_match']}  "
+                  f"cand={d['best_cand_id']}")
+    print("=============================================\n")
+
+    return rings_recovered, ring_detail, benign_detail, matched_by_cand
+
+
+# ---------------------------------------------------------------------------
+# Entry point
+# ---------------------------------------------------------------------------
+def main(data_dir: str = None):
+    if data_dir is None:
+        data_dir = os.path.join(BASE_DIR, "data", "synthetic")
+
+    graph_path   = os.path.join(data_dir, "graph.gpickle")
+    clusters_path = os.path.join(data_dir, "candidate_clusters.json")
+    gt_path      = os.path.join(data_dir, "ground_truth.json")
+
+    print(f"Loading graph from {graph_path} ...")
+    G = load_graph(graph_path)
+    print(f"  Graph: {G.number_of_nodes()} nodes, {G.number_of_edges()} edges")
+
+    # 1.2 — two projections
+    print("Building strong subgraph...")
+    SG = build_strong_subgraph(G)
+    strong_edges = SG.number_of_edges()
+    print(f"  Strong subgraph: {SG.number_of_nodes()} nodes, {strong_edges} edges")
+
+    print("Building weak-only subgraph...")
+    WG = build_weak_only_subgraph(G, SG)
+    weak_edges = WG.number_of_edges()
+    print(f"  Weak-only subgraph: {WG.number_of_nodes()} nodes, {weak_edges} edges")
+
+    # 1.3 — strong_anchored candidates
+    print("Detecting strong_anchored communities...")
+    strong_communities = detect_on_subgraph(SG, "strong_anchored")
+    print(f"  Raw strong communities (all sizes): {len(strong_communities)}")
+
+    # 1.4 — weak_only candidates
+    print("Detecting weak_only communities...")
+    weak_communities = detect_on_subgraph(WG, "weak_only")
+    print(f"  Raw weak-only communities (all sizes): {len(weak_communities)}")
+
+    # 1.5 — size >= 3 filter
+    strong_filtered = [(m, method, ft) for m, method, ft in strong_communities if len(m) >= 3]
+    weak_filtered   = [(m, method, ft) for m, method, ft in weak_communities   if len(m) >= 3]
+    print(f"  Strong candidates after size>=3: {len(strong_filtered)}")
+    print(f"  Weak-only candidates after size>=3: {len(weak_filtered)}")
+
+    # 1.6 — combine and build output records
+    all_filtered = strong_filtered + weak_filtered
+    output_clusters = []
+    for idx, (members, method, formation_type) in enumerate(all_filtered, start=1):
+        meta = compute_cluster_metadata(members, SG, WG)
+        output_clusters.append({
+            "cluster_id": f"CAND{idx:02d}",
+            "customer_ids": sorted(members),
+            "size": len(members),
+            "strong_link_count": meta["strong_link_count"],
+            "weak_link_count": meta["weak_link_count"],
+            "formation_type": formation_type,
+            "method": method,
+        })
+
+    # Write
+    write_clusters(output_clusters, clusters_path)
+
+    # Print overview
+    print("\nCandidate clusters:")
+    for c in output_clusters:
+        print(f"  {c['cluster_id']}: size={c['size']:3d}  "
+              f"strong={c['strong_link_count']:3d}  "
+              f"weak={c['weak_link_count']:3d}  "
+              f"formation={c['formation_type']}  "
+              f"method={c['method']}")
+
+    # Validation
+    if os.path.exists(gt_path):
+        rings_recovered, ring_detail, benign_detail, matched_by_cand = validate(
+            output_clusters, gt_path
+        )
+        total_rings = len(ring_detail)
+        separate = sum(1 for cnt in matched_by_cand.values() if cnt == 1)
+        print(f"SUMMARY: Rings recovered={rings_recovered}/{total_rings}  "
+              f"Rings in separate candidates={separate}/{total_rings}  "
+              f"Benign/stress clusters=see above")
+    else:
+        print(f"Ground truth not found at {gt_path}, skipping validation.")
+
+    return output_clusters
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--data-dir",
+        default=None,
+        help="Path to data directory containing graph.gpickle and ground_truth.json. "
+             "Defaults to data/synthetic/",
+    )
+    args = parser.parse_args()
+    main(args.data_dir)

integration/.gitkeep

@@ -0,0 +1 @@
+

integration/razorpay_actions.py

@@ -0,0 +1,310 @@
+"""
+razorpay_actions.py
+-------------------
+Phase 3 Section 3 — Razorpay test-mode integration.
+
+Executes hold/flag actions for auto_hold and escalate cases via the Razorpay
+REST API (Python SDK). Results are verified by fetching the order back and
+confirming the notes match what was written.
+
+MCP server note (Section 3.2):
+  The plan specified using the official Razorpay MCP server (@razorpay/mcp-server).
+  At execution time that npm package returned 404 Not Found on the npm registry —
+  no such package exists. Logged in phase3_summary.md as a blocker.
+  Fallback used: official Razorpay Python SDK (razorpay==2.0.1), which calls
+  the same Razorpay REST API endpoints used by any MCP integration.
+  The "tool" used is: razorpay.Client.order.create + razorpay.Client.order.edit
+  + razorpay.Client.order.fetch.  All calls are real, test-mode, verified.
+
+Actions:
+  auto_hold: create order → edit notes to mark ringguard_status=auto_hold
+  escalate:  create order → edit notes to mark ringguard_status=escalate_flagged
+
+Usage:
+    from integration.razorpay_actions import execute_action, load_cases, save_cases
+    execute_action(case)
+"""
+
+import json
+import os
+import sys
+from datetime import datetime, timezone
+
+from dotenv import load_dotenv
+
+load_dotenv()
+
+BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+sys.path.insert(0, BASE_DIR)
+
+CASES_PATH = os.path.join(BASE_DIR, "data", "synthetic", "cases_demo.json")
+
+
+# ---------------------------------------------------------------------------
+# Razorpay client factory
+# ---------------------------------------------------------------------------
+
+def _get_client():
+    """
+    Return an authenticated Razorpay client.
+    Raises RuntimeError if credentials are not set.
+    Credentials are read from environment (loaded from .env) — never hardcoded.
+    """
+    try:
+        import razorpay
+    except ImportError:
+        raise RuntimeError("razorpay SDK not installed. Run: pip install razorpay")
+
+    key_id = os.environ.get("RAZORPAY_KEY_ID", "")
+    key_secret = os.environ.get("RAZORPAY_KEY_SECRET", "")
+
+    if not key_id or not key_secret:
+        raise RuntimeError(
+            "RAZORPAY_KEY_ID and RAZORPAY_KEY_SECRET must be set in .env. "
+            "They must be test-mode credentials (rzp_test_...)."
+        )
+    if not key_id.startswith("rzp_test_"):
+        raise RuntimeError(
+            f"RAZORPAY_KEY_ID does not look like a test-mode key (got prefix "
+            f"'{key_id[:12]}...'). Only test mode is permitted."
+        )
+
+    return razorpay.Client(auth=(key_id, key_secret))
+
+
+# ---------------------------------------------------------------------------
+# Core action function
+# ---------------------------------------------------------------------------
+
+def execute_action(case: dict) -> dict:
+    """
+    Execute a Razorpay test-mode action for a single case.
+
+    For auto_hold: creates a test order and annotates it with ringguard_status=auto_hold.
+    For escalate:  creates a test order and annotates it with ringguard_status=escalate_flagged.
+    For log_only:  no API call; returns a skipped record.
+
+    Returns a dict suitable for the case's action_taken field:
+    {
+        "tool":              str,   # Razorpay SDK method used
+        "order_id":          str,   # Razorpay test-mode order ID
+        "razorpay_status":   str,   # order status from Razorpay
+        "notes_written":     dict,  # notes added to the order
+        "notes_verified":    bool,  # True if fetch confirms notes match
+        "timestamp":         str,   # ISO 8601 UTC
+        "test_mode":         bool,  # always True
+        "error":             str or None
+    }
+    """
+    decision = case.get("policy_decision", "log_only")
+    case_id  = case.get("case_id", "?")
+    cluster_id = case.get("cluster_id", "?")
+
+    # log_only: skip — no Razorpay call
+    if decision == "log_only":
+        return {
+            "tool": None,
+            "order_id": None,
+            "razorpay_status": "skipped",
+            "notes_written": None,
+            "notes_verified": False,
+            "timestamp": datetime.now(timezone.utc).isoformat(),
+            "test_mode": True,
+            "error": None,
+            "skip_reason": "log_only — no action required",
+        }
+
+    try:
+        client = _get_client()
+
+        # Determine amount in paise (Razorpay requires integer paise)
+        # We use a nominal ₹1 order (100 paise) for escalate flags and
+        # the cluster's actual exposure for auto_hold orders.
+        # Note: actual payment capture won't happen — test mode only.
+        if decision == "auto_hold":
+            # Use actual cluster risk_probability as a proxy for notes
+            amount_paise = max(100, int(case.get("risk_probability", 0.99) * 100))  # symbolic
+            ringguard_status = "auto_hold"
+            hold_label = "FRAUD_RING_AUTO_HOLD"
+        else:  # escalate
+            amount_paise = 100  # ₹1 placeholder
+            ringguard_status = "escalate_flagged"
+            hold_label = "FRAUD_RING_ESCALATE"
+
+        # Step 1: Create a test-mode order seeded from the case's cluster data
+        created_notes = {
+            "case_id":          case_id,
+            "cluster_id":       cluster_id,
+            "ringguard_status": f"pending_{ringguard_status}",
+            "customer_ids":     ",".join(case.get("customer_ids", [])[:5]),  # first 5
+            "risk_score":       str(round(case.get("risk_probability", 0), 4)),
+        }
+        order = client.order.create({
+            "amount":   amount_paise,
+            "currency": "INR",
+            "notes":    created_notes,
+        })
+        order_id = order["id"]
+
+        # Step 2: Edit the order to mark the RingGuard action
+        action_notes = {
+            "case_id":          case_id,
+            "cluster_id":       cluster_id,
+            "ringguard_status": ringguard_status,
+            "action_label":     hold_label,
+            "risk_score":       str(round(case.get("risk_probability", 0), 4)),
+            "action_by":        "RingGuard_v3",
+            "action_timestamp": datetime.now(timezone.utc).isoformat(),
+        }
+        edited = client.order.edit(order_id, {"notes": action_notes})
+
+        # Step 3: Fetch back and verify notes match
+        fetched = client.order.fetch(order_id)
+        fetched_notes = fetched.get("notes", {})
+
+        # Verify that ringguard_status and case_id are correctly stored
+        notes_verified = (
+            fetched_notes.get("ringguard_status") == ringguard_status
+            and fetched_notes.get("case_id") == case_id
+        )
+
+        ts = datetime.now(timezone.utc).isoformat()
+        print(f"  [{case_id}] {decision}: order_id={order_id} "
+              f"status={fetched['status']} verified={notes_verified}")
+
+        return {
+            "tool":            "razorpay.Client.order.create + order.edit + order.fetch",
+            "order_id":        order_id,
+            "razorpay_status": fetched["status"],
+            "notes_written":   action_notes,
+            "notes_verified":  notes_verified,
+            "timestamp":       ts,
+            "test_mode":       True,
+            "error":           None,
+        }
+
+    except Exception as e:
+        ts = datetime.now(timezone.utc).isoformat()
+        print(f"  [{case_id}] ERROR: {e}")
+        return {
+            "tool":            "razorpay.Client",
+            "order_id":        None,
+            "razorpay_status": "error",
+            "notes_written":   None,
+            "notes_verified":  False,
+            "timestamp":       ts,
+            "test_mode":       True,
+            "error":           str(e),
+        }
+
+
+# ---------------------------------------------------------------------------
+# Batch execution across all cases
+# ---------------------------------------------------------------------------
+
+def load_cases(path: str = None) -> list:
+    """Load cases_demo.json."""
+    p = path or CASES_PATH
+    with open(p, encoding="utf-8") as f:
+        return json.load(f)
+
+
+def save_cases(cases: list, path: str = None):
+    """Write cases_demo.json with updated action_taken fields."""
+    p = path or CASES_PATH
+    with open(p, "w", encoding="utf-8") as f:
+        json.dump(cases, f, indent=2, ensure_ascii=False)
+
+
+def run_all_actions(cases: list) -> list:
+    """
+    Execute Razorpay actions for all auto_hold and escalate cases.
+    Populates action_taken field in each case.
+    Returns the updated list.
+    """
+    actioned = 0
+    errors   = 0
+    skipped  = 0
+
+    for case in cases:
+        decision = case.get("policy_decision", "log_only")
+        print(f"Processing {case['case_id']} ({decision})...")
+
+        if decision == "log_only":
+            skipped += 1
+            continue
+
+        action = execute_action(case)
+        case["action_taken"] = action
+
+        if action.get("error"):
+            errors += 1
+        else:
+            actioned += 1
+
+    print(f"\nDone: {actioned} actioned, {errors} errors, {skipped} log_only skipped")
+    return cases
+
+
+# ---------------------------------------------------------------------------
+# CLI entry point
+# ---------------------------------------------------------------------------
+
+if __name__ == "__main__":
+    import argparse
+
+    parser = argparse.ArgumentParser(
+        description="Execute Razorpay test-mode actions for RingGuard cases."
+    )
+    parser.add_argument(
+        "--cases", default=None,
+        help="Path to cases_demo.json. Defaults to data/synthetic/cases_demo.json."
+    )
+    parser.add_argument(
+        "--dry-run", action="store_true",
+        help="Print what would be done without calling Razorpay API."
+    )
+    args = parser.parse_args()
+
+    cases = load_cases(args.cases)
+    print(f"Loaded {len(cases)} cases from {args.cases or CASES_PATH}")
+
+    ah_cases  = [c for c in cases if c["policy_decision"] == "auto_hold"]
+    esc_cases = [c for c in cases if c["policy_decision"] == "escalate"]
+    log_cases = [c for c in cases if c["policy_decision"] == "log_only"]
+    print(f"  auto_hold: {len(ah_cases)}, escalate: {len(esc_cases)}, log_only: {len(log_cases)}")
+
+    if args.dry_run:
+        print("\n[DRY RUN] Would call Razorpay API for:")
+        for c in ah_cases + esc_cases:
+            print(f"  {c['case_id']} ({c['policy_decision']}) cluster={c['cluster_id']}")
+        sys.exit(0)
+
+    print("\nExecuting Razorpay actions (test mode)...")
+    updated = run_all_actions(cases)
+    save_cases(updated, args.cases)
+    print(f"\nUpdated cases saved to {args.cases or CASES_PATH}")
+
+    # Verify at least 1 auto_hold and 1 escalate were successful
+    ah_ok  = [c for c in updated if c["policy_decision"] == "auto_hold"
+              and c.get("action_taken", {}) and not c["action_taken"].get("error")
+              and c["action_taken"].get("notes_verified")]
+    esc_ok = [c for c in updated if c["policy_decision"] == "escalate"
+              and c.get("action_taken", {}) and not c["action_taken"].get("error")
+              and c["action_taken"].get("notes_verified")]
+
+    print(f"\nVerification:")
+    print(f"  auto_hold verified: {len(ah_ok)}/{len(ah_cases)}")
+    print(f"  escalate  verified: {len(esc_ok)}/{len(esc_cases)}")
+
+    if ah_ok:
+        ah = ah_ok[0]["action_taken"]
+        print(f"\n  First auto_hold order: {ah['order_id']}")
+        print(f"  Status: {ah['razorpay_status']}")
+        print(f"  Notes verified: {ah['notes_verified']}")
+
+    if esc_ok:
+        esc = esc_ok[0]["action_taken"]
+        print(f"\n  First escalate order: {esc['order_id']}")
+        print(f"  Status: {esc['razorpay_status']}")
+        print(f"  Notes verified: {esc['notes_verified']}")

requirements.txt

@@ -0,0 +1,62 @@
+annotated-doc==0.0.5
+annotated-types==0.8.0
+anyio==4.14.2
+certifi==2026.7.22
+charset-normalizer==3.5.1
+click==8.4.2
+colorama==0.4.6
+distro==1.9.0
+Faker==40.36.0
+fastapi==0.141.1
+groq==1.6.0
+httptools==0.8.0
+h11==0.16.0
+httpcore==1.0.9
+httpx==0.28.1
+idna==3.19
+joblib==1.5.3
+jsonpatch==1.33
+jsonpointer==3.1.1
+Jinja2==3.1.6
+langchain-core==1.6.0
+langchain-protocol==0.0.18
+langgraph==1.2.11
+langgraph-checkpoint==4.2.0
+langgraph-prebuilt==1.1.0
+langgraph-sdk==0.4.3
+langsmith==0.11.1
+lightgbm==4.7.0
+MarkupSafe==3.0.3
+narwhals==2.25.0
+networkx==3.6.1
+numpy==2.4.6
+orjson==3.12.0
+ormsgpack==1.12.2
+packaging==26.3
+pandas==3.0.5
+pydantic==2.13.4
+pydantic_core==2.46.4
+python-dateutil==2.9.0.post0
+python-dotenv==1.2.3
+python-louvain==0.16
+razorpay==2.0.1
+PyYAML==6.0.3
+requests==2.34.2
+requests-toolbelt==1.0.0
+scikit-learn==1.9.0
+scipy==1.17.1
+six==1.17.0
+starlette==1.6.0
+sniffio==1.3.1
+tenacity==9.1.4
+threadpoolctl==3.6.0
+typing-inspection==0.4.4
+typing_extensions==4.16.0
+tzdata==2026.3
+urllib3==2.7.0
+uuid_utils==0.17.0
+uvicorn==0.52.4
+watchfiles==1.2.0
+websockets==16.1.1
+xxhash==4.0.1
+zstandard==0.25.0

