| Node | Kind | State | Records | Tools | ms |
|---|---|---|---|---|---|
| intake | transform | passed | 1 | 0 | |
| map_terrain | agent | passed | 1 | 10 | 490307 |
| hypothesize | agent | passed | 3 | 0 | 194540 |
| gather_witnesses | agent | passed | 3 | 29 | 5111976 |
| witness_gate | gate | passed | 1 | 0 | |
| adjudicate | agent | passed | 1 | 0 | 265843 |
| citation_gate | gate | passed | 1 | 0 | |
| propose_remediation | agent | passed | 1 | 2 | 311820 |
| human_approval write | human | passed | 1 | 0 | |
| write_back write | tool | passed | 1 | 0 | |
| report | transform | passed | 1 | 0 |
| id | Claim | Source | Supports | Refutes |
|---|---|---|---|---|
| e1 | The marts.checkout_completions metric is a SQL view defined as `SELECT day, browser_version, COUNT(*) AS completions FROM checkout_events WHERE step='completed' GROUP BY day, browser_version`, containing no user-agent parsing, browser_version normalization, or bucket-relabeling logic. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.marts.checkout_completion |
- | h1 |
| e2 | raw.checkout_events stores browser_version directly as a NOT NULL TEXT column and has no user_agent (raw UA) column; its full column set is event_id, day, step, browser_version, customer_code. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.checkout_events, |
- | h1 |
| e3 | Across day 1–36 exactly four browser_version values appear among completed checkouts — chrome-127, chrome-126, firefox-128, safari-17 — each present from first_day=1 through last_day=36, with no new, unknown, or null browser_version bucket appearing at any point. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.checkout_events, |
- | h1 |
| e4 | In every 6-day period from day 1 to 36 there are exactly 4 distinct browser_version values, and total completed checkouts per period are flat (349, 361, 398, 364, 386, 373) with no period showing a shortfall. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.checkout_events, |
- | h1 |
| e5 | Per-browser completions per 6-day period are approximately flat for all four browsers over day 1–36, so no browser_version bucket falls ~40% from its own baseline. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.marts.checkout_completion |
- | h1 |
| e6 | For the browser_version values fully returned (chrome-126, chrome-127, firefox-128), started/payment/completed event counts per 6-day period all remain stable across day 1–36, with no funnel step collapsing for any browser. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.checkout_events, |
- | h1 |
| e11 | The marts.checkout_completions metric is defined as COUNT(*) of raw.checkout_events rows filtered to step='completed', grouped by day and browser_version — a count of completion-tagged events, not of downstream-confirmed orders. | config urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.marts.checkout_completion |
- | - |
| e12 | In marts.checkout_completions, day 19-36 vs day 1-18 summed completions are safari-17 177 vs 214 (-17%), chrome-126 303 vs 328, chrome-127 339 vs 295, firefox-128 304 vs 271; the all-bucket total is 1123 vs 1108, i.e. flat, so there is no real aggregate ~40% loss even in the recorded metric. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.marts.checkout_completion |
h1 | h2 |
| e13 | Confirmed orders in raw.orders — keyed by customer_code and carrying no browser_version column, so immune to any user-agent tagging break — are flat over the window for the cohort of customers who ever checked out on safari-17 (1128 in day 19-36 vs 1147 in day 1-18) and in total (1271 vs 1297). | warehouse_query urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.orders,PROD) |
- | h2 |
| e14 | Attributing raw.orders to browser via web_sessions, day 19-36 vs day 1-18 order counts are flat for every browser including safari-17 (1263 vs 1291), giving no per-browser real drop. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.orders,PROD) |
- | h2 |
| e15 | For safari-17 in raw.checkout_events, weekly (6-day) started and payment counts hold roughly steady while the payment->completed rate falls only modestly from ~55% (weeks 1-3) to ~49% (weeks 4-6), showing no sharp onset-coincident late-step funnel collapse of the magnitude h2 predicts. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.checkout_events, |
- | h2 |
| e21 | marts.checkout_completions is a plain pass-through view defined as SELECT day, browser_version, COUNT(*) AS completions FROM checkout_events WHERE step='completed' GROUP BY day, browser_version, so its completion counts equal raw checkout_events row counts with no filtering or bucketing in between. | datahub_schema urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.marts.checkout_completion |
- | - |
| e22 | In raw.checkout_events, completed-step row counts for every browser_version are flat or rising across the incident window, and no browser falls ~40%. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.checkout_events, |
- | h3 |
| e23 | The independent orders table is flat across the window and roughly agrees with flat raw completed checkouts, so raw.checkout_events is not deficient in completions relative to an independent source of truth. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.checkout_events, |
- | h3 |
| e24 | ingest_job_runs records only status='ok' runs for every logged job across all 36 days, with zero rejected/missing/dead-lettered runs. | run_log urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.checkout_events, |
- | h3 |
| e25 | DataHub lineage UPSTREAM of raw.checkout_events returns zero upstream entities, so the producer/ingest job that writes the table is not catalogued and cannot be inspected via metadata. | datahub_lineage urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.checkout_events, |
- | - |
| e26 | All four browser_version values carry completed-step rows spanning the full window (first_day=1, last_day=36) with non-null, well-formed browser_version values; no new/unknown/truncated bucket appears. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.checkout_events, |
- | h3 |
This page is built from the run ledger alone. It never reads the agent's own summary of what it did.