loopable / platform /harness /amazon_schema.py
fsanyoto's picture
Deploy AIOS web (React glide grid + FastAPI slice)
3e19b1e verified
Raw History Blame Contribute Delete
45.4 kB
"""amazon_schema.py β€” the pinned field catalogue for the Amazon connector, and the honest map of
which tables actually have one.
python platform/harness/amazon_schema.py --refresh re-pin from amzn/selling-partner-api-models
python platform/harness/amazon_schema.py --report what each table's columns come from
python platform/harness/amazon_schema.py --selftest prove the flattener and the provenance map
⭐⭐ **WHY THIS FILE EXISTS: R6, AND THE TEST IT CHANGED.** Owner ruling R6 (wave 52) gives Amazon
the onboarding disclosure screen fed by Amazon's PUBLISHED report schemas, and amends
[[connector-onboarding-schema]]'s R4 so the test is *"is there a real schema artifact"* rather than
*"is there a live `fields_get`"*. SP-API has no introspection endpoint at all, so a probe is not
available at any price; a versioned, vendor-authored schema file is the artifact.
β›”β›” **AND THE ENUMERATION SAYS ONLY TWO OF SEVEN TABLES HAVE ONE.** Measured 2026-09-04 against
`schemas/reports` at commit `3659f96`: 22 artifacts exist and they are all JSON-format analytics
reports. **Every flat-file (TSV) report we depend on β€” orders, listings, settlements, the FBA
ledger β€” has NO schema artifact anywhere in the repository.** That is not a gap to paper over: a
transcribed column list is exactly what W31's R8 forbids, and `meta_store.py`'s header spends three
paragraphs on why (Graph's 285 columns were ASKED-AND-ACCEPTED, never read off a doc).
⭐ **THE RESOLUTION, AND IT NEEDS NO CREDENTIAL TO BE CORRECT: A FLAT-FILE REPORT CARRIES ITS OWN
HEADER ROW.** Line 1 of the downloaded document IS the column list, straight from Amazon. So those
tables are declared with **no columns at all** until the first sync, and `amazon_store._measure`
reads the header. That is genuinely measured rather than transcribed, it is the same discipline as
Meta's, and it means this connector ships zero invented columns.
**Every table therefore declares its PROVENANCE, and the disclosure screen prints it:**
| provenance | meaning | tables |
|---|---|---|
| `schema_artifact` | columns derived mechanically from a pinned vendor JSON Schema | pricing, brand_analytics |
| `measured_at_sync` | no artifact exists; columns are read from the report document's own header | orders, order_items, listings, settlements, inventory |
β›” **THERE IS NO THIRD STATE, AND `transcribed` IS NOT ONE OF THEM.** It was considered and refused:
a column list copied out of documentation prose looks identical on the screen to one that was
measured, which is [[a-seeds-upload-date-is-not-its-cut-date]] wearing a different hat. A table
with no artifact shows an empty column list and says why.
⚠ **THE CATALOGUE IS A FILE, NOT A LIVE CALL, so it carries `fetched_utc` and the blob sha of every
artifact it was built from** β€” R6's own warning, and the screen must show that date.
β›” NO NETWORK AT IMPORT. `--refresh` is the only path that reaches GitHub; everything else reads
the pinned JSON off disk, so a gate, a screen and a sync all work with no connectivity.
"""
from __future__ import annotations
import argparse
import json
import sys
import urllib.request
from datetime import datetime, timezone
from pathlib import Path
_HERE = Path(__file__).resolve().parent
CATALOGUE = _HERE / "proto" / "amazon-report-schemas.json"
if str(_HERE.parent) not in sys.path:
sys.path.insert(0, str(_HERE.parent))
from harness.amazon_client import GROUP_A_TABLES, DECLARED_DATASETS
from core import connector_spec
_inbound_spec = connector_spec.spec_for("ut_amzn_inbound_shipments")
#: The vendor repository, pinned. ⚠ `main` is what `--refresh` reads, and the resulting file
#: records the exact commit and per-file blob shas it saw, so a later refresh that changes a column
#: list is VISIBLE as a sha change rather than a silent redefinition.
REPO = "amzn/selling-partner-api-models"
RAW = "https://raw.githubusercontent.com/" + REPO + "/{ref}/{path}"
API = "https://api.github.com/repos/" + REPO
#: table -> (artifact path in the vendor repo, the root property holding the row array).
#: β›” ONLY TABLES WITH A REAL ARTIFACT APPEAR HERE. The others are declared in PROVENANCE below
#: with an empty column list on purpose; adding a row here without an artifact is the one edit
#: this file exists to prevent.
ARTIFACTS = {
"pricing": ("schemas/reports/sellerSalesAndTrafficReport.json", "salesAndTrafficByAsin"),
"brand_analytics": ("schemas/reports/sellingPartnerSearchTermsReport.json",
"dataByDepartmentAndSearchTerm"),
}
#: Every table the connector ships, and where its columns come from. The seven are owner ruling
#: R4; the provenance split is the enumeration above.