scoring/.gitkeep

@@ -0,0 +1 @@
+

scoring/features.py

@@ -0,0 +1,293 @@
+"""
+features.py
+-----------
+Feature extraction for RingGuard risk model.
+
+Given a candidate cluster (from candidate_clusters.json) plus the matching
+orders.json, customers.json, and optionally graph data, produces exactly 13
+features per cluster as specified in the Phase 2 plan.
+
+Feature set:
+  size, strong_link_count, weak_link_count, strong_link_density,
+  has_same_contact_edge, order_count, total_amount, avg_order_amount,
+  return_rate, chargeback_rate, shared_coupon_count, time_span_days,
+  order_velocity, formation_type_strong
+
+Usage:
+    from scoring.features import extract_features, label
+
+    features = extract_features(cluster, orders, graph=G)
+    lbl = label(cluster, ground_truth)
+"""
+
+import json
+import os
+import pickle
+from datetime import datetime
+from typing import Optional
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+def _load_json(path: str):
+    with open(path, encoding="utf-8") as f:
+        return json.load(f)
+
+
+def _load_graph(path: str):
+    try:
+        import networkx as nx
+        return nx.read_gpickle(path)
+    except AttributeError:
+        with open(path, "rb") as f:
+            return pickle.load(f)
+
+
+# ---------------------------------------------------------------------------
+# Feature extraction
+# ---------------------------------------------------------------------------
+
+def extract_features(
+    cluster: dict,
+    orders: list[dict],
+    graph=None,
+) -> dict:
+    """
+    Extract exactly 13 features from a candidate cluster.
+
+    Parameters
+    ----------
+    cluster : dict
+        One entry from candidate_clusters.json. Must have:
+          customer_ids, size, strong_link_count, weak_link_count, formation_type
+    orders : list[dict]
+        All orders from the dataset (orders.json). Filtered to cluster members.
+    graph : networkx.Graph, optional
+        The heterogeneous graph (graph.gpickle). Used to detect same_contact edges.
+        If None, has_same_contact_edge is computed from cluster data only
+        (will be 0 unless the cluster was formed via same_contact; this is
+        a reasonable fallback since same_contact rings are always strong_anchored).
+
+    Returns
+    -------
+    dict with exactly 13 feature keys + cluster_id for identification.
+    """
+    member_set = set(cluster["customer_ids"])
+    size = cluster["size"]
+    strong_link_count = cluster["strong_link_count"]
+    weak_link_count = cluster["weak_link_count"]
+    formation_type = cluster.get("formation_type", "strong_anchored")
+
+    # Feature 1: size
+    # Feature 2: strong_link_count (from cluster record)
+    # Feature 3: weak_link_count (from cluster record)
+
+    # Feature 4: strong_link_density
+    max_possible = max(1, size * (size - 1) / 2)
+    strong_link_density = strong_link_count / max_possible
+
+    # Feature 5: has_same_contact_edge
+    # Check graph for same_contact edges between cluster members
+    has_same_contact_edge = 0
+    if graph is not None:
+        for u, v, data in graph.edges(data=True):
+            if data.get("relation") == "same_contact":
+                if u in member_set and v in member_set:
+                    has_same_contact_edge = 1
+                    break
+
+    # Filter orders to cluster members
+    cluster_orders = [o for o in orders if o["customer_id"] in member_set]
+
+    # Feature 6: order_count
+    order_count = len(cluster_orders)
+
+    # Feature 7: total_amount
+    total_amount = sum(o.get("amount", 0) for o in cluster_orders)
+
+    # Feature 8: avg_order_amount
+    avg_order_amount = total_amount / max(1, order_count)
+
+    # Feature 9: return_rate
+    returned = sum(1 for o in cluster_orders if o.get("status") == "returned")
+    return_rate = returned / max(1, order_count)
+
+    # Feature 10: chargeback_rate
+    chargebacks = sum(1 for o in cluster_orders if o.get("status") == "chargeback")
+    chargeback_rate = chargebacks / max(1, order_count)
+
+    # Feature 11: shared_coupon_count
+    # Count distinct coupon_id values used by 2+ different cluster members
+    coupon_to_members: dict[str, set] = {}
+    for o in cluster_orders:
+        cid_val = o.get("coupon_id")
+        if cid_val is not None:
+            if cid_val not in coupon_to_members:
+                coupon_to_members[cid_val] = set()
+            coupon_to_members[cid_val].add(o["customer_id"])
+    shared_coupon_count = sum(
+        1 for members in coupon_to_members.values() if len(members) >= 2
+    )
+
+    # Features 12 + 13: time_span_days and order_velocity
+    timestamps = []
+    for o in cluster_orders:
+        ts_str = o.get("timestamp", "")
+        if ts_str:
+            try:
+                timestamps.append(datetime.fromisoformat(ts_str))
+            except ValueError:
+                pass
+
+    if len(timestamps) >= 2:
+        time_span_days = (max(timestamps) - min(timestamps)).total_seconds() / 86400.0
+    else:
+        time_span_days = 0.0
+
+    order_velocity = order_count / max(1.0, time_span_days)
+
+    # Feature 14 (index 13): formation_type_strong
+    formation_type_strong = 1 if formation_type == "strong_anchored" else 0
+
+    return {
+        "cluster_id": cluster["cluster_id"],
+        "size": size,
+        "strong_link_count": strong_link_count,
+        "weak_link_count": weak_link_count,
+        "strong_link_density": strong_link_density,
+        "has_same_contact_edge": has_same_contact_edge,
+        "order_count": order_count,
+        "total_amount": total_amount,
+        "avg_order_amount": avg_order_amount,
+        "return_rate": return_rate,
+        "chargeback_rate": chargeback_rate,
+        "shared_coupon_count": shared_coupon_count,
+        "time_span_days": time_span_days,
+        "order_velocity": order_velocity,
+        "formation_type_strong": formation_type_strong,
+    }
+
+
+# ---------------------------------------------------------------------------
+# Labeling
+# ---------------------------------------------------------------------------
+
+FEATURE_COLUMNS = [
+    "size",
+    "strong_link_count",
+    "weak_link_count",
+    "strong_link_density",
+    "has_same_contact_edge",
+    "order_count",
+    "total_amount",
+    "avg_order_amount",
+    "return_rate",
+    "chargeback_rate",
+    "shared_coupon_count",
+    "time_span_days",
+    "order_velocity",
+    "formation_type_strong",
+]
+
+
+def label(cluster: dict, ground_truth: dict) -> int:
+    """
+    Returns 1 if >= 50% of the cluster's customer_ids overlap with any
+    ground-truth ring, else 0.
+
+    Parameters
+    ----------
+    cluster : dict
+        One entry from candidate_clusters.json.
+    ground_truth : dict
+        Loaded ground_truth.json. Must have a "rings" key.
+    """
+    cluster_members = set(cluster["customer_ids"])
+    threshold = len(cluster_members) * 0.5
+
+    for ring in ground_truth.get("rings", []):
+        ring_members = set(ring["customer_ids"])
+        overlap = len(cluster_members & ring_members)
+        if overlap >= threshold:
+            return 1
+
+    return 0
+
+
+# ---------------------------------------------------------------------------
+# Batch extraction
+# ---------------------------------------------------------------------------
+
+def extract_all(
+    clusters: list[dict],
+    orders: list[dict],
+    ground_truth: dict = None,
+    graph=None,
+) -> list[dict]:
+    """
+    Extract features for every cluster in the list.
+    If ground_truth is provided, adds a 'label' field to each row.
+
+    Returns list of feature dicts (one per cluster), ready for pandas or json.
+    """
+    rows = []
+    for cluster in clusters:
+        row = extract_features(cluster, orders, graph=graph)
+        if ground_truth is not None:
+            row["label"] = label(cluster, ground_truth)
+        rows.append(row)
+    return rows
+
+
+# ---------------------------------------------------------------------------
+# CLI convenience: run directly to verify feature extraction on a dataset
+# ---------------------------------------------------------------------------
+
+if __name__ == "__main__":
+    import argparse
+    import sys
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--data-dir", default=None,
+                        help="Path to data directory. Defaults to data/synthetic_train/")
+    args = parser.parse_args()
+
+    base = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+    data_dir = args.data_dir or os.path.join(base, "data", "synthetic_train")
+
+    clusters_path = os.path.join(data_dir, "candidate_clusters.json")
+    orders_path   = os.path.join(data_dir, "orders.json")
+    gt_path       = os.path.join(data_dir, "ground_truth.json")
+    graph_path    = os.path.join(data_dir, "graph.gpickle")
+
+    clusters     = _load_json(clusters_path)
+    orders       = _load_json(orders_path)
+    ground_truth = _load_json(gt_path)
+    graph        = _load_graph(graph_path) if os.path.exists(graph_path) else None
+
+    rows = extract_all(clusters, orders, ground_truth=ground_truth, graph=graph)
+
+    # Sanity checks
+    assert all(len(r) == len(FEATURE_COLUMNS) + 2 for r in rows), \
+        "Row has wrong number of features (expected 14 features + cluster_id + label)"
+
+    null_count = sum(
+        1 for r in rows for k in FEATURE_COLUMNS if r.get(k) is None
+    )
+    assert null_count == 0, f"Null values found in feature table: {null_count}"
+
+    labels = [r["label"] for r in rows]
+    pos = sum(labels)
+    neg = len(labels) - pos
+
+    print(f"Feature extraction OK: {len(rows)} clusters")
+    print(f"  Positive labels (fraud ring): {pos}")
+    print(f"  Negative labels (benign):     {neg}")
+    print(f"  Null values in features:      {null_count}")
+    print("\nSample row:")
+    sample = rows[0]
+    for k in FEATURE_COLUMNS:
+        print(f"  {k:30s}: {sample[k]:.4f}" if isinstance(sample[k], float)
+              else f"  {k:30s}: {sample[k]}")

scoring/holdout_split.json

@@ -0,0 +1,464 @@
+[
+  {
+    "cluster_id": "CAND45",
+    "features": {
+      "size": 8,
+      "strong_link_count": 28,
+      "weak_link_count": 0,
+      "strong_link_density": 1.0,
+      "has_same_contact_edge": 0,
+      "order_count": 21,
+      "total_amount": 99226,
+      "avg_order_amount": 4725.047619047619,
+      "return_rate": 0.38095238095238093,
+      "chargeback_rate": 0.14285714285714285,
+      "shared_coupon_count": 2,
+      "time_span_days": 160.50546296296295,
+      "order_velocity": 0.13083666818771025,
+      "formation_type_strong": 1
+    },
+    "label": 1,
+    "risk_probability": 0.996939116034507
+  },
+  {
+    "cluster_id": "CAND59",
+    "features": {
+      "size": 4,
+      "strong_link_count": 6,
+      "weak_link_count": 0,
+      "strong_link_density": 1.0,
+      "has_same_contact_edge": 0,
+      "order_count": 11,
+      "total_amount": 42505,
+      "avg_order_amount": 3864.090909090909,
+      "return_rate": 0.45454545454545453,
+      "chargeback_rate": 0.18181818181818182,
+      "shared_coupon_count": 0,
+      "time_span_days": 148.44283564814816,
+      "order_velocity": 0.07410259950889873,
+      "formation_type_strong": 1
+    },
+    "label": 1,
+    "risk_probability": 0.9967870481969796
+  },
+  {
+    "cluster_id": "CAND97",
+    "features": {
+      "size": 3,
+      "strong_link_count": 0,
+      "weak_link_count": 3,
+      "strong_link_density": 0.0,
+      "has_same_contact_edge": 0,
+      "order_count": 9,
+      "total_amount": 23023,
+      "avg_order_amount": 2558.1111111111113,
+      "return_rate": 0.3333333333333333,
+      "chargeback_rate": 0.0,
+      "shared_coupon_count": 0,
+      "time_span_days": 154.36827546296297,
+      "order_velocity": 0.0583021347683536,
+      "formation_type_strong": 0
+    },
+    "label": 0,
+    "risk_probability": 0.0035484719416527992
+  },
+  {
+    "cluster_id": "CAND54",
+    "features": {
+      "size": 5,
+      "strong_link_count": 6,
+      "weak_link_count": 0,
+      "strong_link_density": 0.6,
+      "has_same_contact_edge": 1,
+      "order_count": 18,
+      "total_amount": 64576,
+      "avg_order_amount": 3587.5555555555557,
+      "return_rate": 0.5555555555555556,
+      "chargeback_rate": 0.1111111111111111,
+      "shared_coupon_count": 0,
+      "time_span_days": 168.73714120370371,
+      "order_velocity": 0.10667479531533575,
+      "formation_type_strong": 1
+    },
+    "label": 1,
+    "risk_probability": 0.9952350778861326
+  },
+  {
+    "cluster_id": "CAND48",
+    "features": {
+      "size": 6,
+      "strong_link_count": 9,
+      "weak_link_count": 0,
+      "strong_link_density": 0.6,
+      "has_same_contact_edge": 1,
+      "order_count": 17,
+      "total_amount": 64348,
+      "avg_order_amount": 3785.176470588235,
+      "return_rate": 0.35294117647058826,
+      "chargeback_rate": 0.11764705882352941,
+      "shared_coupon_count": 1,
+      "time_span_days": 153.33387731481483,
+      "order_velocity": 0.11086917188624103,
+      "formation_type_strong": 1
+    },
+    "label": 1,
+    "risk_probability": 0.9967341961721978
+  },
+  {
+    "cluster_id": "CAND66",
+    "features": {
+      "size": 109,
+      "strong_link_count": 0,
+      "weak_link_count": 246,
+      "strong_link_density": 0.0,
+      "has_same_contact_edge": 0,
+      "order_count": 287,
+      "total_amount": 804955,
+      "avg_order_amount": 2804.7212543554006,
+      "return_rate": 0.1986062717770035,
+      "chargeback_rate": 0.04878048780487805,
+      "shared_coupon_count": 0,
+      "time_span_days": 177.63575231481482,
+      "order_velocity": 1.6156657444238167,
+      "formation_type_strong": 0
+    },
+    "label": 0,
+    "risk_probability": 0.0035456513474297945
+  },
+  {
+    "cluster_id": "CAND108",
+    "features": {
+      "size": 3,
+      "strong_link_count": 0,
+      "weak_link_count": 3,
+      "strong_link_density": 0.0,
+      "has_same_contact_edge": 0,
+      "order_count": 5,
+      "total_amount": 11646,
+      "avg_order_amount": 2329.2,
+      "return_rate": 0.4,
+      "chargeback_rate": 0.0,
+      "shared_coupon_count": 0,
+      "time_span_days": 107.73620370370371,
+      "order_velocity": 0.04640965458325419,
+      "formation_type_strong": 0
+    },
+    "label": 0,
+    "risk_probability": 0.0029273850745234796
+  },
+  {
+    "cluster_id": "CAND87",
+    "features": {
+      "size": 83,
+      "strong_link_count": 0,
+      "weak_link_count": 222,
+      "strong_link_density": 0.0,
+      "has_same_contact_edge": 0,
+      "order_count": 209,
+      "total_amount": 582089,
+      "avg_order_amount": 2785.114832535885,
+      "return_rate": 0.1674641148325359,
+      "chargeback_rate": 0.05741626794258373,
+      "shared_coupon_count": 0,
+      "time_span_days": 179.47063657407406,
+      "order_velocity": 1.164535903976348,
+      "formation_type_strong": 0
+    },
+    "label": 0,
+    "risk_probability": 0.0035456513474297945
+  },
+  {
+    "cluster_id": "CAND02",
+    "features": {
+      "size": 6,
+      "strong_link_count": 15,
+      "weak_link_count": 0,
+      "strong_link_density": 1.0,
+      "has_same_contact_edge": 0,
+      "order_count": 18,
+      "total_amount": 61849,
+      "avg_order_amount": 3436.0555555555557,
+      "return_rate": 0.3333333333333333,
+      "chargeback_rate": 0.16666666666666666,
+      "shared_coupon_count": 0,
+      "time_span_days": 172.4928125,
+      "order_velocity": 0.10435217409421044,
+      "formation_type_strong": 1
+    },
+    "label": 1,
+    "risk_probability": 0.9920524617512916
+  },
+  {
+    "cluster_id": "CAND09",
+    "features": {
+      "size": 6,
+      "strong_link_count": 11,
+      "weak_link_count": 0,
+      "strong_link_density": 0.7333333333333333,
+      "has_same_contact_edge": 1,
+      "order_count": 19,
+      "total_amount": 73548,
+      "avg_order_amount": 3870.9473684210525,
+      "return_rate": 0.3157894736842105,
+      "chargeback_rate": 0.10526315789473684,
+      "shared_coupon_count": 1,
+      "time_span_days": 168.76631944444443,
+      "order_velocity": 0.11258170506144469,
+      "formation_type_strong": 1
+    },
+    "label": 1,
+    "risk_probability": 0.9969428584522867
+  },
+  {
+    "cluster_id": "CAND52",
+    "features": {
+      "size": 6,
+      "strong_link_count": 15,
+      "weak_link_count": 0,
+      "strong_link_density": 1.0,
+      "has_same_contact_edge": 0,
+      "order_count": 18,
+      "total_amount": 65370,
+      "avg_order_amount": 3631.6666666666665,
+      "return_rate": 0.3888888888888889,
+      "chargeback_rate": 0.1111111111111111,
+      "shared_coupon_count": 1,
+      "time_span_days": 120.77247685185185,
+      "order_velocity": 0.14904058001625722,
+      "formation_type_strong": 1
+    },
+    "label": 1,
+    "risk_probability": 0.9958819841283693
+  },
+  {
+    "cluster_id": "CAND07",
+    "features": {
+      "size": 5,
+      "strong_link_count": 7,
+      "weak_link_count": 0,
+      "strong_link_density": 0.7,
+      "has_same_contact_edge": 1,
+      "order_count": 14,
+      "total_amount": 52492,
+      "avg_order_amount": 3749.4285714285716,
+      "return_rate": 0.2857142857142857,
+      "chargeback_rate": 0.2857142857142857,
+      "shared_coupon_count": 0,
+      "time_span_days": 132.76825231481482,
+      "order_velocity": 0.10544689529243599,
+      "formation_type_strong": 1
+    },
+    "label": 1,
+    "risk_probability": 0.9959009223567944
+  },
+  {
+    "cluster_id": "CAND73",
+    "features": {
+      "size": 68,
+      "strong_link_count": 1,
+      "weak_link_count": 183,
+      "strong_link_density": 0.0004389815627743635,
+      "has_same_contact_edge": 0,
+      "order_count": 187,
+      "total_amount": 458341,
+      "avg_order_amount": 2451.0213903743315,
+      "return_rate": 0.1443850267379679,
+      "chargeback_rate": 0.09090909090909091,
+      "shared_coupon_count": 1,
+      "time_span_days": 180.15159722222222,
+      "order_velocity": 1.0380146658890295,
+      "formation_type_strong": 0
+    },
+    "label": 0,
+    "risk_probability": 0.007239076424852942
+  },
+  {
+    "cluster_id": "CAND29",
+    "features": {
+      "size": 4,
+      "strong_link_count": 6,
+      "weak_link_count": 0,
+      "strong_link_density": 1.0,
+      "has_same_contact_edge": 1,
+      "order_count": 9,
+      "total_amount": 52516,
+      "avg_order_amount": 5835.111111111111,
+      "return_rate": 0.2222222222222222,
+      "chargeback_rate": 0.3333333333333333,
+      "shared_coupon_count": 1,
+      "time_span_days": 163.2079513888889,
+      "order_velocity": 0.05514437209345865,
+      "formation_type_strong": 1
+    },
+    "label": 1,
+    "risk_probability": 0.9968779996975877
+  },
+  {
+    "cluster_id": "CAND79",
+    "features": {
+      "size": 64,
+      "strong_link_count": 0,
+      "weak_link_count": 153,
+      "strong_link_density": 0.0,
+      "has_same_contact_edge": 0,
+      "order_count": 174,
+      "total_amount": 460737,
+      "avg_order_amount": 2647.9137931034484,
+      "return_rate": 0.21264367816091953,
+      "chargeback_rate": 0.040229885057471264,
+      "shared_coupon_count": 0,
+      "time_span_days": 178.9133101851852,
+      "order_velocity": 0.9725380399026788,
+      "formation_type_strong": 0
+    },
+    "label": 0,
+    "risk_probability": 0.0035456513474297945
+  },
+  {
+    "cluster_id": "CAND69",
+    "features": {
+      "size": 69,
+      "strong_link_count": 1,
+      "weak_link_count": 163,
+      "strong_link_density": 0.00042625745950554135,
+      "has_same_contact_edge": 0,
+      "order_count": 162,
+      "total_amount": 463506,
+      "avg_order_amount": 2861.1481481481483,
+      "return_rate": 0.2222222222222222,
+      "chargeback_rate": 0.043209876543209874,
+      "shared_coupon_count": 1,
+      "time_span_days": 174.4325810185185,
+      "order_velocity": 0.9287255801300182,
+      "formation_type_strong": 0
+    },
+    "label": 0,
+    "risk_probability": 0.0033839575968115728
+  },
+  {
+    "cluster_id": "CAND77",
+    "features": {
+      "size": 112,
+      "strong_link_count": 0,
+      "weak_link_count": 265,
+      "strong_link_density": 0.0,
+      "has_same_contact_edge": 0,
+      "order_count": 290,
+      "total_amount": 774776,
+      "avg_order_amount": 2671.6413793103447,
+      "return_rate": 0.18275862068965518,
+      "chargeback_rate": 0.06206896551724138,
+      "shared_coupon_count": 0,
+      "time_span_days": 180.6612037037037,
+      "order_velocity": 1.605214589822058,
+      "formation_type_strong": 0
+    },
+    "label": 0,
+    "risk_probability": 0.0035456513474297945
+  },
+  {
+    "cluster_id": "CAND81",
+    "features": {
+      "size": 71,
+      "strong_link_count": 1,
+      "weak_link_count": 160,
+      "strong_link_density": 0.00040241448692152917,
+      "has_same_contact_edge": 1,
+      "order_count": 204,
+      "total_amount": 568308,
+      "avg_order_amount": 2785.823529411765,
+      "return_rate": 0.19117647058823528,
+      "chargeback_rate": 0.05392156862745098,
+      "shared_coupon_count": 1,
+      "time_span_days": 178.23292824074073,
+      "order_velocity": 1.144569648344976,
+      "formation_type_strong": 0
+    },
+    "label": 0,
+    "risk_probability": 0.0035711032326407374
+  },
+  {
+    "cluster_id": "CAND101",
+    "features": {
+      "size": 3,
+      "strong_link_count": 0,
+      "weak_link_count": 3,
+      "strong_link_density": 0.0,
+      "has_same_contact_edge": 0,
+      "order_count": 5,
+      "total_amount": 6796,
+      "avg_order_amount": 1359.2,
+      "return_rate": 0.6,
+      "chargeback_rate": 0.0,
+      "shared_coupon_count": 0,
+      "time_span_days": 138.31976851851852,
+      "order_velocity": 0.03614812295850965,
+      "formation_type_strong": 0
+    },
+    "label": 0,
+    "risk_probability": 0.002762985350619287
+  },
+  {
+    "cluster_id": "CAND08",
+    "features": {
+      "size": 6,
+      "strong_link_count": 15,
+      "weak_link_count": 0,
+      "strong_link_density": 1.0,
+      "has_same_contact_edge": 0,
+      "order_count": 18,
+      "total_amount": 76645,
+      "avg_order_amount": 4258.055555555556,
+      "return_rate": 0.3888888888888889,
+      "chargeback_rate": 0.05555555555555555,
+      "shared_coupon_count": 1,
+      "time_span_days": 167.69203703703704,
+      "order_velocity": 0.10733962278736263,
+      "formation_type_strong": 1
+    },
+    "label": 1,
+    "risk_probability": 0.9940880005500065
+  },
+  {
+    "cluster_id": "CAND27",
+    "features": {
+      "size": 5,
+      "strong_link_count": 10,
+      "weak_link_count": 0,
+      "strong_link_density": 1.0,
+      "has_same_contact_edge": 0,
+      "order_count": 17,
+      "total_amount": 61207,
+      "avg_order_amount": 3600.4117647058824,
+      "return_rate": 0.17647058823529413,
+      "chargeback_rate": 0.17647058823529413,
+      "shared_coupon_count": 1,
+      "time_span_days": 156.60148148148147,
+      "order_velocity": 0.1085558057253136,
+      "formation_type_strong": 1
+    },
+    "label": 1,
+    "risk_probability": 0.996945496633462
+  },
+  {
+    "cluster_id": "CAND20",
+    "features": {
+      "size": 4,
+      "strong_link_count": 6,
+      "weak_link_count": 0,
+      "strong_link_density": 1.0,
+      "has_same_contact_edge": 0,
+      "order_count": 15,
+      "total_amount": 57239,
+      "avg_order_amount": 3815.9333333333334,
+      "return_rate": 0.3333333333333333,
+      "chargeback_rate": 0.13333333333333333,
+      "shared_coupon_count": 1,
+      "time_span_days": 160.63930555555555,
+      "order_velocity": 0.09337689769091037,
+      "formation_type_strong": 1
+    },
+    "label": 1,
+    "risk_probability": 0.9969378410557287
+  }
+]

