| Node | Kind | State | Records | Tools | ms |
|---|---|---|---|---|---|
| intake | transform | passed | 1 | 0 | |
| map_terrain | agent | passed | 1 | 9 | 523419 |
| hypothesize | agent | passed | 3 | 0 | 265537 |
| gather_witnesses | agent | passed | 3 | 27 | 4817982 |
| witness_gate | gate | passed | 1 | 0 | |
| adjudicate | agent | passed | 1 | 0 | 290462 |
| citation_gate | gate | passed | 1 | 0 | |
| propose_remediation | agent | passed | 1 | 5 | 505127 |
| human_approval write | human | passed | 1 | 0 | |
| write_back write | tool | passed | 1 | 0 | |
| report | transform | passed | 1 | 0 |
| id | Claim | Source | Supports | Refutes |
|---|---|---|---|---|
| e11 | The signups metric is a pure pass-through row count of raw.channel_events_raw with no filter, join, or bot/is_automated exclusion: signups = events_daily = COUNT(*) of channel_events_raw grouped by date(occurred_at) and channel. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.marts.signups,PROD) |
h2 | - |
| e12 | All four acquisition channels are present in raw.channel_events_raw across the entire day-1→day-36 window; none dropped to zero or went dark. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.channel_events_r |
- | h2 |
| e13 | Per-channel weekly counts in raw.channel_events_raw are continuous and flat-to-rising across full weeks 0-4, with no channel stepping down to zero at any date in the window. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.channel_events_r |
- | h2 |
| e14 | channel_ingest_job ran every day of the window with zero missing runs and never wrote zero rows, so there is no silently dead ingest task feeding channel_events_raw. | run_log urn:li:dataJob:(urn:li:dataFlow:(airflow,witnessgraph,PROD),channel_ingest_job |
- | h2 |
| e15 | No schema migration in the window altered the channel-ingest pipeline; the sole migration (M-0001, day 2) created an index on the unrelated orders table. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.channel_events_r |
- | h2 |
| e16 | The warehouse signups metric fed by this upstream shows week-over-week growth across full weeks 0-4, not a 24% decline, so the reported drop is not present in the raw→signups path. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.marts.signups,PROD) |
h3 | h2 |
| e21 | The marts `signups` view is a pure pass-through of events_daily: it renames columns and applies no filter, join, or aggregation of its own, so it cannot under-count rows it receives from events_daily. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.marts.signups,PROD) |
- | h3 |
| e22 | The `events_daily` view is `SELECT date(occurred_at) AS bucket_date, channel, COUNT(*) AS events FROM channel_events_raw GROUP BY bucket_date, channel` — a plain COUNT(*) of raw rows grouped by UTC calendar day and channel, with NO WHERE filter and NO JOIN, so there is no join/filter key for a schema migration to silently break. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.staging.events_daily,PROD |
- | h3 |
| e23 | channel_events_raw.occurred_at is declared UTC (`TEXT NOT NULL -- ISO8601 in UTC`) and there is a `day INTEGER -- UTC day` column, so events_daily's `date(occurred_at)` bucketing operates on UTC and applies no local-time/DST conversion anywhere in the path — contradicting the DataHub blind-spot note that staging 'buckets on local time'. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.channel_events_r |
- | h3 |
| e24 | The signups mart's weekly totals over the window are 593, 538, 608, 683, 732 for the five full weeks (2026-06-01..07-05) and 105 for week 6, which spans only the single day 2026-07-06 (first_date=last_date), across 4 channels every week. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.marts.signups,PROD) |
- | - |
| e25 | A per-week COUNT(*) of channel_events_raw yields 593, 538, 608, 683, 732, 105 for weeks 1-6 with bad_dates=0 (no NULL date(occurred_at)) every week — identical to the signups mart's weekly totals (e24) — so every raw row is counted in the mart and no rows are lost or misbucketed across the raw→mart transformation; whatever pattern the metric shows is already present in the raw input. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.channel_events_r |
h1, h2 | h3 |
| e26 | The only schema migration recorded in the window is M-0001 on day 2, `CREATE INDEX idx_orders_customer ON orders(customer_code)`, which targets the `orders` table — not in the signups upstream path (channel_events_raw → events_daily → signups) — and alters no column, type, join key, or filter that the signups computation depends on. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.marts.signups,PROD) |
- | h3 |
| e1 | Weekly session counts in raw.web_sessions stayed within ~400–456 for every channel across days 1–35 with no dip in the Jun 8–14 week, so top-of-funnel traffic did not fall when signups fell. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.web_sessions,PRO |
h2, h3 | h1 |
| e2 | In marts.signups the week-over-week decline into Jun 9–15 is confined to organic (133→74) and referral (156→115) while affiliate (145→179) and paid_social (157→166) rose, so the drop is channel-concentrated rather than broad-based. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.marts.signups,PROD) |
h2 | h1 |
| e3 | In raw.channel_events_raw the two declining channels have absent and near-zero days (organic missing 2026-06-10; Jun-08=6, Jun-09=6, Jun-15=3 vs ~20 baseline; referral missing 2026-06-13) while affiliate and paid_social have a continuous non-zero row for every day Jun 7–17. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:postgres,witnessgraph.raw.channel_events_r |
h2, h3 | h1 |
| e4 | marts.signups is a pass-through of events_daily, which is COUNT(*) of channel_events_raw grouped by date and channel with no signup-specific filter and no pricing-page or funnel input, so the metric cannot observe pricing-page viewing behavior and a fall in it is equally consistent with fewer rows landing. | warehouse_query urn:li:dataset:(urn:li:dataPlatform:dbt,witnessgraph.marts.signups,PROD) |
h3 | h1 |
This page is built from the run ledger alone. It never reads the agent's own summary of what it did.