PROVENANCE = {
"orders": "measured_at_sync",
"order_items": "measured_at_sync",
"listings": "measured_at_sync",
"settlements": "measured_at_sync",
"inventory": "measured_at_sync",
"pricing": "schema_artifact",
"brand_analytics": "schema_artifact",
"economics": "measured_at_sync",
"transactions": "measured_at_sync",
"fba_fee_estimates": "measured_at_sync",
"referral_fees": "measured_at_sync",
"storage_fees": "measured_at_sync",
"search_query_performance": "measured_at_sync",
"sku_keywords": "derived",
"keyword_ranks": "apify_dataset",
"competitor_snapshots": "competitor_run",
"competitor_images": "competitor_run",
"fee_estimates_api": "measured_at_sync",
}
PROVENANCE.update(dict.fromkeys(GROUP_A_TABLES, "measured_at_sync"))
PROVENANCE.update(dict.fromkeys(DECLARED_DATASETS, "declared"))
PROVENANCE.update(dict.fromkeys((t + "_history" for t in connector_spec.AMAZON_HISTORY_SOURCES), "history"))
REVIEW_UNAVAILABLE_NOTE = (
"Amazon's API does not supply written reviews, average star rating or review count. "
"Read them on the product page or in Seller Central.")
SQP_LIMIT_NOTE = "Amazon lists at most 100 searches a month for each product."
#: Human copy for the disclosure screen, per provenance. β›” NO EM DASH OR EN DASH: this reaches a
#: screen and standing rule 2 governs it.
PROVENANCE_COPY = {
"competitor_run": ("Daily competitor facts saved by the competitor run. Every snapshot day is kept. "
"Image rows keep each distinct variant and link, dated when first saved."),
"apify_dataset": ("Apify Amazon search dataset joined to each SKU's latest matching keyword row on or before the capture date. "
"Sponsored rank uses the vendor rank or ordered sponsored placements in the captured dataset. "
"Global position is separate. Unmatched observations remain in the raw mirror."),
"derived": ("Derived from the saved listing items and Search Query Performance databases. "
"ASIN query metrics are shared across matching SKUs without allocation. "
"Aggregate distinct SQP source grains for monthly totals."),
"declared": "This dataset is listed for completeness; its named reason explains why no rows are loaded.",
"schema_artifact": "Columns come from Amazon's published report schema, pinned below.",
"measured_at_sync": ("Columns are measured from the document header or JSON record keys "
"when this database syncs; no column artifact is pinned here."),
"history": ("Every change to the source table. Each row is one version, "
"valid from one date until the next change."),
}
_JSON_SCALAR = {"string": "VARCHAR", "integer": "BIGINT", "number": "DOUBLE", "boolean": "VARCHAR"}
def _snake(path):
"""`salesByAsin.orderedProductSales.amount` -> `sales_by_asin_ordered_product_sales_amount`.
A dotted JSON path becomes one flat snake_case column, because the mirror is a table and the
grid that reads it has no notion of nesting. Collisions are impossible by construction: the
full path is preserved, only its punctuation changes.
"""
out = []
for part in str(path).split("."):
buf = ""
for i, ch in enumerate(part):
if ch.isupper() and i and not part[i - 1].isupper():
buf += "_"
buf += ch.lower()
out.append(buf)
return "_".join(x for x in out if x)
def _resolve(node, defs, seen=()):
"""Follow a local `$ref` chain to its definition. -> `(node, names_followed)`.
⚠ CYCLE-GUARDED, AND THE GUARD HAD TO MOVE ONCE. A first version tracked visited names inside
this function alone, which is useless: `walk` calls it afresh at every level, so `A.a -> A`
resolved cleanly each time and recursed 1000 deep before Python stopped it. **NC-AS4 caught it
on the first run** β€” the check discriminates, which is the whole reason it was written before
the code was trusted. The names followed are now RETURNED so the caller can carry them down
its own recursion.
"""
names = []
n = node or {}
while isinstance(n, dict) and n.get("$ref"):
name = str(n["$ref"]).rsplit("/", 1)[-1]
if name in seen or name in names:
return {}, names
names.append(name)
n = defs.get(name) or {}
return n or {}, names
def flatten(schema, root):
"""One vendor JSON Schema -> an ordered list of `(column, duckdb_type)`.
⭐ MECHANICAL, WITH NO HAND-WRITTEN COLUMN ANYWHERE. That is the whole point of preferring an
artifact to a doc: the list cannot drift from what Amazon published, and a refresh that changes
it changes the pinned sha too.
⚠ An array of objects nested INSIDE a row is stored as compact JSON text, the same call
`meta_store._cell` makes. Dropping it would violate "every field the source returns"; exploding
it would silently multiply the row count.
"""
defs = (schema or {}).get("definitions") or {}
top = ((schema or {}).get("properties") or {}).get(root) or {}
node, first = _resolve(top.get("items") or top, defs)
cols = []
def walk(n, prefix, seen):
n, names = _resolve(n, defs, seen)
seen = seen + tuple(names)
kind = n.get("type")
props = n.get("properties") or {}
if kind == "object" or props:
for name, child in props.items():
walk(child, f"{prefix}.{name}" if prefix else name, seen)
return
if kind == "array":
cols.append((_snake(prefix), "VARCHAR")) # nested rows -> JSON text
return
if not prefix:
return # an unresolvable root yields nothing
cols.append((_snake(prefix), _JSON_SCALAR.get(str(kind), "VARCHAR")))
walk(node, "", tuple(first))
seen, out = set(), []
for c, t in cols:
if c and c not in seen:
seen.add(c)
out.append((c, t))
return out
def vendor_examples(table):
"""AMAZON'S OWN example rows for a `schema_artifact` table, from the pinned artifact. -> [row]
⭐⭐ THIS IS THE ONLY FIXTURE THIS CONNECTOR IS ALLOWED TO HAVE, AND THAT IS THE POINT. A
hand-authored fixture agrees with the parser that reads it by construction, which makes the
gate's ruler the module under test [[a-gate-whose-ruler-is-the-module-under-test]]. These rows
are Amazon's: they ship inside the vendor schema's own `examples` blocks, pinned by the same
blob sha as the column list, so a change to either is visible as a sha change.