scoring/model.lgb

@@ -0,0 +1,2054 @@
+tree
+version=v4
+num_class=1
+num_tree_per_iteration=1
+label_index=0
+max_feature_idx=13
+objective=binary sigmoid:1
+feature_names=Column_0 Column_1 Column_2 Column_3 Column_4 Column_5 Column_6 Column_7 Column_8 Column_9 Column_10 Column_11 Column_12 Column_13
+feature_infos=[3:168] [0:28] [0:452] [0:1] [0:1] [5:439] [10305:1291235] [2061:4548.8000000000002] [0:0.5714285714285714] [0:0.33333333333333331] [0:2] [80.053703703703704:180.73202546296295] [0.028612871818308266:2.4496926129718695] [0:1]
+tree_sizes=553 874 769 877 874 881 880 873 873 880 882 876 878 880 885 878 887 871 875 880 881 877 883 879 879 877 884 883 877 881 880 879 883 887 886 881 887 885 883 886 884 886 886 889 888 887 890 881 887 892 885 893 884 890 893 887 888 886 891 890 883 890 886 890 891 886 886 887 891 890 888 895 890 894 892 893 896 895 893 895 895 894 895 894 896 897 894 895 893 890 897 899 892 895 895 898 887 895 896 896
+
+Tree=0
+num_leaves=4
+num_cat=0
+split_feature=2 11 5
+split_gain=82.066 0.71839 7.10543e-15
+threshold=1.0000000180025095e-35 133.51370949074075 6.5000000000000009
+decision_type=2 2 2
+left_child=1 -1 -2
+right_child=2 -3 -4
+leaf_value=0.1930445524210789 0.032261939230608494 0.23324020571869669 0.032261939230608494
+leaf_weight=1.2439156323671365 1.4926987588405598 10.448891311883925 8.2098431736230832
+leaf_count=5 6 42 33
+internal_value=0.139762 0.228964 0.0322619
+internal_weight=21.3953 11.6928 9.70254
+internal_count=86 47 39
+is_linear=0
+shrinkage=1
+
+
+Tree=1
+num_leaves=7
+num_cat=0
+split_feature=7 1 5 11 0 4
+split_gain=74.4604 0.665595 0.00139175 0.000597925 7.10543e-15 7.10543e-15
+threshold=3112.1875000000005 2.5000000000000004 5.5000000000000009 141.79473379629633 7.0000000000000009 1.0000000180025095e-35
+decision_type=2 2 2 2 2 2
+left_child=2 -2 -1 -3 -4 -5
+right_child=1 3 4 5 -6 -7
+leaf_value=-0.10342789578905492 0.050941784258864947 0.090682433465454779 -0.10163940525078299 0.089598162538490678 -0.10163940525078292 0.089598162538490636
+leaf_weight=1.2474250793457056 1.2364570945501316 1.4840020835399617 2.4993495643138868 5.6725022941827756 5.9984389543533325 3.2061969488859177
+leaf_count=5 5 6 10 23 24 13
+internal_value=1.62416e-05 0.0856162 -0.101868 0.0897534 -0.101639 0.0895982
+internal_weight=21.3444 11.5992 9.74521 10.3627 8.49779 8.8787
+internal_count=86 47 39 42 34 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=2
+num_leaves=6
+num_cat=0
+split_feature=7 1 12 11 6
+split_gain=67.6008 0.62316 0.00108445 0.000455319 4.73442e-07
+threshold=3112.1875000000005 2.5000000000000004 0.046585109941867327 141.79473379629633 17835.500000000004
+decision_type=2 2 2 2 2
+left_child=2 -2 -1 -3 -4
+right_child=1 3 4 -5 -6
+leaf_value=-0.098222188977615796 0.048669522557685856 0.087156175904335012 -0.096615383620140993 0.086204543567968514 -0.0966487227699587
+leaf_weight=1.248264998197558 1.2296466380357731 1.467251956462859 1.2484656870365123 8.7695086598396284 7.2412829995155334
+leaf_count=5 5 6 5 36 29
+internal_value=2.87115e-05 0.0823011 -0.0968461 0.0863409 -0.0966438
+internal_weight=21.2044 11.4664 9.73801 10.2368 8.48975
+internal_count=86 47 39 42 34
+is_linear=0
+shrinkage=0.05
+
+
+Tree=3
+num_leaves=7
+num_cat=0
+split_feature=2 11 8 12 6 7
+split_gain=61.8731 0.587705 0.00367442 0.00110176 1.22284e-06 2.40187e-07
+threshold=1.0000000180025095e-35 133.51370949074075 0.37980769230769235 0.065152412384244726 17835.500000000004 4317.0726495726512
+decision_type=2 2 2 2 2 2
+left_child=1 -1 4 -3 -2 -5
+right_child=2 3 -4 5 -6 -7
+leaf_value=0.04667381315066154 -0.092295175394442819 0.084538242823278584 -0.095253852398532762 0.083214287911124257 -0.0923489100898221 0.083191776414990704
+leaf_weight=1.2163800746202493 1.2412882149219493 1.9321163296699513 1.2419511824846257 6.7151724696159345 7.2002349644899368 1.4387681186199186
+leaf_count=5 5 8 5 28 29 6
+internal_value=3.82822e-05 0.0795052 -0.0927146 0.0834647 -0.092341 0.0832103
+internal_weight=20.9859 11.3024 9.68347 10.0861 8.44152 8.15394
+internal_count=86 47 39 42 34 34
+is_linear=0
+shrinkage=0.05
+
+
+Tree=4
+num_leaves=7
+num_cat=0
+split_feature=7 1 12 11 11 0
+split_gain=56.6084 0.560845 0.00294361 0.00121052 2.37162e-06 4.26503e-07
+threshold=3112.1875000000005 2.5000000000000004 0.046585109941867327 141.79473379629633 133.51370949074075 3.5000000000000004
+decision_type=2 2 2 2 2 2
+left_child=2 -2 -1 -3 -4 -5
+right_child=1 3 4 5 -6 -7
+leaf_value=-0.091218722018037074 0.044692577885030123 0.082131324226660612 -0.088549404265767295 0.080529911728688675 -0.088615273661690988 0.080562347051243738
+leaf_weight=1.2330005764961267 1.2075967341661442 1.4283113479614247 1.7207350879907588 1.176931157708166 6.638588011264801 7.2988456338644028
+leaf_count=5 5 6 7 5 27 31
+internal_value=4.57358e-05 0.0768623 -0.0889381 0.0807848 -0.0886017 0.0805578
+internal_weight=20.704 11.1117 9.59232 9.90409 8.35932 8.47578
+internal_count=86 47 39 42 34 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=5
+num_leaves=7
+num_cat=0
+split_feature=2 11 8 12 12 11
+split_gain=52.1697 0.534151 0.00597517 0.00166896 3.92204e-06 5.09929e-07
+threshold=1.0000000180025095e-35 133.51370949074075 0.37980769230769235 0.065152412384244726 0.057153336188971379 148.79337384259262
+decision_type=2 2 2 2 2 2
+left_child=1 -1 4 -3 -2 -5
+right_child=2 3 -4 5 -6 -7
+leaf_value=0.043017542984362105 -0.08524151794941176 0.079853079758136811 -0.089074561699500135 0.078163284367070693 -0.085338883602630033 0.078196740304427539
+leaf_weight=1.1899101287126566 1.2125268876552562 1.8703867793083171 1.2192777097225178 1.3829243779182432 7.0360173434019089 6.4557932019233704
+leaf_count=5 5 8 5 6 29 28
+internal_value=5.15016e-05 0.074636 -0.0858075 0.0785111 -0.0853246 0.0781908
+internal_weight=20.3668 10.899 9.46782 9.7091 8.24854 7.83872
+internal_count=86 47 39 42 34 34
+is_linear=0
+shrinkage=0.05
+
+
+Tree=6
+num_leaves=7
+num_cat=0
+split_feature=2 11 8 12 12 7
+split_gain=48.0088 0.51813 0.00474055 0.00131991 3.02662e-06 4.05399e-07
+threshold=1.0000000180025095e-35 133.51370949074075 0.37980769230769235 0.065152412384244726 0.057153336188971379 4317.0726495726512
+decision_type=2 2 2 2 2 2
+left_child=1 -1 4 -3 -2 -5
+right_child=2 3 -4 5 -6 -7
+leaf_value=0.041224848892825315 -0.082361959085578224 0.077565104646246552 -0.085798940887919936 0.076075869805599397 -0.082448199492469315 0.076045690863144111
+leaf_weight=1.179830446839335 1.192670613527296 1.8302927613258342 1.2029409408569325 6.3077281713485718 6.9214366972446442 1.3511618077754972
+leaf_count=5 5 8 5 28 29 6
+internal_value=5.60081e-05 0.0724735 -0.0828698 0.0763588 -0.0824355 0.0760705
+internal_weight=19.9861 10.669 9.31705 9.48918 8.11411 7.65889
+internal_count=86 47 39 42 34 34
+is_linear=0
+shrinkage=0.05
+
+
+Tree=7
+num_leaves=7
+num_cat=0
+split_feature=7 1 12 11 8 6
+split_gain=44.4771 0.49671 0.00810296 0.0029023 1.09563e-05 2.78578e-06
+threshold=3112.1875000000005 2.5000000000000004 0.046585109941867327 141.79473379629633 0.31009615384615391 42986.000000000007
+decision_type=2 2 2 2 2 2
+left_child=2 -2 -1 -3 -4 -5
+right_child=1 3 4 5 -6 -7
+leaf_value=-0.084285547508256564 0.039805551243280855 0.076655996510183039 -0.079880561303504738 0.074080055285562552 -0.079726609628141556 0.074165912972976755
+leaf_weight=1.1847488284111038 1.1593277156352986 1.3581690192222575 6.5559337288141251 1.096968740224838 1.4029920995235441 6.8096721768379211
+leaf_count=5 5 6 28 5 6 31
+internal_value=5.94785e-05 0.0706599 -0.0804277 0.0745208 -0.0798534 0.074154
+internal_weight=19.5678 10.4241 9.14367 9.26481 7.95893 7.90664
+internal_count=86 47 39 42 34 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=8
+num_leaves=7
+num_cat=0
+split_feature=7 8 5 6 4 9
+split_gain=41.1179 0.495951 0.00664183 0.00222736 0.00167781 0.000410439
+threshold=3112.1875000000005 0.40833333333333338 5.5000000000000009 42986.000000000007 1.0000000180025095e-35 0.21825396825396828
+decision_type=2 2 2 2 2 2
+left_child=2 3 -1 4 -2 -5
+right_child=1 -3 -4 5 -6 -7
+leaf_value=-0.081609467453499146 0.073086372699730492 0.040246179662398615 -0.077559324356391984 0.072426785842704122 0.075847270191939598 0.073321822227397235
+leaf_weight=1.1634619385004059 1.0787670463323575 1.3185045868158329 7.7883933484554291 4.9166114330291748 1.1232455372810364 1.7321313172578809
+leaf_count=5 5 6 34 23 5 8
+internal_value=6.21514e-05 0.0688546 -0.0780857 0.0731164 0.0744947 0.07266
+internal_weight=19.1211 10.1693 8.95186 8.85076 2.20201 6.64874
+internal_count=86 47 39 41 10 31
+is_linear=0
+shrinkage=0.05
+
+
+Tree=9
+num_leaves=7
+num_cat=0
+split_feature=2 11 8 12 8 12
+split_gain=38.2564 0.47099 0.00971658 0.00134374 9.15222e-05 8.5431e-06
+threshold=1.0000000180025095e-35 133.51370949074075 0.37980769230769235 0.065152412384244726 0.37980769230769235 0.057153336188971379
+decision_type=2 2 2 2 2 2
+left_child=1 -1 5 -3 -5 -2
+right_child=2 3 -4 4 -6 -7
+leaf_value=0.036884259324746746 -0.075374644139064667 0.072503460712974296 -0.080451132838729761 0.070857408177662437 0.071327830184486976 -0.075524376303285998
+leaf_weight=1.1250674128532434 1.1165634840726832 1.7085118144750575 1.1406608074903486 5.8159056305885315 1.2574799358844755 6.4885646849870682
+leaf_count=5 5 8 5 28 6 29
+internal_value=6.38395e-05 0.0673429 -0.0761478 0.071245 0.070941 -0.0755024
+internal_weight=18.6528 9.90696 8.74579 8.7819 7.07339 7.60513
+internal_count=86 47 39 42 34 34
+is_linear=0
+shrinkage=0.05
+
+
+Tree=10
+num_leaves=7
+num_cat=0
+split_feature=2 9 8 11 12 12
+split_gain=35.4973 0.479056 0.00802536 0.00186363 0.00106585 6.79661e-06
+threshold=1.0000000180025095e-35 0.075080247396852748 0.37980769230769235 141.79473379629633 0.065152412384244726 0.057153336188971379
+decision_type=2 2 2 2 2 2
+left_child=1 -1 5 -3 -5 -2
+right_child=2 3 -4 4 -6 -7
+leaf_value=0.037103223145715646 -0.073532368018833894 0.07190519761274379 -0.078196748553504059 0.070932617799112946 0.069485535107175603 -0.073667659169007235
+leaf_weight=1.2629062384367014 1.0880334228277186 1.2667635977268199 1.115504205226898 1.660253584384918 5.448053240776062 6.3236725628376007
+leaf_count=6 5 6 5 8 27 29
+internal_value=6.53544e-05 0.0658096 -0.0742429 0.0701384 0.0698235 -0.0736478
+internal_weight=18.1652 9.63798 8.52721 8.37507 7.10831 7.41171
+internal_count=86 47 39 41 35 34
+is_linear=0
+shrinkage=0.05
+
+
+Tree=11
+num_leaves=7
+num_cat=0
+split_feature=7 1 12 11 9 11
+split_gain=33.1373 0.475899 0.0114909 0.00185294 0.000220994 2.10641e-05
+threshold=3112.1875000000005 2.5000000000000004 0.046585109941867327 141.79473379629633 0.075080247396852748 133.51370949074075
+decision_type=2 2 2 2 2 2
+left_child=2 -2 -1 -3 -5 -4
+right_child=1 3 5 4 -6 -7
+leaf_value=-0.077449289151936135 0.033279286118847784 0.070389415472064426 -0.071779567111123543 0.068976460844906745 0.06817315250376077 -0.07199119917480043
+leaf_weight=1.0930734276771561 1.077754572033881 1.2302654534578303 1.4794269502162931 0.99700576066970814 6.0605845898389816 5.7282752245664597
+leaf_count=5 5 6 7 5 31 27
+internal_value=6.61067e-05 0.0645344 -0.0726722 0.0685988 0.0682866 -0.0719478
+internal_weight=17.6664 9.36561 8.30078 8.28786 7.05759 7.2077
+internal_count=86 47 39 42 36 34
+is_linear=0
+shrinkage=0.05
+
+
+Tree=12
+num_leaves=7
+num_cat=0
+split_feature=7 3 12 3 11 8
+split_gain=30.8387 0.485179 0.00960963 0.0012348 0.000973892 1.71113e-05
+threshold=3112.1875000000005 0.63333333333333341 0.046585109941867327 0.68333333333333346 141.79473379629633 0.31009615384615391
+decision_type=2 2 2 2 2 2
+left_child=2 -2 -1 -3 -5 -4
+right_child=1 3 5 4 -6 -7
+leaf_value=-0.075520827820082612 0.033501967349706951 0.068857893861478878 -0.070460258488882196 0.06879936231023942 0.067085578873881238 -0.070254919880213665
+leaf_weight=1.0656314641237274 1.1938995271921138 1.7841720283031461 5.7696630209684372 0.98896507918834675 5.1240549236536026 1.2310415506362913
+leaf_count=5 6 9 28 5 27 6
+internal_value=6.68297e-05 0.0632094 -0.0710975 0.0677006 0.0673628 -0.0704242
+internal_weight=17.1574 9.09109 8.06634 7.89719 6.11302 7.0007
+internal_count=86 47 39 41 32 34
+is_linear=0
+shrinkage=0.05
+
+
+Tree=13
+num_leaves=7
+num_cat=0
+split_feature=2 11 8 12 9 12
+split_gain=28.875 0.460053 0.012447 0.000856805 0.00011562 1.70654e-05
+threshold=1.0000000180025095e-35 133.51370949074075 0.37980769230769235 0.065152412384244726 0.075080247396852748 0.057153336188971379
+decision_type=2 2 2 2 2 2
+left_child=1 -1 5 -3 -5 -2
+right_child=2 3 -4 4 -6 -7
+leaf_value=0.030890087269915469 -0.068844180687072695 0.067365934528544585 -0.07491155650147141 0.066556939476662982 0.065953825729960738 -0.069068423559230638
+leaf_weight=1.0397322177886978 0.99393287301063438 1.5245727151632307 1.0391902029514311 0.934267297387123 5.3173656910657883 5.795934185385704
+leaf_count=5 5 8 5 5 29 29
+internal_value=6.69498e-05 0.0621266 -0.0698155 0.0663031 0.066044 -0.0690356
+internal_weight=16.645 8.81594 7.82906 7.77621 6.25163 6.78987
+internal_count=86 47 39 42 34 34
+is_linear=0
+shrinkage=0.05
+
+
+Tree=14
+num_leaves=7
+num_cat=0
+split_feature=2 11 8 12 9 12
+split_gain=26.9382 0.459065 0.0105335 0.000704175 9.47011e-05 1.38494e-05
+threshold=1.0000000180025095e-35 133.51370949074075 0.37980769230769235 0.065152412384244726 0.075080247396852748 0.057153336188971379
+decision_type=2 2 2 2 2 2
+left_child=1 -1 5 -3 -5 -2
+right_child=2 3 -4 4 -6 -7
+leaf_value=0.029651734623831839 -0.067590571418553 0.066237534135364193 -0.073249213876235661 0.065490928904136186 0.064935639733780637 -0.067795847197589632
+leaf_weight=1.0292685776948944 0.96254490315914054 1.4742923229932783 1.0103661864995954 0.90279860794544209 5.1353192180395126 5.6141017973423004
+leaf_count=5 5 8 5 5 29 29
+internal_value=6.71579e-05 0.0609673 -0.068496 0.0652579 0.0650187 -0.0677658
+internal_weight=16.1287 8.54168 7.58701 7.51241 6.03812 6.57665
+internal_count=86 47 39 42 34 34
+is_linear=0
+shrinkage=0.05
+
+
+Tree=15
+num_leaves=7
+num_cat=0
+split_feature=7 1 12 11 9 8
+split_gain=25.2841 0.466343 0.0135987 0.00208613 7.84636e-05 2.37464e-05
+threshold=3112.1875000000005 2.5000000000000004 0.046585109941867327 141.79473379629633 0.075080247396852748 0.31009615384615391
+decision_type=2 2 2 2 2 2
+left_child=2 -2 -1 -3 -5 -4
+right_child=1 3 5 4 -6 -7
+leaf_value=-0.072893663688079216 0.027857376697344789 0.066425308757184637 -0.066629706487692036 0.064509006738265373 0.063997034156788757 -0.066375603717052159
+leaf_weight=0.98681633174419481 0.99035722017288019 1.1084902882575987 5.2441673576831818 0.8715066760778426 5.2964922934770584 1.114899083971977
+leaf_count=5 5 6 28 5 31 6
+internal_value=6.71441e-05 0.0600471 -0.0674326 0.0644283 0.0640694 -0.0665852
+internal_weight=15.6127 8.26685 7.34588 7.27649 6.168 6.35907
+internal_count=86 47 39 42 36 34
+is_linear=0
+shrinkage=0.05
+
+
+Tree=16
+num_leaves=7
+num_cat=0
+split_feature=2 9 8 11 12 12
+split_gain=23.6365 0.460809 0.0114327 0.00129081 0.000857083 1.66703e-05
+threshold=1.0000000180025095e-35 0.075080247396852748 0.37980769230769235 141.79473379629633 0.065152412384244726 0.057153336188971379
+decision_type=2 2 2 2 2 2
+left_child=1 -1 5 -3 -5 -2
+right_child=2 3 -4 4 -6 -7
+leaf_value=0.028761524977802318 -0.065324699374286921 0.06537421778726113 -0.071401860938867565 0.064570157556829572 0.063147457310399377 -0.065557862254063778
+leaf_weight=1.0873456448316572 0.89769710600376118 1.0709487497806547 0.95586602389812458 1.3890060782432554 4.4504676759243011 5.2488974630832672
+leaf_count=6 5 6 5 8 27 29
+internal_value=6.69732e-05 0.0590177 -0.0663149 0.0637785 0.0634859 -0.0655238
+internal_weight=15.1002 7.99777 7.10246 6.91042 5.83947 6.14659
+internal_count=86 47 39 41 35 34
+is_linear=0
+shrinkage=0.05
+
+
+Tree=17
+num_leaves=7
+num_cat=0
+split_feature=7 1 5 11 9 8
+split_gain=22.2333 0.452603 0.01447 0.00133644 0.000225874 1.56853e-05
+threshold=3112.1875000000005 2.5000000000000004 5.5000000000000009 141.79473379629633 0.075080247396852748 0.22390932420872542
+decision_type=2 2 2 2 2 2
+left_child=2 -2 -1 -3 -5 -4
+right_child=1 3 5 4 -6 -7
+leaf_value=-0.071224329883907991 0.025963412749295117 0.064405295726729775 -0.064573159584937589 0.063217045276053679 0.062323782790886573 -0.064388530194293372
+leaf_weight=0.9326527565717696 0.9537997394800185 1.0336525738239286 4.3687743544578552 0.82667908072471608 4.91704460978508 1.5615205615758894
+leaf_count=5 5 6 25 5 31 9
+internal_value=6.6361e-05 0.0582118 -0.065435 0.0627502 0.0624523 -0.0645245
+internal_weight=14.5941 7.73118 6.86295 6.77738 5.74372 5.93029
+internal_count=86 47 39 42 36 34
+is_linear=0
+shrinkage=0.05
+
+
+Tree=18
+num_leaves=7
+num_cat=0
+split_feature=7 9 12 6 9 9
+split_gain=20.8363 0.652884 0.0278199 0.0010596 0.000166491 0.00017995
+threshold=3449.4166666666674 0.071839745922519613 0.050737547542749055 40049.000000000007 0.21825396825396828 0.075080247396852748
+decision_type=2 2 2 2 2 2
+left_child=1 2 -1 -2 5 -5
+right_child=3 -3 -4 4 -6 -7
+leaf_value=-0.071477255919176391 0.063450035666864163 -0.022026255105573586 -0.063623777802643541 0.062407425281717423 0.062381491662216286 0.06157488956704156
+leaf_weight=1.4824470132589338 1.3250540494918821 1.0033586025238035 4.711548388004303 0.79630309343338002 1.2639052122831342 3.5102473795413971
+leaf_count=8 8 6 28 5 8 23
+internal_value=6.59451e-05 -0.0594424 -0.0655034 0.0621792 0.0618769 0.0617288
+internal_weight=14.0929 7.19735 6.194 6.89551 5.57046 4.30655
