| Node | Kind | State | Records | Tools | ms |
|---|---|---|---|---|---|
| intake | transform | passed | 1 | 0 | |
| map_terrain | agent | passed | 1 | 6 | 281615 |
| hypothesize | agent | passed | 3 | 0 | 186033 |
| gather_witnesses | agent | failed | 3 | 28 | 3828261 |
| witness_gate | gate | passed | 1 | 0 | |
| adjudicate | agent | passed | 1 | 0 | 56447 |
| citation_gate | gate | passed | 1 | 0 | |
| propose_remediation | agent | passed | 1 | 1 | 159488 |
| human_approval write | human | passed | 1 | 0 | |
| write_back write | tool | passed | 1 | 0 | |
| report | transform | passed | 1 | 0 |
| id | Claim | Source | Supports | Refutes |
|---|---|---|---|---|
| e21 | marts.signups (warehouse view `signups`) is defined as a straight pass-through of the channel column from events_daily with no channel-bucketing CASE/WHEN, WHERE filter, or attribution predicate. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.marts.signups,PROD) |
- | h3 |
| e22 | staging.events_daily (warehouse view `events_daily`) derives channel by GROUP BY on the raw channel column with no re-mapping, merging, or attribution logic that could re-scope the affiliate bucket. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.staging.events_daily,PROD |
- | h3 |
| e23 | metric_definition_history records no definition change for signups or events_daily, and no metric-definition change dated on or near day 14. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.marts.signups,PROD) |
- | h3 |
| e24 | Across the full 36-day window the signups view contains only four channels, affiliate ends exactly at 2026-06-13, and no other channel's day-count or total volume increases to absorb affiliate's prior ~25-30 signups/day. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.marts.signups,PROD) |
- | h3 |
| e25 | Affiliate rows are present in raw.channel_events_raw through 2026-06-13 and entirely absent from 2026-06-14 onward, with no other existing channel gaining affiliate's volume and no new channel label appearing after day 14. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.channel_events_r |
h1 | h3 |
| e26 | The only deploys near day 14 (D-0012 on day 12, D-0020 on day 20) ship unrelated product features, and none modifies channel attribution or the signups/events_daily definition on day 14. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.marts.signups,PROD) |
- | h3 |
| e11 | In the warehouse the metric is derived, not measured: `signups` is `CREATE VIEW signups AS SELECT bucket_date AS day, channel, events AS signups FROM events_daily`, and `events_daily` is `SELECT date(occurred_at) AS bucket_date, channel, COUNT(*) AS events FROM channel_events_raw GROUP BY bucket_date, channel`, so an affiliate signup count is exactly the number of affiliate rows in channel_events_raw for that day, with no channel-remapping CASE/WHEN and no independent user-behaviour input. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.marts.signups,PROD) |
h1, h3 | h2 |
| e12 | In channel_events_raw the affiliate collector's output stops dead at the day-13/14 boundary while every sibling collector keeps writing for three more weeks: affiliate n=299 spans 2026-06-01..2026-06-13 only, whereas organic (n=1039), paid_social (n=1033) and referral (n=1026) each span 2026-06-01..2026-07-06. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.channel_events_r |
h1 | - |
| e13 | raw.web_sessions — an instrument independent of the per-channel collector — shows affiliate-referred demand did NOT collapse after day 14: affiliate sessions are 791 across days 1-13 (~60.8/day) and 1407 across days 14-36 (~61.2/day), all human (1407 of 1407 with is_automated=0), and the per-day series stays in the 41-75 range straight across the boundary (day 13=48, day 14=43, day 15=67) with zero bot sessions on every day. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.web_sessions,PRO |
h1 | h2 |
| e14 | The affiliate signups series is an abrupt cliff, not the decay or noisy near-zero a genuine behavioural loss produces: signups WHERE channel='affiliate' returns one row per day for 2026-06-01..2026-06-13 at a healthy 16-31/day (19,23,22,23,21,16,18,26,23,27,29,31,21) and then NO rows at all from 2026-06-14 onward. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.marts.signups,PROD) |
h1, h3 | h2 |
This page is built from the run ledger alone. It never reads the agent's own summary of what it did.