⚠ RETURNS [] FOR A `measured_at_sync` TABLE, and that is correct rather than a gap. Those
reports publish no schema and therefore no examples, and inventing a flat-file header here
would be the transcription the whole design refuses.
"""
spec = (catalogue().get("tables") or {}).get(str(table)) or {}
if spec.get("provenance") != "schema_artifact":
return []
return [dict(r) for r in (spec.get("examples") or [])]
def _fetch(url, timeout=45):
req = urllib.request.Request(url, headers={"User-Agent": "AIOS-Loopable/1.0 (+read-only)",
"Accept": "application/json"})
with urllib.request.urlopen(req, timeout=timeout) as r:
return json.loads(r.read().decode("utf-8"))
def refresh(log=print):
"""Re-pin the catalogue from the vendor repository. The ONLY path here that uses the network."""
head = _fetch(f"{API}/commits/main")
ref = str(head["sha"])
log(f"{REPO} @ {ref[:7]} ({head['commit']['committer']['date']})")
listing = {e["path"]: e for e in _fetch(f"{API}/contents/schemas/reports")}
tables = {}
for table, (path, root) in ARTIFACTS.items():
schema = _fetch(RAW.format(ref=ref, path=path))
cols = flatten(schema, root)
blob = (listing.get(path) or {}).get("sha")
# ⭐ THE VENDOR'S OWN EXAMPLE ROWS, pinned beside the columns they belong to. They are
# what `verify_amazon`'s pipeline section drives, so the end-to-end proof runs on Amazon's
# data rather than on a fixture written by the person who wrote the parser.
defs = schema.get("definitions") or {}
top = ((schema.get("properties") or {}).get(root) or {})
node, _refs = _resolve(top.get("items") or top, defs)
examples = [e for e in (node.get("examples") or []) if isinstance(e, dict)]
tables[table] = {
"provenance": "schema_artifact",
"artifact": path,
"artifact_sha": blob,
"root": root,
"columns": [{"name": c, "type": t} for c, t in cols],
"examples": examples,
}
log(f" {table:<16} {len(cols):>3} columns {path.rsplit('/', 1)[-1]}")
for table, prov in PROVENANCE.items():
if prov == "measured_at_sync":
tables[table] = {"provenance": prov, "artifact": None, "artifact_sha": None,
"root": None, "columns": []}
log(f" {table:<16} 0 columns NO ARTIFACT PUBLISHED: measured at first sync")
out = {
"_comment": ("Pinned Amazon report field catalogue. Rebuild with "
"`python platform/harness/amazon_schema.py --refresh`. "
"Never hand-edit: every column here is derived from a vendor schema."),
"repo": REPO,
"commit": ref,
"fetched_utc": datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ"),
"artifacts_in_repo": len(listing),
"tables": tables,
}
CATALOGUE.parent.mkdir(parents=True, exist_ok=True)
CATALOGUE.write_text(json.dumps(out, indent=2) + "\n", encoding="utf-8")
log(f"\nwrote {CATALOGUE}")
return out
def catalogue():
"""The pinned catalogue off disk, or an empty shell when it has never been refreshed."""
try:
result = json.loads(CATALOGUE.read_text(encoding="utf-8"))
except Exception: # noqa: BLE001
result = {"repo": REPO, "commit": None, "fetched_utc": None, "tables": {}}
from core.connector_spec import AMAZON_FIELDS
descriptors = {row.key: row for row in connector_spec.tables_for("amazon")}
for name, provenance in PROVENANCE.items():
key = "ut_amzn_" + name
table = result.setdefault("tables", {}).setdefault(name, {})
table["provenance"] = provenance
if name in DECLARED_DATASETS:
declared = DECLARED_DATASETS[name]
label = declared["label"]
native = ", ".join(declared.get("report_types") or declared.get("operations") or ())
else:
descriptor = descriptors[key]
label, native = descriptor.label, descriptor.native
table[key] = {"label": label, "native": native}
if provenance in {"derived", "apify_dataset", "competitor_run"}:
table["columns"] = [dict(field) for field in AMAZON_FIELDS.get(key, ())
if field["name"] != STORE]
table["note"] = PROVENANCE_COPY[provenance]
table["local_columns"] = [dict(field) for field in AMAZON_FIELDS.get(key, ())
if field["name"] == STORE]
return result
def columns_for(table):
"""The pinned column list for one table, as `[(name, type)]`. Empty when measured at sync."""
if PROVENANCE.get(str(table)) in {"derived", "apify_dataset", "competitor_run"}:
return []
spec = (catalogue().get("tables") or {}).get(str(table)) or {}
return [(c["name"], c.get("type") or "VARCHAR") for c in (spec.get("columns") or [])]
# --------------------------------------------------------------------------- contract C1
#: The one translation from C1's short report keys to the authored database namespace.
#: A-C1-KEYS preserves stored enablement keys and route paths while lane D adopts this map.