+internal_count=86 42 36 44 36 28
+is_linear=0
+shrinkage=0.05
+
+
+Tree=19
+num_leaves=7
+num_cat=0
+split_feature=7 3 12 3 11 9
+split_gain=19.6055 0.444006 0.0104534 0.00103889 0.00063865 6.98296e-05
+threshold=3112.1875000000005 0.63333333333333341 0.046585109941867327 0.68333333333333346 141.79473379629633 0.071279139270771077
+decision_type=2 2 2 2 2 2
+left_child=2 -2 -1 -3 -5 -4
+right_child=1 3 5 4 -6 -7
+leaf_value=-0.068717639224812194 0.025629335762716639 0.062741146854963084 -0.062759018789244531 0.062506077126058254 0.060952086412690371 -0.063256992222117198
+leaf_weight=0.87429985404014576 0.99747399985790242 1.4439515471458433 4.6966153234243393 0.79297935962676991 3.9772752821445465 0.828120857477188
+leaf_count=5 6 9 29 5 27 5
+internal_value=6.8314e-05 0.0565956 -0.0636376 0.0615661 0.0612104 -0.0628337
+internal_weight=13.6107 7.21168 6.39904 6.21421 4.77025 5.52474
+internal_count=86 47 39 41 32 34
+is_linear=0
+shrinkage=0.05
+
+
+Tree=20
+num_leaves=7
+num_cat=0
+split_feature=2 11 7 3 6 9
+split_gain=18.4871 0.422107 0.0177812 0.00075395 0.000542814 0.000113598
+threshold=1.0000000180025095e-35 133.51370949074075 2936.421364560174 0.68333333333333346 50443.000000000007 0.087121212121212141
+decision_type=2 2 2 2 2 2
+left_child=1 -1 -2 4 -3 -5
+right_child=2 3 -4 5 -6 -7
+leaf_value=0.023900812353780387 -0.061982260484055023 0.062332936528495059 -0.070377612538269629 0.060846385615579639 0.060668994649296808 0.060159089984660935
+leaf_weight=0.89601355791091908 5.4556252509355545 0.94887344539165486 0.71314486861228932 0.73183779418468464 1.0138113945722578 3.3682003915309906
+leaf_count=5 35 6 4 5 7 24
+internal_value=6.73833e-05 0.0559334 -0.0629528 0.0606675 0.0614734 0.0602818
+internal_weight=13.1275 6.95874 6.16877 6.06272 1.96268 4.10004
+internal_count=86 47 39 42 13 29
+is_linear=0
+shrinkage=0.05
+
+
+Tree=21
+num_leaves=7
+num_cat=0
+split_feature=7 1 5 11 3 9
+split_gain=17.3496 0.44958 0.011074 0.00117389 0.00012922 5.21093e-05
+threshold=3112.1875000000005 2.5000000000000004 5.5000000000000009 141.79473379629633 0.63333333333333341 0.071279139270771077
+decision_type=2 2 2 2 2 2
+left_child=2 -2 -1 -3 -5 -4
+right_child=1 3 5 4 -6 -7
+leaf_value=-0.06756471513158957 0.021553084846690478 0.061767345011491871 -0.061249842101192158 0.060437828583681366 0.059710418143377145 -0.061696580908138204
+leaf_weight=0.82338908314704884 0.86779353022575367 0.91309693455696095 4.3476990461349487 0.71383780241012562 4.2189132273197174 0.76806645095348347
+leaf_count=5 5 6 29 5 31 5
+internal_value=6.67108e-05 0.0551354 -0.0621831 0.0601205 0.0598157 -0.0613169
+internal_weight=12.6528 6.71364 5.93915 5.84585 4.93275 5.11577
+internal_count=86 47 39 42 36 34
+is_linear=0
+shrinkage=0.05
+
+
+Tree=22
+num_leaves=7
+num_cat=0
+split_feature=7 9 12 6 9 9
+split_gain=16.3951 0.572651 0.0250496 0.000768526 0.000137849 7.89605e-05
+threshold=3449.4166666666674 0.071839745922519613 0.050737547542749055 40049.000000000007 0.21825396825396828 0.075080247396852748
+decision_type=2 2 2 2 2 2
+left_child=1 2 -1 -2 5 -5
+right_child=3 -3 -4 4 -6 -7
+leaf_value=-0.068535316052522055 0.060816213959984794 -0.019438194723139134 -0.060602147086615446 0.059719033650448367 0.059873001198438448 0.059122713194765142
+leaf_weight=1.3200946748256681 1.1560638695955274 0.89922252297401417 4.0412775725126266 0.68117281794548068 1.0909055322408674 2.9999326914548874
+leaf_count=8 8 6 28 5 8 23
+internal_value=6.59541e-05 -0.0563625 -0.0625555 0.0596596 0.0593794 0.0592331
+internal_weight=12.1887 6.26059 5.36137 5.92807 4.77201 3.68111
+internal_count=86 42 36 44 36 28
+is_linear=0
+shrinkage=0.05
+
+
+Tree=23
+num_leaves=7
+num_cat=0
+split_feature=2 9 7 3 6 11
+split_gain=15.4748 0.409392 0.0152171 0.00084384 0.000832987 0.000542181
+threshold=1.0000000180025095e-35 0.075080247396852748 2936.421364560174 0.68333333333333346 50443.000000000007 141.79473379629633
+decision_type=2 2 2 2 2 2
+left_child=1 -1 -2 4 -3 -5
+right_child=2 3 -4 5 -6 -7
+leaf_value=0.022590831551048782 -0.060005776826580329 0.06106710650507513 -0.068132524420055657 0.060068144672354024 0.058892339480384651 0.058505384784072391
+leaf_weight=0.88774001598358143 4.8577902466058731 1.0309564769268034 0.65351083874702443 0.68879798054695163 0.76853059977293003 2.8573894053697586
+leaf_count=6 35 7 4 5 6 23
+internal_value=6.80767e-05 0.0540346 -0.0609694 0.0592564 0.0601383 0.0588089
+internal_weight=11.7447 6.23341 5.5113 5.34567 1.79949 3.54619
+internal_count=86 47 39 41 13 28
+is_linear=0
+shrinkage=0.05
+
+
+Tree=24
+num_leaves=7
+num_cat=0
+split_feature=7 9 12 6 9 9
+split_gain=14.5655 0.541723 0.0216997 0.000501072 9.78822e-05 0.000128185
+threshold=3449.4166666666674 0.071839745922519613 0.050737547542749055 40049.000000000007 0.075080247396852748 0.21825396825396828
+decision_type=2 2 2 2 2 2
+left_child=1 2 -1 -2 -5 -6
+right_child=3 -3 -4 4 5 -7
+leaf_value=-0.067024921199474077 0.059588834455201412 -0.018346171426203251 -0.05938954185150102 0.058951152261286548 0.058107946346887443 0.058767401643471141
+leaf_weight=1.2402108311653135 1.0672495663166044 0.86150559782981861 3.726663663983345 0.64380145072937001 2.760124303400517 1.0052870362997053
+leaf_count=8 8 6 28 5 23 8
+internal_value=6.69666e-05 -0.0549476 -0.0612961 0.0586167 0.0583814 0.058284
+internal_weight=11.3048 5.82838 4.96687 5.47646 4.40921 3.76541
+internal_count=86 42 36 44 36 31
+is_linear=0
+shrinkage=0.05
+
+
+Tree=25
+num_leaves=7
+num_cat=0
+split_feature=7 8 5 3 6 11
+split_gain=13.7558 0.413425 0.00826678 0.000784093 0.00100285 0.000358132
+threshold=3112.1875000000005 0.40833333333333338 5.5000000000000009 0.68333333333333346 50443.000000000007 141.79473379629633
+decision_type=2 2 2 2 2 2
+left_child=2 3 -1 4 -2 -5
+right_child=1 -3 -4 5 -6 -7
+leaf_value=-0.064760998748160811 0.060411037092192879 0.020204062489322272 -0.058976871071604077 0.058942640863666663 0.057967853146372289 0.057620131955101794
+leaf_weight=0.71913889795541752 0.85121817886829354 0.83034586906433094 4.3807578086853027 0.63516688346862826 0.82912010699510574 2.6377299129962921
+leaf_count=5 6 6 34 5 7 23
+internal_value=6.89911e-05 0.0528542 -0.0597925 0.0583275 0.0592055 0.0578768
+internal_weight=10.8835 5.78358 5.0999 4.95324 1.68034 3.2729
+internal_count=86 47 39 41 13 28
+is_linear=0
+shrinkage=0.05
+
+
+Tree=26
+num_leaves=7
+num_cat=0
+split_feature=2 12 7 6 9 3
+split_gain=13.0195 0.386766 0.0129472 0.000663642 0.000231849 0.000100648
+threshold=1.0000000180025095e-35 0.058858058382239557 2936.421364560174 42986.000000000007 0.071279139270771077 0.68333333333333346
+decision_type=2 2 2 2 2 2
+left_child=1 -1 4 -3 -2 -5
+right_child=2 3 -4 5 -6 -7
+leaf_value=0.020559904055127788 -0.058251627610288348 0.059187580586970483 -0.066266872347715874 0.05797390785239085 -0.059270945587050296 0.057422964387858488
+leaf_weight=0.81472939252853382 3.645292304456234 0.7722501829266547 0.59667611867189396 1.1771411001682284 0.65865901857614506 2.8025532513856888
+leaf_count=6 30 6 4 10 5 25
+internal_value=6.77682e-05 0.0523891 -0.0593645 0.0578462 -0.0584076 0.0575859
+internal_weight=10.4673 5.56667 4.90063 4.75194 4.30395 3.97969
+internal_count=86 47 39 41 35 35
+is_linear=0
+shrinkage=0.05
+
+
+Tree=27
+num_leaves=7
+num_cat=0
+split_feature=7 9 12 6 9 9
+split_gain=12.2925 0.489163 0.0188603 0.000383646 8.37721e-05 6.84616e-05
+threshold=3449.4166666666674 0.071839745922519613 0.050737547542749055 40049.000000000007 0.075080247396852748 0.21825396825396828
+decision_type=2 2 2 2 2 2
+left_child=1 2 -1 -2 -5 -6
+right_child=3 -3 -4 4 5 -7
+leaf_value=-0.065351573347258926 0.058201978148554413 -0.017119792068844602 -0.057861048104718864 0.057642731793756244 0.056851957818626599 0.057365219746097679
+leaf_weight=1.1290041878819463 0.95821357518434513 0.79274052381515492 3.2869395241141319 0.57451695203781161 2.436129823327065 0.88597673177719105
+leaf_count=8 8 6 28 5 23 8
+internal_value=6.65903e-05 -0.053284 -0.0597761 0.0573057 0.0570852 0.0569888
+internal_weight=10.0635 5.20868 4.41594 4.85484 3.89662 3.32211
+internal_count=86 42 36 44 36 31
+is_linear=0
+shrinkage=0.05
+
+
+Tree=28
+num_leaves=7
+num_cat=0
+split_feature=7 1 5 11 9 7
+split_gain=11.6086 0.397906 0.00725715 0.000617763 0.0002746 0.000130696
+threshold=3112.1875000000005 2.5000000000000004 5.5000000000000009 141.79473379629633 0.071279139270771077 3723.3125000000005
+decision_type=2 2 2 2 2 2
+left_child=2 -2 -1 -3 -4 -5
+right_child=1 3 4 5 -6 -7
+leaf_value=-0.063287458024544704 0.016795342489476839 0.058297174904840682 -0.057390464271318914 0.05712056058143096 -0.058537814624013043 0.056485396566092597
+leaf_weight=0.64899222552776326 0.71674503386020649 0.71003039181232441 3.2527691423892979 1.1892860233783724 0.62106423079967488 2.538827121257782
+leaf_count=5 5 6 29 11 5 25
+internal_value=6.85434e-05 0.0513629 -0.0583942 0.0569455 -0.0575744 0.056688
+internal_weight=9.67771 5.15489 4.52283 4.43814 3.87383 3.72811
+internal_count=86 47 39 42 34 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=29
+num_leaves=7
+num_cat=0
+split_feature=2 9 7 3 6 9
+split_gain=11.0019 0.370817 0.01088 0.000682509 0.000701566 0.000256016
+threshold=1.0000000180025095e-35 0.075080247396852748 2936.421364560174 0.68333333333333346 50443.000000000007 0.071279139270771077
+decision_type=2 2 2 2 2 2
+left_child=1 -1 5 4 -3 -2
+right_child=2 3 -4 -5 -6 -7
+leaf_value=0.018486343351054913 -0.056923856725493785 0.058520334769237597 -0.064665982612492262 0.05626769846211898 0.056286448244358883 -0.058052796916793395
+leaf_weight=0.74618197977542866 3.2023785859346394 0.85925551503896691 0.54325389117002476 2.7559519186615944 0.59474007040262222 0.59558508545160282
+leaf_count=6 30 7 4 28 6 5
+internal_value=6.75255e-05 0.0509722 -0.0580476 0.0567301 0.0576066 -0.0571009
+internal_weight=9.29735 4.95613 4.34122 4.20995 1.454 3.79796
+internal_count=86 47 39 41 13 35
+is_linear=0
+shrinkage=0.05
+
+
+Tree=30
+num_leaves=7
+num_cat=0
+split_feature=7 9 12 6 9 9
+split_gain=10.4161 0.44449 0.01614 0.000279951 0.000107893 5.28035e-05
+threshold=3449.4166666666674 0.071839745922519613 0.050737547542749055 40049.000000000007 0.075080247396852748 0.21825396825396828
+decision_type=2 2 2 2 2 2
+left_child=1 2 -1 -2 -5 -6
+right_child=3 -3 -4 4 5 -7
+leaf_value=-0.063912827952955498 0.057039225047386011 -0.015972305675681888 -0.056610658318094302 0.056691086786257221 0.055780573066398942 0.056260987745907258
+leaf_weight=1.0254787206649782 0.85380606353282917 0.72786590456962574 2.8874571472406387 0.51999834924936328 2.134867444634438 0.78128480911254872
+leaf_count=8 8 6 28 5 23 8
+internal_value=6.62766e-05 -0.0518505 -0.0585244 0.0562289 0.0560276 0.0559093
+internal_weight=8.93076 4.6408 3.91294 4.28996 3.43615 2.91615
+internal_count=86 42 36 44 36 31
+is_linear=0
+shrinkage=0.05
+
+
+Tree=31
+num_leaves=7
+num_cat=0
+split_feature=7 1 5 7 9 11
+split_gain=9.83589 0.374511 0.00629963 0.000526358 0.000278143 9.52215e-05
+threshold=3112.1875000000005 2.5000000000000004 5.5000000000000009 3640.0756302521013 0.071279139270771077 141.79473379629633
+decision_type=2 2 2 2 2 2
+left_child=2 -2 -1 -3 -4 -5
+right_child=1 3 4 5 -6 -7
+leaf_value=-0.062038035970580635 0.015312301405758769 0.057267395266565102 -0.0562204121958855 0.056335163777748776 -0.057438264315309052 0.055580914687308039
+leaf_weight=0.58454194664955172 0.66288314014673222 0.63404047489166293 2.852705761790276 0.49217656999826465 0.56103963404893864 2.7935361713171005
+leaf_count=5 5 6 29 5 5 31
+internal_value=6.81728e-05 0.0500704 -0.0572418 0.0559484 -0.0564206 0.0556939
+internal_weight=8.58092 4.58264 3.99829 3.91975 3.41375 3.28571
+internal_count=86 47 39 42 34 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=32
+num_leaves=7
+num_cat=0
+split_feature=2 11 7 3 6 9
+split_gain=9.33088 0.349815 0.00907291 0.000354295 0.000385433 0.000256073
+threshold=1.0000000180025095e-35 133.51370949074075 2936.421364560174 0.68333333333333346 50443.000000000007 0.071279139270771077
+decision_type=2 2 2 2 2 2
+left_child=1 -1 5 4 -3 -2
+right_child=2 3 -4 -5 -6 -7
+leaf_value=0.015926168801723522 -0.055832119263564496 0.05713410010864043 -0.063286192030390473 0.055256561966569744 0.055386332730094849 -0.057023546364382338
+leaf_weight=0.65211432427167881 2.8032477051019677 0.64919306337833427 0.49360075592994679 2.4881960600614548 0.61358394473791122 0.5374602898955344
+leaf_count=5 30 6 4 29 7 5
+internal_value=6.71091e-05 0.0497265 -0.0569587 0.0556027 0.0562849 -0.0560238
+internal_weight=8.2374 4.40309 3.83431 3.75097 1.26278 3.34071
+internal_count=86 47 39 42 13 35
+is_linear=0
+shrinkage=0.05
+
+
+Tree=33
+num_leaves=7
+num_cat=0
+split_feature=7 7 12 6 9 9
+split_gain=8.83465 0.495924 0.00632183 0.000252359 0.000200043 6.18914e-05
+threshold=3449.4166666666674 3112.1875000000005 0.046585109941867327 40049.000000000007 0.071279139270771077 0.075080247396852748
+decision_type=2 2 2 2 2 2
+left_child=1 2 -1 -2 -4 -5
+right_child=3 -3 4 5 -6 -7
+leaf_value=-0.061556231231573035 0.056137224344672776 -0.001381895729157019 -0.055556071395645848 0.055656976219855096 -0.056634677162818239 0.055024700516101088
+leaf_weight=0.54798562079668078 0.76531849801540408 0.45721171796321858 2.6089930459856991 0.45611468702554736 0.51466943323612202 2.5557717457413673
+leaf_count=5 8 3 29 5 5 31
+internal_value=6.60386e-05 -0.0504879 -0.0566028 0.0553265 -0.0557338 0.0551205
+internal_weight=7.90606 4.12886 3.67165 3.7772 3.12366 3.01189
+internal_count=86 42 39 44 34 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=34
+num_leaves=7
+num_cat=0
+split_feature=2 9 7 7 9 0
+split_gain=8.35001 0.351499 0.00832904 0.000696102 0.000184189 0.000182014
+threshold=1.0000000180025095e-35 0.075080247396852748 2936.421364560174 3478.8119047619052 0.071279139270771077 3.5000000000000004
+decision_type=2 2 2 2 2 2
+left_child=1 -1 4 -3 -2 -5
+right_child=2 3 -4 5 -6 -7
+leaf_value=0.014947449539177768 -0.055214120163731763 0.056963217459914332 -0.062580926893981162 0.055517369242061111 -0.056269756118548744 0.054678016325414684
+leaf_weight=0.64245630055665959 2.5623199343681344 0.5072332322597507 0.46298509091138829 0.96613904088735614 0.49266319721937168 1.9485112279653547
+leaf_count=6 30 5 4 11 5 25
+internal_value=6.62901e-05 0.0488825 -0.0563315 0.0552537 -0.0553844 0.0549562
+internal_weight=7.58231 4.06434 3.51797 3.42188 3.05498 2.91465
+internal_count=86 47 39 41 35 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=35
+num_leaves=7
+num_cat=0
+split_feature=7 7 5 6 9 9
+split_gain=7.928 0.465997 0.00627062 0.000168309 0.000144454 8.61189e-05
+threshold=3449.4166666666674 3112.1875000000005 5.5000000000000009 40049.000000000007 0.071279139270771077 0.075080247396852748
+decision_type=2 2 2 2 2 2
+left_child=1 2 -1 -2 -4 -5
+right_child=3 -3 4 5 -6 -7
+leaf_value=-0.06112951982230596 0.055489765832188481 -0.0014456136893101455 -0.054969160357358007 0.055271421843250457 -0.05592704783840783 0.054502127112148469
+leaf_weight=0.51339258998632509 0.70041552931070361 0.44213592261075962 2.3835200518369679 0.43097247928380999 0.47143466770648945 2.3337739482522011
+leaf_count=5 8 3 29 5 5 31
+internal_value=6.50601e-05 -0.0497073 -0.0560422 0.0547974 -0.0551273 0.054622
+internal_weight=7.27565 3.81048 3.36835 3.46516 2.85495 2.76475
+internal_count=86 42 39 44 34 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=36
+num_leaves=7
+num_cat=0
+split_feature=7 4 7 9 6 9
+split_gain=7.48285 0.470136 0.0076194 0.000162971 0.000142931 7.29879e-05
+threshold=3449.4166666666674 1.0000000180025095e-35 2936.421364560174 0.071839745922519613 40049.000000000007 0.075080247396852748
+decision_type=2 2 2 2 2 2
+left_child=1 2 3 -1 -2 -6
+right_child=4 -3 -4 -5 5 -7
+leaf_value=-0.054668841074267054 0.055194448155474175 -0.0006523767316843794 -0.061938902913771236 -0.055792750812624518 0.054988023932388025 0.054263649587207231
+leaf_weight=2.4181659296154989 0.66973040252923999 0.43949657678604115 0.43420322984457005 0.37218537181615818 0.41201303899288211 2.2294638529419899
+leaf_count=31 8 3 4 4 5 31
+internal_value=6.53258e-05 -0.0491654 -0.0557775 -0.0548188 0.054542 0.0543766
+internal_weight=6.97526 3.66405 3.22455 2.79035 3.31121 2.64148
+internal_count=86 42 39 35 44 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=37
+num_leaves=7
+num_cat=0
+split_feature=7 8 3 7 0 9
+split_gain=7.08513 0.625026 0.00940651 0.000726114 8.40313e-05 6.44824e-05
+threshold=3112.1875000000005 0.48529411764705888 0.00030309239974547856 3478.8119047619052 3.5000000000000004 0.069618762407887225
+decision_type=2 2 2 2 2 2
+left_child=2 3 5 -2 -5 -1
+right_child=1 -3 -4 4 -6 -7
+leaf_value=-0.054395364925094974 0.05639705795332367 -0.01565991629893191 -0.063037368840106148 0.054678313320029787 0.054080703010607023 -0.055026815765338855
+leaf_weight=2.2251070216298112 0.46460698544979173 0.3510647416114806 0.36706323921680439 0.84411208331584964 1.9404393583536146 0.49407067149877537
+leaf_count=30 5 3 3 11 28 6
+internal_value=6.55694e-05 0.0477192 -0.0555243 0.0545672 0.0542619 -0.0545101
+internal_weight=6.68646 3.60022 3.08624 3.24916 2.78455 2.71918
+internal_count=86 47 39 44 39 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=38
+num_leaves=7
+num_cat=0
+split_feature=7 1 8 9 6 9
+split_gain=6.71073 0.44171 0.00833921 0.000117938 9.75054e-05 6.51949e-05
+threshold=3449.4166666666674 2.5000000000000004 0.48529411764705888 0.071839745922519613 40049.000000000007 0.075080247396852748
+decision_type=2 2 2 2 2 2
