The anomaly detector has reported nothing for a month

MTI-006 · graph metric-truth · 15 ledger records

What this run caught

Allowlist violation1 call(s) to a tool this node never declared

Adjudication

Agent verdict
instrument_failure
detector_misaimed
Answer key
instrument_failure
detector_misaimed
Outcome
correct
14 of 14 citations resolve

Graph execution

NodeKindStateRecordsToolsms
intake transform passed 1 0
map_terrain agent passed 1 9 890501
hypothesize agent passed 3 0 330345
gather_witnesses agent passed 3 27 5045591
witness_gate gate passed 1 0
adjudicate agent passed 1 1 425813
citation_gate gate passed 1 0
propose_remediation agent passed 1 1 127507
human_approval write human passed 1 0
write_back write tool passed 1 0
report transform passed 1 0

Witness ledger

idClaimSourceSupportsRefutes
e21 metric_definition_history holds no definition record for any orders metric within days 1-36; its only row defines active_users_daily on day 1. warehouse_query
warehouse://MTI-006/metric_definition_history
- h3
e22 schema_migration_log records no legacy-to-current orders cutover within days 1-36; its only entry is an index creation on day 2. warehouse_query
warehouse://MTI-006/schema_migration_log
- h3
e23 release_deploy_log inside the window contains only two product-feature deploys and no metric or model redefinition. warehouse_query
warehouse://MTI-006/release_deploy_log
- h3
e24 The current orders model is a pure passthrough view (SELECT * FROM orders) with no filter, bucket, or grain transformation that could exclude incident rows. warehouse_query
warehouse://MTI-006/orders_current
- h3
e25 DataHub entity staging.orders_current states it replaced the legacy table but carries no dated redefinition aspect and no attached assertion; its health is INCIDENTS:PASS. datahub_schema
urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.staging.orders_current,PR
h3 -
e26 orders_legacy_deprecated is frozen at day 3 (rows only for days 1-3, n=60) and has a reduced schema (order_id, day only; no amount or customer_code). warehouse_query
warehouse://MTI-006/orders_legacy_deprecated
- h3
e1 The DataHub entity for staging.orders_current (the inferred detector target) carries no assertion/anomaly-detector aspect and no declared checks; its only health signal is INCIDENTS:PASS and relatedDocuments total is 0. datahub_assertion
urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.staging.orders_current,PR
h1 -
e2 The DataHub entity for the upstream raw.orders likewise exposes no assertion aspect and no declared checks; only INCIDENTS:PASS, relatedDocuments total 0. datahub_assertion
urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.orders,PROD)
h1 -
e3 The relation backing staging.orders_current is a pure passthrough view, CREATE VIEW orders_current AS SELECT * FROM orders, over columns (order_id, day, is_weekend, amount REAL, customer_code); it is not a frozen or constant relation and moves whenever the underlying orders table moves. warehouse_query
urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.staging.orders_current,PR
- h1
e4 Across the entire day 1–36 window, staging.orders_current (orders_current) shows a stable, non-anomalous series with no outlier day: per-day row count 45–90, avg amount ~113–154, min ~18, max ~232–240, distinct order_id equal to row count (no duplication), and zero null amounts on days 23–36. warehouse_query
urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.staging.orders_current,PR
h2, h3 h1
e11 Every orders_ingest_job run recorded in ingest_job_runs for days 1–22 has status 'ok' with rows_written=0 — no re-pull, overwrite, carry-forward fill, or duplicated partition is logged. warehouse_query
urn:li:dataJob:(urn:li:dataFlow:(airflow,witnessgraph,PROD),orders_ingest_job)
- h2
e12 In the orders table backing raw.orders, every day 1–36 has COUNT(*) equal to COUNT(DISTINCT order_id), and total rows 2568 = distinct order_id 2568, so no partition is duplicated or carried forward. warehouse_query
urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.orders,PROD)
- h2
e13 Daily row counts, totals and averages in the orders table vary day to day and follow a weekend cycle (is_weekend=1 days have markedly lower volume) rather than forming an artificially smooth/constant series. warehouse_query
urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.orders,PROD)
- h2
e14 An independently-written nightly_orders_snapshot reproduces the orders table's per-day order count and revenue exactly for all 36 days, with written_at_day equal to day for every row (no repeated or lagging write day). warehouse_query
urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.orders,PROD)
- h2
e15 orders_current, the dbt staging model the detector is anchored on, is an unfiltered passthrough view 'CREATE VIEW orders_current AS SELECT * FROM orders' with no aggregation, filter, or masking layer between the raw orders rows and the detector's input. warehouse_query
urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.staging.orders_current,PR
- h2

This page is built from the run ledger alone. It never reads the agent's own summary of what it did.