DATABASE_PREFIX = "ut_amzn_"
# Internal mirror metadata preserves legacy duplicate ordinals across SQL pages.
IDENTITY_OCCURRENCE = "_amazon_identity_occurrence"
IDENTITY_PID = "_amazon_identity_pid"
CONNECTION_ID = "_connection_id"
STORE = "store"
MIRROR_INTERNAL_COLUMNS = frozenset((IDENTITY_OCCURRENCE, IDENTITY_PID, CONNECTION_ID))
def _table_spec(table, identity, relationships=(), *, grain="entity", line_discriminator=None):
"""Build the common descriptor fields from the LOCAL pin, never a vendor call."""
from core.connector_spec import AMAZON_FIELDS
spec = {
"columns": [{"name": name, "type": kind} for name, kind in columns_for(table)],
"columns_source": {
"path": "platform/harness/proto/amazon-report-schemas.json",
"reader": "amazon_schema.columns_for",
"provenance": PROVENANCE[table],
"assumption": (
"A-QI14: This wave declares from REPORT_TYPES and the pinned artifacts. "
"Measured-at-sync columns are re-measured against the first "
"production sync during the deferred QI13 hotfix (intake RDA6)."),
},
"row_identity": identity,
"row_identity_mode": "composite",
"row_identity_missing": "require_all",
"row_identity_alternatives": (),
"row_identity_status": "declared_pending_first_production_sync",
"grain": grain,
"line_discriminator": line_discriminator,
"line_discriminator_source": ({
"kind": "declared_mirror_metadata",
"document_column": "_report_document_id",
"note": (
"Sync must retain the report document ID and the one-based parsed row position "
"before filtering or deduplication, including identical repeated lines. These "
"are proposed mirror metadata, not measured Amazon header columns. Replaying "
"one document keeps its positions; reconciling overlapping documents is a "
"separate sync responsibility. Never fall back to the order or settlement ID."),
} if line_discriminator else None),
"relationships": relationships,
"provenance": PROVENANCE[table],
"read_through": True,
"row_cap": None,
"read_through_status": "declared_pending_consumer_wiring",
"read_through_reason": (
"Standing rule 1: these are connector records with records_mutable=False. "
"Serve history read-through from the mirror without a record cap."),
}
fields = AMAZON_FIELDS.get(DATABASE_PREFIX + table, ())
spec["columns"].extend(dict(field) for field in fields if field["name"] != STORE)
spec["local_columns"] = [dict(field) for field in fields if field["name"] == STORE]
local_sources = {
"derived": ("listing_items", "search_query_performance"),
"apify_dataset": ("apify:khadinakbar/amazon-search-scraper", "sku_keywords"),
"competitor_run": ("amazon_rivals.run_quotes", "amazon_quotes_loop.run_tenant"),
}
provenance = PROVENANCE[table]
if provenance in local_sources:
spec["columns_source"] = {
"path": "platform/core/connector_spec.py", "reader": "connector_spec.AMAZON_FIELDS",
"provenance": provenance, "assumption": PROVENANCE_COPY[provenance],
"sources": local_sources[provenance],
}
return spec
#: ⭐⭐ TABLE_SPECS IS CONTRACT C1'S SINGLE SOURCE OF TRUTH. One entry per REPORT_TYPES key,
#: checked strictly by verify_amazon. Vendor columns come through columns_for's pinned catalogue.
#: Local projections declare their attribute source through connector_spec.AMAZON_FIELDS.
#: columns_source carries A-QI14 in the
#: descriptor itself, so a consumer cannot mistake this wave's declaration for production evidence.
#:
#: β›” IDENTITIES ARE COMPOSITES, NEVER FALLING PREFERENCES. Every component participates and no
#: proper subset is an alternative identity FOR THAT TABLE. orders is the deduplicated header
#: (_dedupe_orders), while order_items and settlements retain source lines. An order item ID is
#: not guaranteed in the flat-file header, and even identical settlement charges must survive:
#: both line tables therefore declare document + position lineage beside their parent ID. The
#: lineage columns are proposed MIRROR METADATA, not invented vendor columns in the catalogue.
#:
#: ⚠ Identity uniqueness, line metadata and relationship column spellings await the first real
#: sync. All identities say so, including the two tables with published COLUMN artifacts. Nine
#: relationships retain BOTH join-column names and short target keys; the gate compares them to
#: connector_spec.tables_for("amazon"), which amazon_relational still renders. Read-through is a
#: REQUIRED POLICY, not a claim that the existing materialisation consumer already implements it.
TABLE_SPECS = {
"orders": _table_spec("orders", ("amazon_order_id",), (
("sku", "listings", "seller_sku", "many_to_one"),
), grain="order"),
"order_items": _table_spec("order_items", (
"amazon_order_id", "_report_document_id", "_report_row_number",
), (
("amazon_order_id", "orders", "amazon_order_id", "many_to_one"),
("sku", "listings", "seller_sku", "many_to_one"),
), grain="line", line_discriminator="_report_row_number"),
"listings": _table_spec("listings", ("seller_sku",)),
"settlements": _table_spec("settlements", (
"settlement_id", "_report_document_id", "_report_row_number",
), (
("amazon_order_id", "orders", "amazon_order_id", "many_to_one"),
("sku", "listings", "seller_sku", "many_to_one"),
), grain="line", line_discriminator="_report_row_number"),
"inventory": _table_spec("inventory", ("asin", "fnsku", "msku"), (
("msku", "listings", "seller_sku", "many_to_one"),
("asin", "listings", "asin1", "many_to_one"),
)),
# The period, not the parent, is identity: a re-parented child replaces its own window's row.
"pricing": _table_spec("pricing", ("child_asin", "sku", "_period_start", "_period_end"), (
("child_asin", "listings", "asin1", "many_to_one"),
("sku", "listings", "seller_sku", "many_to_one"),
)),
# D-2675: the report period is part of the identity, so each weekly pull of a term keeps its own rows.
"brand_analytics": _table_spec("brand_analytics", (
"department_name", "search_term", "clicked_asin", "click_share_rank",
"_period_start", "_period_end",
)),
"economics": _table_spec("economics", ("start_date", "msku", "marketplace_id"), (
("msku", "listings", "seller_sku", "many_to_one"),
("child_asin", "listings", "asin1", "many_to_one"),
), grain="day_msku"),
"transactions": _table_spec("transactions", ("transaction_id", "item_position"), (