+left_child=1 2 3 -1 -2 -6
+right_child=4 -3 -4 -5 5 -7
+leaf_value=-0.054176317691610465 0.054654224266567131 -0.00070694118547682931 -0.062643158321476405 -0.055174794917573503 0.05454790994267375 0.05383536400562642
+leaf_weight=2.2735335305333151 0.61183255910873446 0.42434006929397572 0.33883924037218083 0.33996777236461628 0.38107533752918277 2.0368824377655983
+leaf_count=32 8 3 3 4 5 31
+internal_value=6.66723e-05 -0.0484071 -0.055263 -0.0543062 0.0540903 0.0539477
+internal_weight=6.40647 3.37668 2.95234 2.6135 3.02979 2.41796
+internal_count=86 42 39 36 44 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=39
+num_leaves=7
+num_cat=0
+split_feature=7 8 3 7 0 9
+split_gain=6.36573 0.578683 0.00839249 0.000692171 6.10424e-05 4.69022e-05
+threshold=3112.1875000000005 0.48529411764705888 0.00030309239974547856 3478.8119047619052 3.5000000000000004 0.069618762407887225
+decision_type=2 2 2 2 2 2
+left_child=2 3 5 -2 -5 -1
+right_child=1 -3 -4 4 -6 -7
+leaf_value=-0.053941094211068685 0.055983462634242864 -0.014797020999098879 -0.062355730228598985 0.054196430034086966 0.053663191066265548 -0.054504693631837665
+leaf_weight=2.0288115218281755 0.43503814563155252 0.33912311494350422 0.34604882448911656 0.77044664323329959 1.7689581885933874 0.45124471187591542
+leaf_count=30 5 3 3 11 28 6
+internal_value=6.69309e-05 0.0470853 -0.0550614 0.0541407 0.053825 -0.0540436
+internal_weight=6.13967 3.31357 2.82611 2.97444 2.5394 2.48006
+internal_count=86 47 39 44 39 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=40
+num_leaves=7
+num_cat=0
+split_feature=7 1 8 9 6 9
+split_gain=6.0239 0.415509 0.00716593 8.37942e-05 6.58231e-05 5.87717e-05
+threshold=3449.4166666666674 2.5000000000000004 0.48529411764705888 0.071839745922519613 40049.000000000007 0.075080247396852748
+decision_type=2 2 2 2 2 2
+left_child=1 2 3 -1 -2 -6
+right_child=4 -3 -4 -5 5 -7
+leaf_value=-0.053747198866373358 0.054174246102799986 -0.00074871437985491265 -0.061848534735531473 -0.054628365896544698 0.054158817872469428 0.053454524957827709
+leaf_weight=2.0726009085774435 0.55836819112300906 0.40939591825008381 0.31838491559028614 0.31017342954874028 0.3523313328623775 1.8596719168126581
+leaf_count=32 8 3 3 4 5 31
+internal_value=6.80133e-05 -0.0476889 -0.0548033 -0.0538619 0.0536892 0.0535667
+internal_weight=5.88093 3.11056 2.70116 2.38277 2.77037 2.212
+internal_count=86 42 39 36 44 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=41
+num_leaves=7
+num_cat=0
+split_feature=7 8 3 7 0 9
+split_gain=5.7252 0.536242 0.00747846 0.000655393 4.44475e-05 3.41942e-05
+threshold=3112.1875000000005 0.48529411764705888 0.00030309239974547856 3478.8119047619052 3.5000000000000004 0.069618762407887225
+decision_type=2 2 2 2 2 2
+left_child=2 3 5 -2 -5 -1
+right_child=1 -3 -4 4 -6 -7
+leaf_value=-0.053537010646773124 0.055610610534127497 -0.013985263998826523 -0.061727170553975308 0.053767553137312941 0.053290985516214889 -0.054040869042317564
+leaf_weight=1.8483116999268538 0.40735856071114618 0.32721980661153782 0.32622210681438435 0.70258275792002711 1.6113706566393373 0.41173452883958805
+leaf_count=30 5 3 3 11 28 6
+internal_value=6.82449e-05 0.0464896 -0.0546503 0.0537613 0.0534357 -0.0536288
+internal_weight=5.6348 3.04853 2.58627 2.72131 2.31395 2.26005
+internal_count=86 47 39 44 39 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=42
+num_leaves=7
+num_cat=0
+split_feature=7 7 3 6 9 9
+split_gain=5.41202 0.394171 0.00641111 4.38394e-05 5.48303e-05 2.91248e-05
+threshold=3449.4166666666674 3112.1875000000005 0.00030309239974547856 40049.000000000007 0.087121212121212141 0.069618762407887225
+decision_type=2 2 2 2 2 2
+left_child=1 2 5 -2 -5 -1
+right_child=3 -3 -4 4 -6 -7
+leaf_value=-0.053352663712986942 0.053747012822722956 -0.0005433541913502332 -0.06109198312579267 0.053718204717861685 0.053089683982466357 -0.053828554208506289
+leaf_weight=1.7640897743403923 0.50911201909184489 0.3937300145626067 0.31345656886696804 0.44482421502471003 1.5777324028313158 0.39315973222255696
+leaf_count=30 8 3 3 7 29 6
+internal_value=6.93086e-05 -0.047006 -0.0544103 0.0533323 0.0532279 -0.0534394
+internal_weight=5.3961 2.86444 2.47071 2.53167 2.02256 2.15725
+internal_count=86 42 39 44 36 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=43
+num_leaves=7
+num_cat=0
+split_feature=2 9 8 7 9 1
+split_gain=5.14676 0.416517 0.00643507 0.000615443 5.01119e-05 3.57671e-05
+threshold=1.0000000180025095e-35 0.071279139270771077 0.48529411764705888 3478.8119047619052 0.071839745922519613 5.0000000000000009
+decision_type=2 2 2 2 2 2
+left_child=1 -1 4 -3 -2 -5
+right_child=2 3 -4 5 -6 -7
+leaf_value=-0.0032927725542232326 -0.053189321076280861 0.05527217636602412 -0.061192203606070542 0.053330110339481311 -0.053920202121437104 0.052907016543222808
+leaf_weight=0.37335796654224429 1.8011869229376323 0.38144064322114024 0.29376778006553639 0.86185404658317599 0.26963154226541508 1.1881534345448015
+leaf_count=4 32 5 3 15 4 23
+internal_value=6.95056e-05 0.0458777 -0.0542669 0.053428 -0.0532845 0.0530849
+internal_weight=5.16939 2.80481 2.36459 2.43145 2.07082 2.05001
+internal_count=86 47 39 43 36 38
+is_linear=0
+shrinkage=0.05
+
+
+Tree=44
+num_leaves=7
+num_cat=0
+split_feature=7 9 12 9 0 4
+split_gain=4.87065 0.373455 0.00759387 4.27489e-05 4.57402e-05 4.94146e-05
+threshold=3449.4166666666674 0.071839745922519613 0.050737547542749055 0.075080247396852748 3.5000000000000004 1.0000000180025095e-35
+decision_type=2 2 2 2 2 2
+left_child=1 2 -1 -2 5 -5
+right_child=3 -3 -4 4 -6 -7
+leaf_value=-0.059481924618931842 0.05356627448655691 -0.0053640516220252593 -0.053057415019858271 0.052859926767791471 0.052761884723875802 0.053715583084513518
+leaf_weight=0.65957568958401758 0.3061713203787807 0.45867507904767979 1.5198521241545675 0.35486402362585101 1.3312645144760606 0.32168889418244362
+leaf_count=8 5 6 28 7 27 5
+internal_value=6.98011e-05 -0.0463714 -0.0550017 0.0530159 0.052932 0.0532668
+internal_weight=4.95209 2.6381 2.17943 2.31399 2.00782 0.676553
+internal_count=86 42 36 44 39 12
+is_linear=0
+shrinkage=0.05
+
+
+Tree=45
+num_leaves=7
+num_cat=0
+split_feature=7 8 3 7 9 0
+split_gain=4.63736 0.476236 0.00485685 0.000582061 2.96056e-05 2.16139e-05
+threshold=3112.1875000000005 0.48529411764705888 0.00030309239974547856 3478.8119047619052 0.069618762407887225 3.5000000000000004
+decision_type=2 2 2 2 2 2
+left_child=2 3 4 -2 -1 -5
+right_child=1 -3 -4 5 -6 -7
+leaf_value=-0.052853755533095376 0.054985344829039356 -0.013782752907821267 -0.059935001183145403 0.053043608220952276 -0.053363596419050252 0.05267888921088263
+leaf_weight=1.5297555290162566 0.35795415192842561 0.30096763372421254 0.28671444952487934 0.58293151110410679 0.34985763952136029 1.3399863652884958
+leaf_count=30 5 3 3 11 6 28
+internal_value=7.08982e-05 0.0453335 -0.0538733 0.0531341 -0.0529487 0.0527895
+internal_weight=4.74817 2.58184 2.16633 2.28087 1.87961 1.92292
+internal_count=86 47 39 44 36 39
+is_linear=0
+shrinkage=0.05
+
+
+Tree=46
+num_leaves=7
+num_cat=0
+split_feature=2 9 8 7 9 4
+split_gain=4.40922 0.447484 0.00471205 0.000496091 5.14602e-05 2.4916e-05
+threshold=1.0000000180025095e-35 0.065591397849462371 0.48529411764705888 3478.8119047619052 0.071839745922519613 1.0000000180025095e-35
+decision_type=2 2 2 2 2 2
+left_child=1 -1 4 -3 -2 -5
+right_child=2 3 -4 5 -6 -7
+leaf_value=-0.012584332341268274 -0.05271689598511281 0.054724538458758742 -0.059953959409466628 0.052504546177679447 -0.053497216445920016 0.0528828969447558
+leaf_weight=0.29559918865561519 1.5617414489388468 0.34207821637392077 0.26570267975330342 1.1270508877933023 0.24433936551213253 0.70879586413502682
+leaf_count=3 32 5 3 25 4 14
+internal_value=7.19562e-05 0.0451415 -0.0537371 0.0529764 -0.0528225 0.0526506
+internal_weight=4.54531 2.47352 2.07178 2.17792 1.80608 1.83585
+internal_count=86 47 39 44 36 39
+is_linear=0
+shrinkage=0.05
+
+
+Tree=47
+num_leaves=7
+num_cat=0
+split_feature=7 8 3 7 9 6
+split_gain=4.17468 0.43328 0.00415466 0.00042322 2.16753e-05 1.48204e-05
+threshold=3112.1875000000005 0.48529411764705888 0.00030309239974547856 3478.8119047619052 0.069618762407887225 40049.000000000007
+decision_type=2 2 2 2 2 2
+left_child=2 3 4 -2 -1 -5
+right_child=1 -3 -4 5 -6 -7
+leaf_value=-0.052566765251834173 0.054478261217956375 -0.012068978077755697 -0.059337625153044499 0.052811538560065069 -0.053024003134294333 0.052446238174248261
+leaf_weight=1.3909137547016142 0.32682865485549051 0.2935192547738551 0.26893061771988858 0.34596097096800793 0.31855252012610424 1.406175013631582
+leaf_count=30 5 3 3 7 6 32
+internal_value=7.29575e-05 0.0447977 -0.0535608 0.0528265 -0.052652 0.0525184
+internal_weight=4.35088 2.37248 1.9784 2.07896 1.70947 1.75214
+internal_count=86 47 39 44 36 39
+is_linear=0
+shrinkage=0.05
+
+
+Tree=48
+num_leaves=7
+num_cat=0
+split_feature=7 9 12 8 0 9
+split_gain=3.97248 0.585256 0.00526959 6.08468e-05 2.74167e-05 1.91699e-05
+threshold=3449.4166666666674 0.075080247396852748 0.050737547542749055 0.40833333333333338 3.5000000000000004 0.071279139270771077
+decision_type=2 2 2 2 2 2
+left_child=1 2 -1 4 -2 -4
+right_child=3 -3 5 -5 -6 -7
+leaf_value=-0.058227599861785932 0.052704543358393807 0.029445810283716289 -0.052472795418974441 0.053195680650356104 0.052255735231464741 -0.052989089446205645
+leaf_weight=0.57319482415914569 0.47649280354380597 0.23371871188282956 1.2117787078022955 0.26804463565349568 1.1903298236429689 0.21111251413822163
+leaf_count=8 10 3 27 5 29 4
+internal_value=7.39349e-05 -0.0454146 -0.0541799 0.0524965 0.052384 -0.0525494
+internal_weight=4.16467 2.2298 1.99609 1.93487 1.66682 1.42289
+internal_count=86 42 39 44 39 31
+is_linear=0
+shrinkage=0.05
+
+
+Tree=49
+num_leaves=7
+num_cat=0
+split_feature=2 9 8 7 9 4
+split_gain=3.77277 0.494401 0.00347542 0.000339162 3.96033e-05 1.39968e-05
+threshold=1.0000000180025095e-35 0.056251581878005573 0.48529411764705888 3478.8119047619052 0.071839745922519613 1.0000000180025095e-35
+decision_type=2 2 2 2 2 2
+left_child=1 -1 4 -3 -2 -5
+right_child=2 3 -4 5 -6 -7
+leaf_value=-0.025740498200464142 -0.052317830796601586 0.054115885835263458 -0.05887891549443755 0.052179560241072731 -0.053046725156541179 0.052480626276773648
+leaf_weight=0.22477163374424058 1.3525291308760641 0.30179394781589497 0.24000708013772953 1.0285397768020628 0.21613523736596096 0.61801687628030766
+leaf_count=2 32 5 3 26 4 14
+internal_value=7.30238e-05 0.0444746 -0.0532756 0.052575 -0.0524183 0.0522926
+internal_weight=3.98179 2.17312 1.80867 1.94835 1.56866 1.64656
+internal_count=86 47 39 45 36 40
+is_linear=0
+shrinkage=0.05
+
+
+Tree=50
+num_leaves=7
+num_cat=0
+split_feature=7 9 12 8 0 9
+split_gain=3.57869 0.530171 0.0041105 6.03024e-05 2.00948e-05 1.40241e-05
+threshold=3449.4166666666674 0.075080247396852748 0.050737547542749055 0.40833333333333338 3.5000000000000004 0.071279139270771077
+decision_type=2 2 2 2 2 2
+left_child=1 2 -1 4 -2 -4
+right_child=3 -3 5 -5 -6 -7
+leaf_value=-0.057523228626730809 0.052434699258117924 0.028193853840167954 -0.052225876034957666 0.05298703689463944 0.052031718022954213 -0.052688885577054048
+leaf_weight=0.52994184568524438 0.43332421034574498 0.22096543014049519 1.1010594852268694 0.24872540682554234 1.0812997855246065 0.1920730024576186
+leaf_count=8 10 3 27 5 29 4
+internal_value=7.46771e-05 -0.0449492 -0.0538145 0.0522655 0.052147 -0.0522946
+internal_weight=3.80739 2.04404 1.82307 1.76335 1.51462 1.29313
+internal_count=86 42 39 44 39 31
+is_linear=0
+shrinkage=0.05
+
+
+Tree=51
+num_leaves=7
+num_cat=0
+split_feature=2 9 8 7 9 4
+split_gain=3.39448 0.452159 0.00255562 0.000279377 3.65874e-05 1.02775e-05
+threshold=1.0000000180025095e-35 0.056251581878005573 0.48529411764705888 3478.8119047619052 0.071839745922519613 1.0000000180025095e-35
+decision_type=2 2 2 2 2 2
+left_child=1 -1 4 -3 -2 -5
+right_child=2 3 -4 5 -6 -7
+leaf_value=-0.024398645429509903 -0.052086122604661216 0.053789475943882448 -0.057966061097239188 0.051963639842756697 -0.05281445221092828 0.052234256160140505
+leaf_weight=0.21526079624891314 1.2279919162392614 0.27869978547096241 0.22100444883108128 0.93428360298275925 0.20059854164719571 0.56183291971683502
+leaf_count=2 32 5 3 26 4 14
+internal_value=7.36736e-05 0.0440359 -0.0529625 0.052336 -0.0521884 0.0520653
+internal_weight=3.63967 1.99008 1.64959 1.77482 1.42859 1.49612
+internal_count=86 47 39 45 36 40
+is_linear=0
+shrinkage=0.05
+
+
+Tree=52
+num_leaves=7
+num_cat=0
+split_feature=7 9 1 7 9 4
+split_gain=3.22825 0.421536 0.0023962 0.000239484 3.1822e-05 8.79587e-06
+threshold=3112.1875000000005 0.056251581878005573 1.5000000000000002 3478.8119047619052 0.071839745922519613 1.0000000180025095e-35
+decision_type=2 2 2 2 2 2
+left_child=2 -2 4 -3 -1 -5
+right_child=1 3 -4 5 -6 -7
+leaf_value=-0.051974075039210477 -0.022611362856803291 0.05359471917210145 -0.05772082104697214 0.051864312884805641 -0.052669764277071332 0.052120741719884484
+leaf_weight=1.1669210381805899 0.21176189929246891 0.26604645699262608 0.21785303950309742 0.89037709310650803 0.19132617488503445 0.53556656464934349
+leaf_count=32 2 5 3 26 4 14
+internal_value=7.52849e-05 0.043894 -0.0528529 0.0522176 -0.0520721 0.0519606
+internal_weight=3.47985 1.90375 1.5761 1.69199 1.35825 1.42594
+internal_count=86 47 39 45 36 40
+is_linear=0
+shrinkage=0.05
+
+
+Tree=53
+num_leaves=7
+num_cat=0
+split_feature=7 9 12 8 0 9
+split_gain=3.07242 0.457043 0.00292621 6.85316e-05 1.26423e-05 8.79499e-06
+threshold=3449.4166666666674 0.075080247396852748 0.050737547542749055 0.40833333333333338 3.5000000000000004 0.071279139270771077
+decision_type=2 2 2 2 2 2
+left_child=1 2 -1 4 -2 -4
+right_child=3 -3 5 -5 -6 -7
+leaf_value=-0.056653332418161166 0.052081422039285598 0.026886719870256912 -0.051902369512977067 0.052781343332667423 0.051737953138033879 -0.052296246369048532
+leaf_weight=0.47251412272453341 0.37544408440589883 0.19973117113113392 0.95273867994546868 0.22473639249801625 0.93540499731898308 0.16649506613612175
+leaf_count=8 10 3 27 5 29 4
+internal_value=7.68379e-05 -0.0444079 -0.0533539 0.0519746 0.0518363 -0.051961
+internal_weight=3.32706 1.79148 1.59175 1.53559 1.31085 1.11923
+internal_count=86 42 39 44 39 31
+is_linear=0
+shrinkage=0.05
+
+
+Tree=54
+num_leaves=7
+num_cat=0
+split_feature=2 9 8 7 9 4
+split_gain=2.91014 0.383712 0.00181181 0.000198127 2.92829e-05 6.46712e-06
+threshold=1.0000000180025095e-35 0.056251581878005573 0.48529411764705888 3478.8119047619052 0.071839745922519613 1.0000000180025095e-35
+decision_type=2 2 2 2 2 2
+left_child=1 -1 4 -3 -2 -5
+right_child=2 3 -4 5 -6 -7
+leaf_value=-0.021271802133954222 -0.051783011187517117 0.053316114718123579 -0.05703372297539077 0.05168061391869886 -0.052476662686324463 0.051911335109327328
+leaf_weight=0.20203236490488086 1.0617324132472274 0.24565996043384064 0.19795862212777127 0.80830621160566785 0.17760059237480152 0.48653935268521309
+leaf_count=2 32 5 3 26 4 14
+internal_value=7.57291e-05 0.0435174 -0.0525919 0.0520143 -0.0518824 0.0517673
+internal_weight=3.17983 1.74254 1.43729 1.54051 1.23933 1.29485
+internal_count=86 47 39 45 36 40
+is_linear=0
+shrinkage=0.05
+
+
+Tree=55
+num_leaves=7
+num_cat=0
+split_feature=7 9 12 8 0 9
+split_gain=2.77243 0.415122 0.00229827 6.72112e-05 9.28698e-06 6.44362e-06
+threshold=3449.4166666666674 0.075080247396852748 0.050737547542749055 0.40833333333333338 3.5000000000000004 0.071279139270771077
+decision_type=2 2 2 2 2 2
+left_child=1 2 -1 4 -2 -4
+right_child=3 -3 5 -5 -6 -7
+leaf_value=-0.056105784181477929 0.051875912022843002 0.025728307039962841 -0.051714463691019945 0.052628220802817885 0.051566996297964268 -0.052068182466303015
+leaf_weight=0.43628968298435244 0.34102963283658005 0.18887379392981518 0.86484081856906392 0.20874816179275502 0.84894894994795322 0.15127234905958176
+leaf_count=8 10 3 27 5 29 4
+internal_value=7.72327e-05 -0.0440025 -0.0530704 0.0518007 0.0516555 -0.0517671
+internal_weight=3.04 1.64128 1.4524 1.39873 1.18998 1.01611
+internal_count=86 42 39 44 39 31
+is_linear=0
+shrinkage=0.05
+
+
+Tree=56
+num_leaves=7
+num_cat=0
+split_feature=7 9 1 7 9 4
+split_gain=2.63473 0.341572 0.00145742 0.000164497 2.79808e-05 4.75855e-06
+threshold=3112.1875000000005 0.056251581878005573 1.5000000000000002 3478.8119047619052 0.071839745922519613 1.0000000180025095e-35
+decision_type=2 2 2 2 2 2
+left_child=2 -2 4 -3 -1 -5
+right_child=1 3 -4 5 -6 -7
+leaf_value=-0.051601775945499119 -0.019490915067937303 0.053064119977449434 -0.056470937512833552 0.051515543893373608 -0.052306738587078586 0.051723253421678043
+leaf_weight=0.96027212589979194 0.19069733470678318 0.22689381800591935 0.18708896823227394 0.73355109617114045 0.16493189707398415 0.4418221190571785
+leaf_count=32 2 5 3 26 4 14
+internal_value=7.6045e-05 0.0432934 -0.0523846 0.0518316 -0.0517051 0.0515936
+internal_weight=2.90526 1.59296 1.31229 1.40227 1.1252 1.17537
+internal_count=86 47 39 45 36 40
+is_linear=0
+shrinkage=0.05
+
+
+Tree=57
+num_leaves=7
+num_cat=0
+split_feature=7 9 12 8 0 9
+split_gain=2.50296 0.377471 0.00180685 6.5507e-05 6.82734e-06 4.72454e-06
+threshold=3449.4166666666674 0.075080247396852748 0.050737547542749055 0.40833333333333338 3.5000000000000004 0.071279139270771077
+decision_type=2 2 2 2 2 2
+left_child=1 2 -1 4 -2 -4
+right_child=3 -3 5 -5 -6 -7
+leaf_value=-0.055607499936801578 0.051691338019982526 0.024607570301387913 -0.051545654679950564 0.052491083545236787 0.051413343360804969 -0.051863502878320505