("amazon_order_id", "orders", "amazon_order_id", "many_to_one"),
("sku", "listings", "seller_sku", "many_to_one"),
), grain="transaction_item"),
# S66-T40 (lead, measured live): the fee estimate snapshot has one row per SKU per Amazon
# store, and the referral fee preview spells its SKU `seller_sku`.
"fba_fee_estimates": _table_spec("fba_fee_estimates", ("_snapshot_date", "sku", "amazon_store"), (
("sku", "listings", "seller_sku", "many_to_one"),
), grain="snapshot_sku"),
"referral_fees": _table_spec("referral_fees", ("_snapshot_date", "seller_sku"), (
("seller_sku", "listings", "seller_sku", "many_to_one"),
), grain="snapshot_sku"),
"storage_fees": _table_spec("storage_fees", ("month", "asin", "fnsku", "fulfillment_center"), (
("asin", "listings", "asin1", "many_to_one"),
), grain="month_product_center"),
"search_query_performance": _table_spec("search_query_performance", (
"start_date", "end_date", "asin", "search_query_data_search_query"), (
("asin", "listings", "asin1", "many_to_one"),
), grain="month_asin_search_query"),
}
TABLE_SPECS.update({
"long_term_storage_fees": _table_spec("long_term_storage_fees", ("_period_start", "sku", "fnsku", "asin", "_report_row_number"),
(("sku", "listings", "seller_sku", "many_to_one"), ("asin", "listings", "asin1", "many_to_one"),), grain="month_row"),
"storage_overage_fees": _table_spec("storage_overage_fees", ("_period_start", "storage_type", "_report_row_number"),
(), grain="month_row"),
"reimbursements": _table_spec("reimbursements", ("_period_start", "reimbursement_id", "sku", "asin", "amazon_order_id", "_report_row_number"),
(("sku", "listings", "seller_sku", "many_to_one"), ("asin", "listings", "asin1", "many_to_one"), ("amazon_order_id", "orders", "amazon_order_id", "many_to_one"),), grain="window_row"),
"fba_customer_returns": _table_spec("fba_customer_returns", ("_period_start", "order_id", "sku", "fnsku", "asin", "_report_row_number"),
(("order_id", "orders", "amazon_order_id", "many_to_one"), ("sku", "listings", "seller_sku", "many_to_one"), ("asin", "listings", "asin1", "many_to_one"),), grain="window_row"),
"replacements": _table_spec("replacements", ("_period_start", "replacement_amazon_order_id", "original_amazon_order_id", "sku", "asin", "_report_row_number"),
(("sku", "listings", "seller_sku", "many_to_one"), ("asin", "listings", "asin1", "many_to_one"),), grain="window_row"),
"fba_shipped_sales": _table_spec("fba_shipped_sales", ("_period_start", "amazon_order_id", "sku", "fnsku", "asin", "_report_row_number"),
(("amazon_order_id", "orders", "amazon_order_id", "many_to_one"), ("sku", "listings", "seller_sku", "many_to_one"), ("asin", "listings", "asin1", "many_to_one"),), grain="window_row"),
"fba_shipment_promotions": _table_spec("fba_shipment_promotions", ("_period_start", "amazon_order_id", "sku", "_report_row_number"),
(("amazon_order_id", "orders", "amazon_order_id", "many_to_one"), ("sku", "listings", "seller_sku", "many_to_one"),), grain="window_row"),
"inventory_ledger_detail": _table_spec("inventory_ledger_detail", ("_period_start", "msku", "fnsku", "asin", "_report_row_number"),
(("msku", "listings", "seller_sku", "many_to_one"), ("asin", "listings", "asin1", "many_to_one"),), grain="window_row"),
"inventory_planning": _table_spec("inventory_planning", ("_snapshot_date", "sku", "fnsku", "asin", "_report_row_number"),
(("sku", "listings", "seller_sku", "many_to_one"), ("asin", "listings", "asin1", "many_to_one"),), grain="snapshot_row"),
"manage_inventory": _table_spec("manage_inventory", ("_snapshot_date", "sku", "fnsku", "asin", "_report_row_number"),
(("sku", "listings", "seller_sku", "many_to_one"), ("asin", "listings", "asin1", "many_to_one"),), grain="snapshot_row"),
"reserved_inventory": _table_spec("reserved_inventory", ("_snapshot_date", "sku", "fnsku", "asin", "_report_row_number"),
(("sku", "listings", "seller_sku", "many_to_one"), ("asin", "listings", "asin1", "many_to_one"),), grain="snapshot_row"),
"restock_recommendations": _table_spec("restock_recommendations", ("_snapshot_date", "merchant_sku", "fnsku", "asin", "_report_row_number"),
(("asin", "listings", "asin1", "many_to_one"),), grain="snapshot_row"),
"stranded_inventory": _table_spec("stranded_inventory", ("_snapshot_date", "sku", "fnsku", "asin", "_report_row_number"),
(("sku", "listings", "seller_sku", "many_to_one"), ("asin", "listings", "asin1", "many_to_one"),), grain="snapshot_row"),
"inbound_noncompliance": _table_spec("inbound_noncompliance", ("_period_start", "shipment_id", "sku", "asin", "_report_row_number"),
(("sku", "listings", "seller_sku", "many_to_one"), ("asin", "listings", "asin1", "many_to_one"),), grain="window_row"),
"inventory_expiration_dates": _table_spec("inventory_expiration_dates", ("_snapshot_date", "sku", "fnsku", "_report_row_number"),
(("sku", "listings", "seller_sku", "many_to_one"),), grain="snapshot_row"),
"recommended_removals": _table_spec("recommended_removals", ("_snapshot_date", "sku", "fnsku", "asin", "_report_row_number"),
(("sku", "listings", "seller_sku", "many_to_one"), ("asin", "listings", "asin1", "many_to_one"),), grain="snapshot_row"),
"removal_orders": _table_spec("removal_orders", ("_period_start", "order_id", "sku", "fnsku", "_report_row_number"),
(("sku", "listings", "seller_sku", "many_to_one"),), grain="window_row"),
"removal_shipments": _table_spec("removal_shipments", ("_period_start", "order_id", "sku", "fnsku", "_report_row_number"),
(("sku", "listings", "seller_sku", "many_to_one"),), grain="window_row"),
"subscribe_save_forecast": _table_spec("subscribe_save_forecast", ("_snapshot_date", "sku", "asin", "_report_row_number"),
(("sku", "listings", "seller_sku", "many_to_one"), ("asin", "listings", "asin1", "many_to_one"),), grain="snapshot_row"),
"subscribe_save_performance": _table_spec("subscribe_save_performance", ("_period_start", "sku", "asin", "_report_row_number"),
(("sku", "listings", "seller_sku", "many_to_one"), ("asin", "listings", "asin1", "many_to_one"),), grain="window_row"),
"remote_fulfillment_eligibility": _table_spec("remote_fulfillment_eligibility", ("_snapshot_date", "sku", "asin", "_report_row_number"),
(("sku", "listings", "seller_sku", "many_to_one"), ("asin", "listings", "asin1", "many_to_one"),), grain="snapshot_row"),
"search_catalog_performance": _table_spec("search_catalog_performance", ("_period_start", "asin", "_report_row_number"),
(("asin", "listings", "asin1", "many_to_one"),), grain="month_row"),
"market_basket": _table_spec("market_basket", ("_period_start", "asin", "purchased_with_asin", "_report_row_number"),
(("asin", "listings", "asin1", "many_to_one"),), grain="month_row"),
"repeat_purchase": _table_spec("repeat_purchase", ("_period_start", "asin", "_report_row_number"),
(("asin", "listings", "asin1", "many_to_one"),), grain="month_row"),
"promotion_performance": _table_spec("promotion_performance", ("_period_start", "promotion_id", "_report_row_number"),
(), grain="window_row"),
"coupon_performance": _table_spec("coupon_performance", ("_period_start", "coupon_id", "_report_row_number"),
(), grain="window_row"),
"account_health": _table_spec("account_health", ("_snapshot_date", "marketplace_id", "_json_root", "_report_row_number"),
(), grain="snapshot_row"),
"inactive_listings": _table_spec("inactive_listings", ("_snapshot_date", "seller_sku", "asin1", "_report_row_number"),
(("seller_sku", "listings", "seller_sku", "many_to_one"),), grain="snapshot_row"),
"cancelled_listings": _table_spec("cancelled_listings", ("_snapshot_date", "seller_sku", "asin1", "_report_row_number"),
(("seller_sku", "listings", "seller_sku", "many_to_one"),), grain="snapshot_row"),
"suppressed_listings": _table_spec("suppressed_listings", ("_snapshot_date", "sku", "asin", "_report_row_number"),
(("sku", "listings", "seller_sku", "many_to_one"), ("asin", "listings", "asin1", "many_to_one"),), grain="snapshot_row"),
"listing_defects": _table_spec("listing_defects", ("_snapshot_date", "sku", "_report_row_number"),
(("sku", "listings", "seller_sku", "many_to_one"),), grain="snapshot_row"),
"sold_listings": _table_spec("sold_listings", ("_snapshot_date", "sku", "_report_row_number"),
(("sku", "listings", "seller_sku", "many_to_one"),), grain="snapshot_row"),
"mfn_returns": _table_spec("mfn_returns", ("_period_start", "order_id", "sku", "asin", "_report_row_number"),
(("order_id", "orders", "amazon_order_id", "many_to_one"), ("sku", "listings", "seller_sku", "many_to_one"), ("asin", "listings", "asin1", "many_to_one"),), grain="window_row"),
"b2b_recommended": _table_spec("b2b_recommended", ("_snapshot_date", "asin", "_report_row_number"),
(("asin", "listings", "asin1", "many_to_one"),), grain="snapshot_row"),
"b2b_not_yet_on_amazon": _table_spec("b2b_not_yet_on_amazon", ("_snapshot_date", "ean", "upc", "part_number", "_report_row_number"),
(), grain="snapshot_row"),
"geo_opportunities": _table_spec("geo_opportunities", ("_snapshot_date", "asin", "_report_row_number"),
(("asin", "listings", "asin1", "many_to_one"),), grain="snapshot_row"),
"out_of_stock_mfn": _table_spec("out_of_stock_mfn", ("_snapshot_date", "sku", "asin", "_report_row_number"),
(("sku", "listings", "seller_sku", "many_to_one"), ("asin", "listings", "asin1", "many_to_one"),), grain="snapshot_row"),
"b2b_quotes": _table_spec("b2b_quotes", ("_snapshot_date", "sku", "asin", "_report_row_number"),
(("sku", "listings", "seller_sku", "many_to_one"), ("asin", "listings", "asin1", "many_to_one"),), grain="snapshot_row"),
"fee_discounts": _table_spec("fee_discounts", ("_snapshot_date", "sku", "asin", "_report_row_number"),
(("sku", "listings", "seller_sku", "many_to_one"), ("asin", "listings", "asin1", "many_to_one"),), grain="snapshot_row"),
"order_metrics": _table_spec("order_metrics", ("_period_start", "interval", "_report_row_number"),
(), grain="window_row"),
"fba_inventory": _table_spec("fba_inventory", ("_snapshot_date", "seller_sku", "fn_sku", "asin", "_report_row_number"),
(("seller_sku", "listings", "seller_sku", "many_to_one"), ("asin", "listings", "asin1", "many_to_one"),), grain="snapshot_row"),
"listing_items": _table_spec("listing_items", ("sku",),