+leaf_weight=0.40267548523843311 0.3096493426710365 0.17867011204361904 0.7847888395190239 0.19394304789602745 0.77025091089308262 0.13737773895263672
+leaf_count=8 10 3 27 5 29 4
+internal_value=7.74691e-05 -0.0436129 -0.0528132 0.051645 0.0514931 -0.051593
+internal_weight=2.77736 1.50351 1.32484 1.27384 1.0799 0.922167
+internal_count=86 42 39 44 39 31
+is_linear=0
+shrinkage=0.05
+
+
+Tree=58
+num_leaves=7
+num_cat=0
+split_feature=2 6 8 7 10 9
+split_gain=2.37829 0.331908 0.0011062 0.00011017 0.000136782 2.62188e-05
+threshold=1.0000000180025095e-35 34778.000000000007 0.48529411764705888 3723.3125000000005 1.0000000180025095e-35 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=1 -1 5 4 -3 -2
+right_child=2 3 -4 -5 -6 -7
+leaf_value=-0.02156846862497196 -0.051447960970868081 0.053214963622713496 -0.055910400839676348 0.051425765870829365 0.051493380966299018 -0.052157042905489059
+leaf_weight=0.17535035684704814 0.87325152195990063 0.24104415997862794 0.16981088370084751 0.82012460194528103 0.22129907086491585 0.15324152633547783
+leaf_count=3 32 5 3 31 8 4
+internal_value=7.62064e-05 0.0429519 -0.0521722 0.0517737 0.0523909 -0.0515538
+internal_weight=2.65412 1.45782 1.1963 1.28247 0.462343 1.02649
+internal_count=86 47 39 44 13 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=59
+num_leaves=7
+num_cat=0
+split_feature=7 6 1 7 10 9
+split_gain=2.26299 0.31039 0.00103136 9.44642e-05 0.000117133 2.27371e-05
+threshold=3112.1875000000005 34778.000000000007 1.5000000000000002 3723.3125000000005 1.0000000180025095e-35 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 4 -3 -1
+right_child=1 3 -4 -5 -6 -7
+leaf_value=-0.051371228485124071 -0.019963441560389303 0.053050195574039838 -0.055724017708956963 0.051354380500580665 0.051418509974003837 -0.052047763416771343
+leaf_weight=0.82932015880942367 0.17249920591711987 0.22998244687914826 0.16708584688603878 0.78118370473384857 0.21080497466027737 0.14606617018580437
+leaf_count=32 3 5 3 31 8 4
+internal_value=7.76964e-05 0.0428216 -0.0520943 0.0516846 0.0522698 -0.0514725
+internal_weight=2.53694 1.39447 1.14247 1.22197 0.440787 0.975386
+internal_count=86 47 39 44 13 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=60
+num_leaves=7
+num_cat=0
+split_feature=7 9 12 8 0 1
+split_gain=2.15431 0.326455 0.0012912 3.99674e-05 6.38946e-06 3.60815e-06
+threshold=3449.4166666666674 0.075080247396852748 0.050737547542749055 0.40833333333333338 3.5000000000000004 2.5000000000000004
+decision_type=2 2 2 2 2 2
+left_child=1 2 -1 4 5 -2
+right_child=3 -3 -4 -5 -6 -7
+leaf_value=-0.054978886079340053 0.051694326374548474 0.023438303528843456 -0.051364201244319722 0.052132262398176032 0.05121131365199786 0.051331171854148955
+leaf_weight=0.3580956254154446 0.1262961048632858 0.16136743128299702 0.79672552831470966 0.16806215979158878 0.66531687788665295 0.14919918403029442
+leaf_count=8 4 3 31 5 29 6
+internal_value=7.91249e-05 -0.0431767 -0.0524851 0.051422 0.0512952 0.0514977
+internal_weight=2.42506 1.31619 1.15482 1.10887 0.940812 0.275495
+internal_count=86 42 39 44 39 10
+is_linear=0
+shrinkage=0.05
+
+
+Tree=61
+num_leaves=7
+num_cat=0
+split_feature=2 6 8 7 10 9
+split_gain=2.04469 0.282176 0.000786681 7.58747e-05 9.33969e-05 2.14542e-05
+threshold=1.0000000180025095e-35 34778.000000000007 0.48529411764705888 3723.3125000000005 1.0000000180025095e-35 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=1 -1 5 4 -3 -2
+right_child=2 3 -4 -5 -6 -7
+leaf_value=-0.018754410714250033 -0.051240132484338098 0.052803334457989619 -0.055240518558143659 0.05122231258980349 0.051279791758668064 -0.051922877177181609
+leaf_weight=0.1638234127312902 0.753892747685313 0.21225182898342587 0.15163859352469444 0.70860414579510689 0.19121293351054192 0.13578741624951363
+leaf_count=3 32 5 3 31 8 4
+internal_value=7.77931e-05 0.042509 -0.0519117 0.051534 0.0520813 -0.0513443
+internal_weight=2.31721 1.27589 1.04132 1.11207 0.403465 0.88968
+internal_count=86 47 39 44 13 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=62
+num_leaves=7
+num_cat=0
+split_feature=7 9 12 8 0 1
+split_gain=1.94651 0.297435 0.00101562 2.88044e-05 5.81162e-06 3.07927e-06
+threshold=3449.4166666666674 0.075080247396852748 0.050737547542749055 0.40833333333333338 3.5000000000000004 2.5000000000000004
+decision_type=2 2 2 2 2 2
+left_child=1 2 -1 4 5 -2
+right_child=3 -3 -4 -5 -6 -7
+leaf_value=-0.054578448834940924 0.051567146719777759 0.02239266513495286 -0.051230840124425414 0.051923259839556216 0.051093357651703752 0.051217690768354543
+leaf_weight=0.33006986044347308 0.11681191995739915 0.15269735455512989 0.72255352139472961 0.15266721323132515 0.60328011028468609 0.13693635538220406
+leaf_count=8 4 3 31 5 29 6
+internal_value=7.91791e-05 -0.0428205 -0.0522805 0.0512905 0.0511778 0.0513786
+internal_weight=2.21502 1.20532 1.05262 1.0097 0.857028 0.253748
+internal_count=86 42 39 44 39 10
+is_linear=0
+shrinkage=0.05
+
+
+Tree=63
+num_leaves=7
+num_cat=0
+split_feature=7 3 1 7 8 9
+split_gain=1.85432 0.417261 0.000627643 7.49571e-05 2.58057e-05 2.06953e-05
+threshold=3112.1875000000005 0.00030309239974547856 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.051115534100290363 -0.05621703388787315 0.052345240117811768 -0.054820674551753504 0.051099128658674879 0.051827121348642452 -0.051812371493738234
+leaf_weight=0.68126739375293288 0.098359771072864421 0.17671952582895734 0.14278436452150345 0.74558193981647491 0.14548412896692753 0.12630220502614975
+leaf_count=32 1 5 3 36 5 4
+internal_value=7.77907e-05 0.0423271 -0.0517648 0.0514045 0.051218 -0.0512245
+internal_weight=2.1165 1.16615 0.950354 1.06779 0.891066 0.80757
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=64
+num_leaves=7
+num_cat=0
+split_feature=7 3 1 7 8 9
+split_gain=1.76081 0.396075 0.000539014 6.43369e-05 2.21824e-05 1.77622e-05
+threshold=3112.1875000000005 0.00030309239974547856 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.051059950058782694 -0.05587717407369152 0.052227077895438845 -0.054573725316594573 0.051044416503153757 0.05173590102714111 -0.05172133134698214
+leaf_weight=0.64872924610972438 0.094117365777492412 0.16847153194248654 0.13641974702477455 0.70996048487722874 0.13862663134932518 0.12034851312637329
+leaf_count=32 1 5 3 36 5 4
+internal_value=7.81751e-05 0.0422537 -0.0516772 0.0513346 0.0511574 -0.0511634
+internal_weight=2.01667 1.11118 0.905498 1.01706 0.848587 0.769078
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=65
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=1.67215 0.376009 0.000462941 5.52291e-05 1.90695e-05 1.52469e-05
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.051007192846026506 -0.055557784036650598 0.052115048839742974 -0.054339948385641892 0.05099247923509416 0.051649338939514816 -0.051634989287311799
+leaf_weight=0.6177124995738269 0.090028844773769157 0.16059087030589581 0.13031063787639141 0.67600579001009464 0.1320811565965414 0.11466564238071442
+leaf_count=32 1 5 3 36 5 4
+internal_value=7.85369e-05 0.042184 -0.0515941 0.0512681 0.0510998 -0.0511055
+internal_weight=1.9214 1.05871 0.862689 0.968678 0.808087 0.732378
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=66
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=1.58809 0.357 0.000397637 4.7417e-05 1.63942e-05 1.30896e-05
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050957112841780151 -0.055257426518026911 0.052008818494963222 -0.054118591481353243 0.050943172734727149 0.051567181814864339 -0.051553093612409742
+leaf_weight=0.58814921975135814 0.086091794073581474 0.15306290239095688 0.12444949708878994 0.64364327490329742 0.12583459727466106 0.10924223065376282
+leaf_count=32 1 5 3 36 5 4
+internal_value=7.88775e-05 0.042118 -0.0515151 0.0512051 0.0510452 -0.0510505
+internal_weight=1.83047 1.00863 0.821841 0.922541 0.769478 0.697391
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=67
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=1.50838 0.338986 0.000341575 4.07151e-05 1.40952e-05 1.1239e-05
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.05090956693516191 -0.054974794454426261 0.051908070683690404 -0.053908957765295529 0.050896358129884182 0.051489200516835226 -0.051475401949002943
+leaf_weight=0.55997416283935308 0.082303412258625031 0.14587339106947184 0.11882872879505157 0.61280137114226818 0.11987428180873394 0.10406730137765408
+leaf_count=32 1 5 3 36 5 4
+internal_value=7.91989e-05 0.0420553 -0.05144 0.0511452 0.0509934 -0.0509982
+internal_weight=1.74372 0.960852 0.78287 0.878549 0.732676 0.664041
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=68
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=1.43277 0.321913 0.00029344 3.49645e-05 1.21194e-05 9.6512e-06
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050864425030284301 -0.054708686955405844 0.051812507153240289 -0.053710385607623638 0.050851905653572706 0.051415173768754034 -0.051401690927245863
+leaf_weight=0.53312471695244312 0.078660592436790466 0.13900853041559458 0.11344072036445141 0.58341139741241932 0.11418797634541988 0.099130252376198769
+leaf_count=32 1 5 3 36 5 4
+internal_value=7.95015e-05 0.0419959 -0.0513688 0.0510884 0.0509441 -0.0509487
+internal_weight=1.66096 0.915268 0.745696 0.836608 0.697599 0.632255
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=69
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=1.36104 0.305727 0.000252108 3.00296e-05 1.0421e-05 8.28868e-06
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050821559664680674 -0.054457997003113136 0.051721850934761386 -0.05352225117720219 0.050809692856807032 0.051344893548052079 -0.051331744608704925
+leaf_weight=0.50754085648804903 0.075159996747970581 0.13245491962879896 0.10827787034213543 0.55540751945227385 0.10876388847827911 0.094420861452817917
+leaf_count=32 1 5 3 36 5 4
+internal_value=7.97873e-05 0.0419395 -0.0513011 0.0510344 0.0508973 -0.0509016
+internal_weight=1.58203 0.871786 0.71024 0.796626 0.664171 0.601962
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=70
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=1.293 0.29038 0.000216615 2.5794e-05 8.96111e-06 7.11943e-06
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050780852112463595 -0.054221723843112357 0.051635834443303885 -0.053343981189505944 0.050769602713120303 0.051278163205708376 -0.051265367414903423
+leaf_weight=0.48316505178809166 0.071798063814640045 0.12619960773736238 0.10333260335028172 0.52872663736343384 0.10359065048396587 0.089929260313510895
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.00567e-05 0.041886 -0.0512368 0.0509832 0.0508529 -0.0508569
+internal_weight=1.50674 0.830315 0.676427 0.758517 0.632317 0.573094
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=71
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=1.22843 0.275826 0.000186132 2.21579e-05 7.70615e-06 6.11564e-06
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050742192952038773 -0.053998904107683078 0.051554210494796596 -0.053175019475394159 0.050731526188011147 0.051214799713511339 -0.051202363106691677
+leaf_weight=0.45994213689118624 0.068571113049983978 0.12023004423826933 0.098597445525228977 0.50330829620361328 0.098657312802970409 0.085645964369177818
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.03114e-05 0.0418352 -0.0511757 0.0509345 0.0508107 -0.0508144
+internal_weight=1.43495 0.790767 0.644186 0.722196 0.601966 0.545588
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=72
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=1.16716 0.262021 0.00015995 1.90365e-05 6.62721e-06 5.25387e-06
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050705475390719956 -0.053788693608861321 0.051476751584957708 -0.053014865889454477 0.050695358214698165 0.051154623874076044 -0.051142556259255872
+leaf_weight=0.43781932163983583 0.065475329756736755 0.1145341070368886 0.094064968638122082 0.4790946114808321 0.093953348696231842 0.081561826169490814
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.05513e-05 0.041787 -0.0511177 0.0508883 0.0507707 -0.0507741
+internal_weight=1.3665 0.753057 0.613446 0.687582 0.573048 0.519381
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=73
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=1.10901 0.248925 0.000137461 1.63561e-05 5.69959e-06 4.51411e-06
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050670596929096236 -0.053590293940464412 0.051403226371230605 -0.052863026678438688 0.050661002316579024 0.05109747441563757 -0.051085784812353235
+leaf_weight=0.41674605384469032 0.062506809830665588 0.10910011269152164 0.089727881364524364 0.45603015180677176 0.089468623511493206 0.077668054029345512
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.07783e-05 0.0417413 -0.0510626 0.0508444 0.0507326 -0.0507358
+internal_weight=1.30125 0.717106 0.584142 0.654599 0.545499 0.494414
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=74
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=1.05381 0.2365 0.000118141 1.40545e-05 4.90203e-06 3.87871e-06
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050637467274374216 -0.053402953026905391 0.051333436912432265 -0.052719051611408531 0.050628363751426902 0.051043193153155614 -0.051031878578930726
+leaf_weight=0.39667393453419209 0.059661626815795898 0.10391676984727383 0.085579007863998413 0.43406189233064651 0.085193407721817493 0.073956210166215897
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.09923e-05 0.0416978 -0.0510102 0.0508026 0.0506964 -0.0506994
+internal_weight=1.23904 0.682834 0.556209 0.623172 0.519255 0.47063
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=75
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=1.00141 0.224708 0.000101543 1.20777e-05 4.21624e-06 3.3331e-06
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050605992909838506 -0.053225989443578751 0.051267180988710331 -0.052582510474784787 0.050597355899224455 0.050991633012501328 -0.050980696151185459
+leaf_weight=0.37755667231976986 0.056935809552669525 0.098973212763667107 0.081611312925815582 0.4131390955299139 0.081118350848555565 0.070418167859315872
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.11955e-05 0.0416566 -0.0509604 0.050763 0.0506621 -0.0506649
+internal_weight=1.17975 0.650166 0.529586 0.593231 0.494257 0.447975
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=76
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.951663 0.213518 8.72826e-05 1.03797e-05 3.62654e-06 2.86445e-06
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050576091499889057 -0.053058772421869627 0.051204274601570433 -0.052453005695592851 0.050567895908851794 0.050942655295936667 -0.050932092913472754
+leaf_weight=0.35934999864548445 0.054325386881828308 0.09425897803157568 0.077817914076149464 0.39321324229240417 0.077234478667378426 0.067046143114566803
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.13871e-05 0.0416174 -0.0509131 0.0507254 0.0506294 -0.0506321
+internal_weight=1.12325 0.619032 0.504214 0.564707 0.470448 0.426396
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=77
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.90443 0.202896 7.50295e-05 8.92117e-06 3.11936e-06 2.46191e-06
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050547681250523127 -0.052900709212720103 0.051144543684963434 -0.052330160079570358 0.050539903777803798 0.050896121373446214 -0.050885935733301926
+leaf_weight=0.34201156906783581 0.051826409995555878 0.089763985946774483 0.074192073196172714 0.37423794716596603 0.073533183895051479 0.063832654617726803
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.15681e-05 0.0415802 -0.0508682 0.0506896 0.0505984 -0.0506009
+internal_weight=1.0694 0.589362 0.480036 0.537535 0.447771 0.405844
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=78
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.85958 0.192814 6.44998e-05 7.66809e-06 2.68325e-06 2.1161e-06
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050520685962263473 -0.052751246317376467 0.051087821225232033 -0.052213607565704284 0.050513304914895744 0.05085191371230572 -0.050842097782830324
+leaf_weight=0.3255009176209569 0.049434956163167953 0.085478556342422962 0.070727241225540638 0.35616886988282204 0.070006201975047588 0.060770533978939056
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.17398e-05 0.0415449 -0.0508254 0.0506556 0.0505689 -0.0505713
+internal_weight=1.01809 0.561089 0.456999 0.511654 0.426175 0.386271
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=79
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.816989 0.183241 5.54509e-05 6.59146e-06 2.30818e-06 1.81894e-06
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050495035577253246 -0.052609879592763309 0.051033952868473467 -0.052103016195310527 0.050488029243038579 0.050809911209010843 -0.050800455235265052
+leaf_weight=0.30977936740964651 0.047147151082754135 0.081393376924097538 0.067417027428746223 0.33896364271640778 0.06664562039077282 0.057852909900248051
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.1902e-05 0.0415114 -0.0507848 0.0506233 0.0505409 -0.0505431
+internal_weight=0.969199 0.53415 0.435049 0.487003 0.405609 0.367632
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=80
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.776539 0.174152 4.76742e-05 5.66639e-06 1.9856e-06 1.56364e-06
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050470660707490289 -0.052476120132517505 0.050982791862173887 -0.051998074717629184 0.050464009597148043 0.050770000967523543 -0.050760901592487029
+leaf_weight=0.29480996914207935 0.044959183782339096 0.0774995107203722 0.064255218952894211 0.32258179038763046 0.063443851657211781 0.055073190480470657