(("sku", "listings", "seller_sku", "many_to_one"),), grain="snapshot_row"),
"catalog_items": _table_spec("catalog_items", ("_snapshot_date", "asin", "_report_row_number"),
(("asin", "listings", "asin1", "many_to_one"),), grain="snapshot_row"),
"fee_estimates_api": _table_spec("fee_estimates_api", ("estimate_id",),
(("sku", "listings", "seller_sku", "many_to_one"),), grain="price_estimate"),
"outbound_fulfillment": _table_spec("outbound_fulfillment", ("seller_fulfillment_order_id", "seller_fulfillment_order_item_id"),
(("seller_sku", "listings", "seller_sku", "many_to_one"),), grain="fulfillment_order_item"),
"inbound_shipments": _table_spec("inbound_shipments", _inbound_spec.identity,
tuple((column, target.removeprefix(DATABASE_PREFIX), field, cardinality)
for column, target, field, cardinality in _inbound_spec.links), grain="inbound_shipment_item"),
"replenishment": _table_spec("replenishment", ("metric_id",),
(("sku", "listings", "seller_sku", "many_to_one"),), grain="replenishment_metric"),
"review_topics": _table_spec("review_topics", ("asin", "marketplace_id", "sort_by", "sentiment", "topic"),
(("asin", "listings", "asin1", "many_to_one"),), grain="asin_sort_sentiment_topic"),
"review_trends": _table_spec("review_trends", ("asin", "marketplace_id", "sentiment", "topic", "month"),
(("asin", "listings", "asin1", "many_to_one"),), grain="asin_sentiment_topic_month"),
"browse_node_review_topics": _table_spec("browse_node_review_topics",
("browse_node_id", "marketplace_id", "sort_by", "sentiment", "topic"), grain="browse_node_sort_sentiment_topic"),
"sku_keywords": _table_spec("sku_keywords", ("keyword_id",),
(("sku", "listing_items", "sku", "many_to_one"),), grain="sku_period_asin_query"),
"keyword_ranks": _table_spec("keyword_ranks", ("rank_id",),
(("keyword_id", "sku_keywords", "keyword_id", "many_to_one"),), grain="keyword_placement_per_sku"),
"competitor_snapshots": _table_spec("competitor_snapshots", ("asin", "snapshot_date"),
grain="asin_day"),
"competitor_images": _table_spec("competitor_images", ("asin", "variant", "link"),
grain="asin_variant_link"),
})
# A history table is one row per version, identified by its version key and first valid date.
TABLE_SPECS.update({t + "_history": _table_spec(t + "_history", ("_history_key", "valid_from"), grain="version")
for t in connector_spec.AMAZON_HISTORY_SOURCES})
def row_key_columns_for(table, columns=()):
"""Return declared identity columns, qualifying report positions by their document."""
identity = TABLE_SPECS[table]["row_identity"]
if table == "order_items" and "order_item_id" in columns:
identity = ("order_item_id", *identity)
if ("_report_row_number" in identity and "_report_document_id" not in identity
and "_report_document_id" in columns):
identity = (*identity, "_report_document_id")
return (CONNECTION_ID, *identity) if CONNECTION_ID in columns else identity
def row_fallback_columns_for(table):
"""Return source lineage used when a declared identity is incomplete."""
coordinates = ("_report_document_id", "_report_row_number")
return ("order_item_id", *coordinates) if table == "order_items" else coordinates
def row_identity_for(table, row):
"""Evaluate the declared full composite. Missing lineage is an error, never a parent ID.
This is a descriptor evaluator, not a production uniqueness measurement or a sync adapter.
Return a tuple so delimiters inside values cannot merge two different component sequences.
"""
spec = TABLE_SPECS[table]
if spec["row_identity_mode"] != "composite" or spec["row_identity_missing"] != "require_all":
raise ValueError(f"{table}: unsupported row identity policy")
identity = spec["row_identity"]
if not identity or spec["row_identity_alternatives"]:
raise ValueError(f"{table}: exactly one complete row identity is required")
if CONNECTION_ID in row:
identity = (CONNECTION_ID, *identity)
missing = [name for name in identity if row.get(name) in (None, "")]
if missing:
raise ValueError(f"{table}: missing declared row identity components: {', '.join(missing)}")
return tuple(row[name] for name in identity)
def disclosure():
"""What the onboarding screen shows: one row per table, with its provenance and sync date.
⚠ `synced` IS THE CATALOGUE'S FETCH DATE, NOT TODAY. R6 is explicit that an unstamped
catalogue is [[a-seeds-upload-date-is-not-its-cut-date]] waiting to happen, and the screen is
the only place a stale pin becomes visible.
"""
cat = catalogue()
rows = []
for table in PROVENANCE:
spec = (cat.get("tables") or {}).get(table) or {}
prov = spec.get("provenance") or PROVENANCE[table]
rows.append({
"table": f"ut_amzn_{table}",
"provenance": prov,
"note": PROVENANCE_COPY[prov],
"columns": len(TABLE_SPECS[table]["columns"] if prov == "derived" else spec.get("columns") or []),
"artifact": spec.get("artifact"),
})
return {"source": "Amazon", "synced": cat.get("fetched_utc"), "commit": cat.get("commit"),
"repo": cat.get("repo"), "tables": rows}
# --------------------------------------------------------------------------- selftest
def selftest(log=print):
"""Prove the flattener, the provenance map and the disclosure shape. No network.
β›” EVERY CHECK BELOW WAS DRIVEN RED BEFORE IT WAS TRUSTED. Green on a first run proves a check
exists, never that it discriminates [[revert-one-file-to-see-if-a-gate-discriminates]].
"""
fails = []
def ck(name, cond, detail=""):
"""Detail is ASCII-escaped: a failure message carrying a hostile byte must not kill the
run. Same reasoning, and the same incident, as `amazon_store.selftest`'s reporter."""
if cond:
log(f" ok {name}")
return
try:
shown = str(detail).encode("ascii", "backslashreplace").decode("ascii")
except Exception: # noqa: BLE001
shown = "<undisplayable>"
log(f" FAIL {name} {shown}")
fails.append(name)
log("amazon_schema selftest")
# NC-AS1..3 β€” the flattener, against a miniature of the real artifact shape.
mini = {
"properties": {"rows": {"type": "array", "items": {"$ref": "#/definitions/Row"}}},
"definitions": {
"Row": {"type": "object", "properties": {
"parentAsin": {"type": "string"},
"salesByAsin": {"$ref": "#/definitions/Sales"},
"tags": {"type": "array", "items": {"type": "string"}}}},
"Sales": {"type": "object", "properties": {
"unitsOrdered": {"type": "integer"},
"orderedProductSales": {"$ref": "#/definitions/Amount"}}},
"Amount": {"type": "object", "properties": {
"amount": {"type": "number"}, "currencyCode": {"type": "string"}}},
},
}
got = dict(flatten(mini, "rows"))
ck("NC-AS1 refs resolve through two levels",
"sales_by_asin_ordered_product_sales_amount" in got, sorted(got)[:6])
ck("NC-AS2 scalar types map to DuckDB types",
got.get("sales_by_asin_units_ordered") == "BIGINT"
and got.get("sales_by_asin_ordered_product_sales_amount") == "DOUBLE"
and got.get("parent_asin") == "VARCHAR",
{k: got.get(k) for k in ("parent_asin", "sales_by_asin_units_ordered")})
ck("NC-AS3 a nested array becomes one JSON text column",
got.get("tags") == "VARCHAR" and len(got) == 5, sorted(got))
# NC-AS4 β€” a cycle must terminate rather than hang.
cyc = {"properties": {"rows": {"$ref": "#/definitions/A"}},
"definitions": {"A": {"type": "object", "properties": {"a": {"$ref": "#/definitions/A"},
"x": {"type": "string"}}}}}
ck("NC-AS4 a self-referential schema terminates", dict(flatten(cyc, "rows")).get("x") == "VARCHAR")
# NC-AS5 β€” ARTIFACTS may never claim a table whose provenance is measured-at-sync. This is the
# check that stops the one edit this file exists to prevent.
bad = [t for t in ARTIFACTS if PROVENANCE.get(t) != "schema_artifact"]
ck("NC-AS5 every ARTIFACTS row is declared schema_artifact", not bad, bad)
# NC-AS6 β€” and the converse: a schema_artifact table with no artifact is a lie on the screen.
orphan = [t for t, p in PROVENANCE.items() if p == "schema_artifact" and t not in ARTIFACTS]
ck("NC-AS6 every schema_artifact table has an artifact", not orphan, orphan)
# NC-AS7 β€” the seven tables of R4, no more and no fewer.
ck("NC-AS7 table and declared dataset keys cover group (a)",
set(PROVENANCE) == set(TABLE_SPECS) | set(DECLARED_DATASETS), sorted(PROVENANCE))
# NC-AS8 β€” standing rule 2 governs PROVENANCE_COPY: it reaches the disclosure screen.
dashed = [k for k, v in PROVENANCE_COPY.items() if "β€”" in v or "–" in v]
ck("NC-AS8 no em dash or en dash in screen copy", not dashed, dashed)
# NC-AS9 β€” disclosure() names every table and carries the sync stamp key even when unpinned.
d = disclosure()
ck("NC-AS9 disclosure covers all datasets and carries a synced field",
len(d["tables"]) == len(PROVENANCE) and "synced" in d and all(r["table"].startswith("ut_amzn_")
for r in d["tables"]))
# NC-AS10 β€” a measured_at_sync table must ship ZERO columns. An invented column is the defect.
cat = catalogue()
invented = [t for t, s in (cat.get("tables") or {}).items()
if s.get("provenance") == "measured_at_sync" and (s.get("columns") or [])]
ck("NC-AS10 no measured_at_sync table carries invented columns", not invented, invented)
log(f"\n{'GREEN' if not fails else 'RED'}: {10 - len(fails)}/10")
return 0 if not fails else 1
def main(argv=None):
ap = argparse.ArgumentParser(description=__doc__.splitlines()[0])
ap.add_argument("--refresh", action="store_true", help="re-pin from the vendor repository")
ap.add_argument("--report", action="store_true", help="print the provenance map")
ap.add_argument("--selftest", action="store_true")
a = ap.parse_args(argv)
if a.selftest:
return selftest()
if a.refresh:
refresh()
return 0
d = disclosure()
print(f"Amazon field catalogue repo {d['repo']} commit {str(d['commit'])[:7]} "
f"synced {d['synced']}")
for r in d["tables"]:
print(f" {r['table']:<24} {r['provenance']:<18} {r['columns']:>3} columns "
f"{r['artifact'] or ''}")
return 0
if __name__ == "__main__":
sys.exit(main())