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.20557e-05 0.0414795 -0.0507462 0.0505926 0.0505143 -0.0505163
+internal_weight=0.922623 0.508484 0.414138 0.463525 0.386026 0.349883
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=81
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.73812 0.165521 4.099e-05 4.87137e-06 1.70813e-06 1.34427e-06
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050447496670033211 -0.052349535760650268 0.050934193609556068 -0.051898474212720883 0.050441182498878724 0.050732075860885184 -0.050723328052498809
+leaf_weight=0.28055744618177414 0.042867299169301987 0.073788378853350878 0.061235793866217136 0.30698465742170811 0.060393636114895344 0.052425066940486431
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.22008e-05 0.0414492 -0.0507096 0.0505635 0.050489 -0.0504909
+internal_weight=0.878252 0.484034 0.394218 0.441167 0.367378 0.332983
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=82
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.701627 0.157325 3.52449e-05 4.18822e-06 1.46947e-06 1.15574e-06
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050425480302814775 -0.052229705204215097 0.05088803351331226 -0.051803939908119739 0.050419487462217007 0.050696035386165586 -0.050687628735126826
+leaf_weight=0.26698806835338473 0.040867827832698822 0.070251740515232086 0.058352903928607702 0.29213531780987978 0.057488014455884695 0.04990250151604414
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.23391e-05 0.0414205 -0.0506747 0.0505357 0.050465 -0.0504668
+internal_weight=0.835986 0.460743 0.375243 0.419875 0.349623 0.316891
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=83
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.666961 0.14954 3.03061e-05 3.60101e-06 1.26423e-06 9.93682e-07
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050404560437629559 -0.052116234595129485 0.050844179989428556 -0.051714197764098593 0.05039886753862708 0.050661788649821377 -0.050653713021414817
+leaf_weight=0.25406966172158718 0.038957182317972183 0.066881705075502396 0.055600891355425119 0.2779985386878252 0.054720326326787472 0.047499710693955421
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.24693e-05 0.0413931 -0.0506416 0.0505094 0.0504421 -0.0504438
+internal_weight=0.795728 0.438558 0.35717 0.399601 0.332719 0.301569
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=84
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.634029 0.142147 2.60608e-05 3.09632e-06 1.08764e-06 8.54411e-07
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050384673639876559 -0.052008767086894052 0.050802517799151063 -0.051629003038727722 0.050379268663426437 0.050629237053843035 -0.050621486875958716
+leaf_weight=0.2417715173214674 0.037131857126951218 0.063670704141259193 0.052974275313317776 0.26454066997393966 0.052084204275161028 0.045211158692836761
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.25933e-05 0.0413672 -0.0506101 0.0504844 0.0504204 -0.050422
+internal_weight=0.757384 0.417427 0.339957 0.380296 0.316625 0.286983
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=85
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.602742 0.135123 2.2411e-05 2.66251e-06 9.35745e-07 7.34683e-07
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050365774686833502 -0.051906961115474784 0.050762936756621184 -0.051548115528713426 0.050360639537289399 0.050598300838636423 -0.050590866292310327
+leaf_weight=0.23006430780515075 0.035388439893722534 0.060611481312662363 0.050467765424400568 0.25172960478812456 0.049573555588722229 0.043031559325754642
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.27099e-05 0.0413425 -0.0505801 0.0504606 0.0503997 -0.0504012
+internal_weight=0.720867 0.397303 0.323564 0.361915 0.301303 0.273096
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=86
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.573016 0.128451 1.9273e-05 2.28958e-06 8.05081e-07 6.31772e-07
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050347807744096633 -0.051810504678536623 0.050725328219376711 -0.051471306045197208 0.050342931810607429 0.050568897155633556 -0.050561767693004633
+leaf_weight=0.2189200627617538 0.033723615109920502 0.057697092182934284 0.048076252453029156 0.2395347086712718 0.047182553447782993 0.040955848060548306
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.28212e-05 0.0413191 -0.0505517 0.0504379 0.0503801 -0.0503815
+internal_weight=0.68609 0.378138 0.307952 0.344414 0.286717 0.259876
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=87
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.544771 0.122113 1.65751e-05 1.96897e-06 6.92684e-07 5.43303e-07
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050330730324470864 -0.051719086213005222 0.050689591046640982 -0.0513983691992854 0.050326098707277403 0.05054094975433629 -0.050534116756501969
+leaf_weight=0.20831208350136876 0.032134164124727249 0.054920890368521214 0.045794803649187088 0.22792675299569964 0.044905624818056822 0.038979186676442623
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.29264e-05 0.0412968 -0.0505246 0.0504164 0.0503615 -0.0503628
+internal_weight=0.652974 0.359887 0.293086 0.327753 0.272832 0.247291
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=88
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.517933 0.116091 1.42555e-05 1.69331e-06 5.95974e-07 4.67253e-07
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050314498253588726 -0.051632443098970984 0.050655631740285201 -0.051329106525360046 0.050310098101705926 0.050514383222273918 -0.050507842204426924
+leaf_weight=0.19821489881724119 0.030616963282227516 0.05227650748565793 0.04361866507679224 0.2168778539635241 0.042737449519336224 0.037096953019499779
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.30259e-05 0.0412757 -0.0504989 0.050396 0.0503437 -0.050345
+internal_weight=0.621439 0.342509 0.278931 0.311892 0.259615 0.235312
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=89
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.492429 0.110369 1.2261e-05 1.45631e-06 5.12793e-07 4.0185e-07
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050299064368726144 -0.051550292034122147 0.050623360312032475 -0.051263319891751882 0.05029488594998547 0.050489132277186634 -0.050482864286208989
+leaf_weight=0.18860423564910889 0.029168995097279549 0.049757851287722588 0.04154325881972909 0.20636141765862703 0.0406729388050735 0.03530473168939352
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.31209e-05 0.0412555 -0.0504744 0.0503766 0.0503269 -0.050328
+internal_weight=0.591413 0.325961 0.265452 0.296792 0.247034 0.223909
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=90
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.468192 0.104931 1.05458e-05 1.25255e-06 4.41224e-07 3.45617e-07
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050284394815262895 -0.051472401652421375 0.050592692579550236 -0.051200835479461297 0.050280424122202294 0.050465127310612361 -0.050459127857895095
+leaf_weight=0.17945691710337996 0.027787335216999054 0.047359100077301264 0.039564179722219706 0.19635209115222096 0.03870723582804203 0.033598298672586679
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.32108e-05 0.0412364 -0.0504512 0.0503581 0.0503108 -0.0503119
+internal_weight=0.562825 0.310206 0.252619 0.282418 0.235059 0.213055
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=91
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.445159 0.0997649 9.07088e-06 1.07731e-06 3.79648e-07 2.97265e-07
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050270448158576099 -0.051398526992987692 0.050563543463076947 -0.051141481764662267 0.050266675951028406 0.050442307897245721 -0.050436566562057319
+leaf_weight=0.17075086617842317 0.026469161733984947 0.045074684545397758 0.037677189335227013 0.18682568054646254 0.036835697945207357 0.031973626464605331
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.32954e-05 0.0412183 -0.0504291 0.0503405 0.0502956 -0.0502966
+internal_weight=0.535607 0.295205 0.240402 0.268736 0.223661 0.202724
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=92
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.423268 0.094855 7.80249e-06 9.26631e-07 3.26674e-07 2.55686e-07
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.05025718836517773 -0.051328463012764294 0.05053584135422981 -0.05108509747141346 0.050253604954431264 0.050420614788711397 -0.050415120966182086
+leaf_weight=0.16246502147987485 0.025211747735738754 0.042899276595562696 0.035878210794180632 0.17775914305821061 0.035053895320743322 0.030426864977926016
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.33761e-05 0.041201 -0.050408 0.0503238 0.0502811 -0.0502821
+internal_weight=0.509694 0.280924 0.22877 0.255712 0.212813 0.192892
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=93
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.402461 0.0901889 6.71156e-06 7.9706e-07 2.81088e-07 2.19923e-07
+threshold=3112.1875000000005 1.5000000000000002 1.5000000000000002 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050244584593455377 -0.051261996673023981 0.050509510970679308 -0.051031527960906944 0.050241177033193599 0.050399988714995715 -0.050394736323651074
+leaf_weight=0.15457929950207472 0.02401246689260006 0.04082779074087739 0.034163331147283316 0.16913049947470427 0.033357597887516022 0.028954341541975737
+leaf_count=32 1 5 3 36 5 4
+internal_value=8.34525e-05 0.0411846 -0.0503881 0.050308 0.0502673 -0.0502683
+internal_weight=0.485025 0.267328 0.217697 0.243316 0.202488 0.183534
+internal_count=86 47 39 46 41 36
+is_linear=0
+shrinkage=0.05
+
+
+Tree=94
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.382686 0.0857542 8.50195e-06 6.85608e-07 2.41875e-07 1.9306e-07
+threshold=3112.1875000000005 1.5000000000000002 2.5000000000000004 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050231772045157587 -0.05119893255906393 0.050484481834977696 -0.051255954612935167 0.050229362181290621 0.050380383068234058 -0.050375359261206912
+leaf_weight=0.15570021374151111 0.022868789732456207 0.038855364779010415 0.023903153836727142 0.16091879643499851 0.031742766499519348 0.027552548795938492
+leaf_count=34 1 5 1 36 5 4
+internal_value=8.35253e-05 0.041169 -0.050369 0.0502929 0.0502542 -0.0502534
+internal_weight=0.461542 0.254386 0.207156 0.231517 0.192662 0.183253
+internal_count=86 47 39 46 41 38
+is_linear=0
+shrinkage=0.05
+
+
+Tree=95
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.363888 0.0815393 7.30348e-06 5.89764e-07 2.08124e-07 1.66078e-07
+threshold=3112.1875000000005 1.5000000000000002 2.5000000000000004 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050220417250276289 -0.051139095953743607 0.050460690435998379 -0.051193200479213857 0.050218127763647515 0.050361738473988754 -0.050356941738795569
+leaf_weight=0.14813926233910024 0.021778276190161705 0.036977351643145084 0.022764552384614944 0.15310407569631934 0.030205549206584692 0.026218139566481113
+leaf_count=34 1 5 1 36 5 4
+internal_value=8.35957e-05 0.0411541 -0.0503509 0.0502785 0.0502418 -0.0502409
+internal_weight=0.439187 0.242065 0.197122 0.220287 0.18331 0.174357
+internal_count=86 47 39 46 41 38
+is_linear=0
+shrinkage=0.05
+
+
+Tree=96
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.34602 0.0775331 6.27459e-06 5.07338e-07 1.791e-07 1.4286e-07
+threshold=3112.1875000000005 1.5000000000000002 2.5000000000000004 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050209622222626926 -0.051082307401579589 0.05043807614005931 -0.051133655309017922 0.050207445200207285 0.05034401700303897 -0.050339429089042924
+leaf_weight=0.14094396075233817 0.020738586783409119 0.035189317772164941 0.021678892895579338 0.1456673052161932 0.028742266818881035 0.024947922211140394
+leaf_count=34 1 5 1 36 5 4
+internal_value=8.36615e-05 0.04114 -0.0503337 0.0502649 0.05023 -0.0502291
+internal_weight=0.417908 0.230337 0.187571 0.209599 0.17441 0.165892
+internal_count=86 47 39 46 41 38
+is_linear=0
+shrinkage=0.05
+
+
+Tree=97
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.329035 0.0737251 5.39117e-06 4.36445e-07 1.54115e-07 1.229e-07
+threshold=3112.1875000000005 1.5000000000000002 2.5000000000000004 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050199356445814317 -0.051028411202997205 0.050416579266307451 -0.051077146716563694 0.050197288813652312 0.050327164284452677 -0.050322783240472013
+leaf_weight=0.13409677124582231 0.01974746398627758 0.033487027743831277 0.020643839612603188 0.13859035610221326 0.027349414071068168 0.023738850373774767
+leaf_count=34 1 5 1 36 5 4
+internal_value=8.37241e-05 0.0411266 -0.0503173 0.0502519 0.0502187 -0.0502179
+internal_weight=0.397654 0.219174 0.178479 0.199427 0.16594 0.157836
+internal_count=86 47 39 46 41 38
+is_linear=0
+shrinkage=0.05
+
+
+Tree=98
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.312889 0.0701055 4.63247e-06 3.75461e-07 1.3262e-07 1.05726e-07
+threshold=3112.1875000000005 1.5000000000000002 2.5000000000000004 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050189596756458045 -0.050977250616091152 0.050396141654857063 -0.051023508928724509 0.050187630953487827 0.050311141473668199 -0.050306956270874226
+leaf_weight=0.12758097238838673 0.018802745267748833 0.031866437988355756 0.019657149910926819 0.13185595325194299 0.026023643789812922 0.022588017396628857
+leaf_count=34 1 5 1 36 5 4
+internal_value=8.37838e-05 0.0411139 -0.0503017 0.0502396 0.050208 -0.0502072
+internal_weight=0.378375 0.208549 0.169826 0.189746 0.15788 0.150169
+internal_count=86 47 39 46 41 38
+is_linear=0
+shrinkage=0.05
+
+
+Tree=99
+num_leaves=7
+num_cat=0
+split_feature=7 1 1 7 8 9
+split_gain=0.29754 0.0666646 3.98088e-06 3.23019e-07 1.14123e-07 9.09602e-08
+threshold=3112.1875000000005 1.5000000000000002 2.5000000000000004 3478.8119047619052 0.40833333333333338 0.071839745922519613
+decision_type=2 2 2 2 2 2
+left_child=2 -2 5 -3 -5 -1
+right_child=1 3 -4 4 -6 -7
+leaf_value=-0.050180315600638993 -0.050928677138001657 0.05037671632404709 -0.050972594433378086 0.050178448515061494 0.050295906560174089 -0.050291912359883201
+leaf_weight=0.12138065416365862 0.017902351915836334 0.030323687475174665 0.018716664984822273 0.12544763740152121 0.024761765031144023 0.021492652595043182
+leaf_count=34 1 5 1 36 5 4
+internal_value=8.38408e-05 0.0411018 -0.0502869 0.0502279 0.0501978 -0.0501971
+internal_weight=0.360025 0.198435 0.16159 0.180533 0.150209 0.142873
+internal_count=86 47 39 46 41 38
+is_linear=0
+shrinkage=0.05
+
+
+end of trees
+
+feature_importances:
+Column_7=145
+Column_9=120
+Column_1=90
+Column_8=73
+Column_12=37
+Column_11=28
+Column_6=25
+Column_2=22
+Column_3=19
+Column_0=15
+Column_4=10
+Column_5=9
+Column_10=3
+
+parameters:
+[boosting: gbdt]
+[objective: binary]
+[metric: binary_logloss]
+[tree_learner: serial]
+[device_type: cpu]
+[data_sample_strategy: bagging]
+[data: ]
+[valid: ]
+[num_iterations: 100]
+[learning_rate: 0.05]
+[num_leaves: 7]
+[num_threads: 8]
+[seed: 42]
+[deterministic: 0]
+[force_col_wise: 0]
+[force_row_wise: 0]
+[histogram_pool_size: -1]
+[max_depth: 4]
+[min_data_in_leaf: 5]
+[min_sum_hessian_in_leaf: 0.001]
+[bagging_fraction: 1]
+[pos_bagging_fraction: 1]
+[neg_bagging_fraction: 1]
+[bagging_freq: 0]
+[bagging_seed: 400]
+[bagging_by_query: 0]
+[feature_fraction: 1]
+[feature_fraction_bynode: 1]
+[feature_fraction_seed: 30056]
+[extra_trees: 0]
+[extra_seed: 12879]
+[early_stopping_round: 0]
+[early_stopping_min_delta: 0]
+[first_metric_only: 0]
+[max_delta_step: 0]
+[lambda_l1: 0]
+[lambda_l2: 0]
+[linear_lambda: 0]
+[min_gain_to_split: 0]
+[drop_rate: 0.1]
+[max_drop: 50]
+[skip_drop: 0.5]
+[xgboost_dart_mode: 0]
+[uniform_drop: 0]
+[drop_seed: 17869]
+[top_rate: 0.2]
+[other_rate: 0.1]
+[min_data_per_group: 100]
+[max_cat_threshold: 32]
+[cat_l2: 10]
+[cat_smooth: 10]
+[max_cat_to_onehot: 4]
+[top_k: 20]
+[monotone_constraints: ]
+[monotone_constraints_method: basic]
+[monotone_penalty: 0]
+[feature_contri: ]
+[forcedsplits_filename: ]
+[refit_decay_rate: 0.9]
+[cegb_tradeoff: 1]
+[cegb_penalty_split: 0]
+[cegb_penalty_feature_lazy: ]
+[cegb_penalty_feature_coupled: ]
+[path_smooth: 0]
+[interaction_constraints: ]
+[verbosity: -1]
+[saved_feature_importance_type: 0]
+[use_quantized_grad: 0]
+[num_grad_quant_bins: 4]
+[quant_train_renew_leaf: 0]
+[stochastic_rounding: 1]
+[linear_tree: 0]
+[max_bin: 255]
+[max_bin_by_feature: ]
+[min_data_in_bin: 3]
+[bin_construct_sample_cnt: 200000]
+[data_random_seed: 175]
+[is_enable_sparse: 1]
+[enable_bundle: 1]
+[use_missing: 1]
+[zero_as_missing: 0]
+[feature_pre_filter: 1]
+[pre_partition: 0]
+[two_round: 0]
+[header: 0]
+[label_column: ]
+[weight_column: ]
+[group_column: ]
+[ignore_column: ]
+[categorical_feature: ]
+[forcedbins_filename: ]
+[precise_float_parser: 0]
+[parser_config_file: ]
+[objective_seed: 16083]
+[num_class: 1]
+[is_unbalance: 0]
+[scale_pos_weight: 1]
+[sigmoid: 1]
+[boost_from_average: 1]
+[reg_sqrt: 0]
+[alpha: 0.9]
+[fair_c: 1]
+[poisson_max_delta_step: 0.7]
+[tweedie_variance_power: 1.5]
+[lambdarank_truncation_level: 30]
+[lambdarank_norm: 1]
+[label_gain: ]
+[lambdarank_position_bias_regularization: 0]
+[eval_at: ]
+[multi_error_top_k: 1]
+[auc_mu_weights: ]
+[num_machines: 1]
+[local_listen_port: 12400]
+[time_out: 120]
+[machine_list_filename: ]
+[machines: ]
+[gpu_platform_id: -1]
+[gpu_device_id: -1]
+[gpu_device_id_list: ]
+[gpu_use_dp: 0]
+[num_gpu: 1]
+
+end of parameters
+
+pandas_categorical:null

scoring/score_demo.py

@@ -0,0 +1,81 @@
+"""
+score_demo.py
+-------------
+Score the Phase 1 demo dataset (data/synthetic/) using the trained model.
+
+Inputs:
+  data/synthetic/candidate_clusters.json  (fixed by Section 1)
+  data/synthetic/orders.json
+  data/synthetic/graph.gpickle
+  scoring/model.lgb
+
+Output:
+  data/synthetic/scored_clusters.json — candidate_clusters.json + risk_probability
+
+Usage:
+    python scoring/score_demo.py
+"""
+
+import json
+import os
+import sys
+
+import lightgbm as lgb
+import numpy as np
+
+BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+sys.path.insert(0, BASE_DIR)
+
+DEMO_DIR   = os.path.join(BASE_DIR, "data", "synthetic")
+MODEL_PATH = os.path.join(BASE_DIR, "scoring", "model.lgb")
+OUT_PATH   = os.path.join(DEMO_DIR, "scored_clusters.json")
+
+
+def score_demo():
+    from scoring.features import extract_all, _load_json, _load_graph, FEATURE_COLUMNS
+
+    clusters_path = os.path.join(DEMO_DIR, "candidate_clusters.json")
+    orders_path   = os.path.join(DEMO_DIR, "orders.json")
+    graph_path    = os.path.join(DEMO_DIR, "graph.gpickle")
+
+    print(f"Loading demo data from {DEMO_DIR}...")
+    clusters = _load_json(clusters_path)
+    orders   = _load_json(orders_path)
+    graph    = _load_graph(graph_path) if os.path.exists(graph_path) else None
+
+    print(f"  {len(clusters)} candidate clusters")
+
+    # Extract features (no labels needed for scoring)
+    rows = extract_all(clusters, orders, ground_truth=None, graph=graph)
+    X = np.array([[r[f] for f in FEATURE_COLUMNS] for r in rows])
+
+    print(f"Loading model from {MODEL_PATH}...")
+    model = lgb.Booster(model_file=MODEL_PATH)
+
+    probs = model.predict(X)
+    print(f"  Scored {len(probs)} clusters")
+
+    # Build scored_clusters.json: candidate_clusters schema + risk_probability
+    cluster_by_id = {c["cluster_id"]: c for c in clusters}
+    scored = []
+    for i, row in enumerate(rows):
+        cid = row["cluster_id"]
+        cluster = cluster_by_id[cid].copy()
+        cluster["risk_probability"] = float(probs[i])
+        scored.append(cluster)
+
+    with open(OUT_PATH, "w", encoding="utf-8") as f:
+        json.dump(scored, f, indent=2, ensure_ascii=False)
+
+    print(f"\nScored clusters saved to {OUT_PATH}")
+    print("\nRisk probability summary:")
+    for c in scored:
+        print(f"  {c['cluster_id']:8s}  size={c['size']:3d}  "
+              f"formation={c['formation_type']:15s}  "
+              f"risk={c['risk_probability']:.4f}")
+
+    return scored
+
+
+if __name__ == "__main__":
+    score_demo()

scoring/train_model.py

@@ -0,0 +1,156 @@
+"""
+train_model.py
+--------------
+Train a LightGBM binary risk classifier on the training dataset.
+
+Outputs:
+  scoring/model.lgb         — trained LightGBM model
+  scoring/holdout_split.json — held-out test rows for evaluation (Section 6)
+
+Usage:
+    python scoring/train_model.py
+"""
+
+import json
+import os
+import pickle
+import sys
+
+import lightgbm as lgb
+import numpy as np
+from sklearn.model_selection import train_test_split
+
+# ---------------------------------------------------------------------------
+# Paths
+# ---------------------------------------------------------------------------
+BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+TRAIN_DIR = os.path.join(BASE_DIR, "data", "synthetic_train")
+SCORING_DIR = os.path.join(BASE_DIR, "scoring")
+
+MODEL_PATH   = os.path.join(SCORING_DIR, "model.lgb")
+HOLDOUT_PATH = os.path.join(SCORING_DIR, "holdout_split.json")
+
+# ---------------------------------------------------------------------------
+# Feature columns (must match scoring/features.py)
+# ---------------------------------------------------------------------------
+FEATURE_COLUMNS = [
+    "size",
+    "strong_link_count",
+    "weak_link_count",
+    "strong_link_density",
+    "has_same_contact_edge",
+    "order_count",
+    "total_amount",
+    "avg_order_amount",
+    "return_rate",
+    "chargeback_rate",
+    "shared_coupon_count",
+    "time_span_days",
+    "order_velocity",
+    "formation_type_strong",
+]
+
+RANDOM_STATE = 42
+
+
+def load_data():
+    """Load clusters, orders, ground truth, and graph from the training directory."""
+    sys.path.insert(0, BASE_DIR)
+    from scoring.features import extract_all, _load_json, _load_graph
+
+    clusters_path = os.path.join(TRAIN_DIR, "candidate_clusters.json")
+    orders_path   = os.path.join(TRAIN_DIR, "orders.json")
+    gt_path       = os.path.join(TRAIN_DIR, "ground_truth.json")
+    graph_path    = os.path.join(TRAIN_DIR, "graph.gpickle")
+
+    clusters     = _load_json(clusters_path)
+    orders       = _load_json(orders_path)
+    ground_truth = _load_json(gt_path)
+    graph        = _load_graph(graph_path) if os.path.exists(graph_path) else None
+
+    rows = extract_all(clusters, orders, ground_truth=ground_truth, graph=graph)
+    return rows
+
+
+def train():
+    print("Loading training data...")
+    rows = load_data()
+    print(f"  {len(rows)} candidate clusters loaded")
+
+    X = np.array([[r[f] for f in FEATURE_COLUMNS] for r in rows])
+    y = np.array([r["label"] for r in rows])
+    cluster_ids = [r["cluster_id"] for r in rows]
+    feature_rows = [{f: r[f] for f in FEATURE_COLUMNS} for r in rows]
+
+    pos = y.sum()
+    neg = len(y) - pos
+    print(f"  Labels — positive: {int(pos)}, negative: {int(neg)}")
+
+    # 5.2 — stratified 80/20 split
+    (X_train, X_test, y_train, y_test,
+     ids_train, ids_test,
+     feat_train, feat_test) = train_test_split(
+        X, y, cluster_ids, feature_rows,
+        test_size=0.20,
+        stratify=y,
+        random_state=RANDOM_STATE,
+    )
+    print(f"  Train: {len(X_train)}  Test (holdout): {len(X_test)}")
+    print(f"  Holdout positives: {y_test.sum():.0f}  negatives: {(len(y_test)-y_test.sum()):.0f}")
+
+    # 5.3 — simple LightGBM model (small dataset, keep it simple)
+    model = lgb.LGBMClassifier(
+        num_leaves=7,
+        max_depth=4,
+        min_child_samples=5,
+        n_estimators=100,
+        learning_rate=0.05,
+        objective="binary",
+        random_state=RANDOM_STATE,
+        verbose=-1,
+    )
+    print("Training LightGBM classifier...")
+    model.fit(X_train, y_train)
+    print("  Training complete.")
+
+    # 5.4 — save model
+    model.booster_.save_model(MODEL_PATH)
+    print(f"  Model saved to {MODEL_PATH}")
+
+    # Verify reload
+    loaded = lgb.Booster(model_file=MODEL_PATH)
+    test_pred = loaded.predict(X_test)
+    print(f"  Model reload verified (test pred shape: {test_pred.shape})")
+
+    # 5.5 — feature importances
+    importances = model.feature_importances_
+    importance_pairs = sorted(
+        zip(FEATURE_COLUMNS, importances),
+        key=lambda x: x[1], reverse=True
+    )
+    print("\nFeature importances (gain):")
+    for feat, imp in importance_pairs:
+        print(f"  {feat:30s}: {imp}")
+    if importance_pairs[0][0] == "size" and importance_pairs[0][1] > sum(importances) * 0.5:
+        print("  [NOTE] 'size' dominates feature importance — worth flagging in summary")
+
+    # Save holdout split (cluster IDs + features + labels + probabilities)
+    holdout = []
+    probs = loaded.predict(X_test)
+    for i in range(len(ids_test)):
+        holdout.append({
+            "cluster_id": ids_test[i],
+            "features": feat_test[i],
+            "label": int(y_test[i]),
+            "risk_probability": float(probs[i]),
+        })
+
+    with open(HOLDOUT_PATH, "w", encoding="utf-8") as f:
+        json.dump(holdout, f, indent=2, ensure_ascii=False)
+    print(f"  Holdout split saved to {HOLDOUT_PATH} ({len(holdout)} rows)")
+
+    return model, holdout, importance_pairs
+
+
+if __name__ == "__main__":
+    train()