Stewardship inbox
Every queue a person works, in one place: 7,334 items open. Answers here are the ground truth every automatic rule is measured against.
| Area | Task | Status |
|---|---|---|
| Outlets | General review queue Every other drawn pair, easiest first. Answers become the labels every rule is scored against. |
2,838 open |
| Outlets | Gaps set Merges no rule has been measured on yet (street variants, department listings, a direction on one side, names that contain each other) and pairs the master did not merge. Each answer measures a rule; a 'different' on a merged pair splits it. |
598 open 67 of 665 answered |
| Outlets | Check set Claude's answers on the strata its merge tier would act on. The tier stays off until these pass. |
100 open 200 of 300 answered |
| Outlets | Review set Where Haiku's screen is most likely wrong, checked before Opus confirms anything. |
52 open 148 of 200 answered |
| Outlets | Next best The pairs the review model expects to learn most from. |
10 open 85 of 95 answered |
| Products | General review queue Every other drawn pair, easiest first. Answers become the labels every rule is scored against. |
2,596 open |
| Products | Barcode pairs Products that share a barcode but may not be one item: a 'different' is a barcode that lied. |
300 open 0 of 300 answered |
| Products | Product join questions Whether two catalogs' products are the same item when no barcode joins them. |
239 open 1 of 240 answered |
| Brands | Brands sharing barcodes Two brands on the same barcodes with no word in common: an owner, an alias, or a wrong barcode. Never merged automatically. (Rulings are recorded from the command line today.) |
311 open 0 of 311 answered |
| Menus | Menu checks Haiku's menu readings checked against the printed page. |
284 open 11 of 295 answered |
| Outlets | Agent set The web agent's answers on held merges. Its 'same' calls merge only while these hold up. |
clear 134 of 134 answered |
| Outlets | Calibration set A fixed random draw every judge is scored against. |
clear 200 of 200 answered |
| Outlets | Recheck Earlier answers that a decided Claude answer contradicts. Every gate is measured on these answers, so a wrong one moves all of them. |
clear 9 of 9 answered |
| Sources | census-migration: degraded the latest 1 run(s) degraded
Suggested: 1. Do nothing urgent. The data is complete. Re-run only the failed states: `python -m hoodie_bev.sources.census_migration --state 18,24,25,26,31,37`. Or wait for the next monthly run.
2. Optional, to stop Census outages from degrading future runs: in sources/census_migration.py `_fetch`/`harvest_state` (around lines 101-104 and 196-199), retry 2-3 times with a pause between tries when a request times out or gets a 5xx error.
3. If the same states keep failing on later runs, check whether api.census.gov is having problems. |
low
since Nov 09 |
| Sources | tax-rates: degraded the latest 7 run(s) degraded
Suggested: Nothing needs fixing in the code or the data. Either accept this as known degraded status, or change one of two things:
(a) Data: if someone can get a sourced effective wine rate for MS/NH/PA/UT/WY, add it to the rate_value column in src/hoodie_bev/sources/data/tax_rates_seed.csv.
(b) Code: so the alert stops firing, change sources/tax_rates.py:114-116 so a blank rate_value on a row with is_control_state=True is logged but not added to `warns`. Keep provenance='unverified' so the gap still shows in the table. Only a missing rate in a non-control state would then mark the run degraded. |
low
since Nov 09 |
| Sources | winebow: failed the latest run failed
|
failed
since Nov 10 |
| Sources | mdm-briefing: degraded the latest 3 run(s) degraded
|
degraded
since Nov 14 |
| Sources | doordash-zip: degraded the latest 1 run(s) degraded
|
degraded
since Nov 16 |
| Sources | ubereats: degraded the latest 1 run(s) degraded
|
degraded
since Nov 16 |
Exceptions — held merges 32,616
A merge a rule found but the master declined. Each is retried every build, so these need no action; they say where the rules stop. One outlet's held merges are listed on its own page.
| Reason | Held | What it means |
|---|---|---|
| same-source-conflict | 31,664 | The merge would put two records of one source in one outlet (two licences, two listings). Usually two operations at one address, sometimes one store written two ways. |
| withdrawn | 499 | A rule that made this link no longer passes it (a record changed, or the rule was narrowed). |
| cluster-too-large | 180 | The outlet would grow past the size cap — a mall or a shared address, not one place. |
| duplicates-too-large | 160 | More listings of one name at one address than any real store has: treated as a data fault. |
| refused | 53 | A person answered 'different' on two of its records; the master will not join them again. |
| duplicate-bridges-sources | 47 | A duplicate listing would weld two outlets that each hold their own record of another source. |
| agent-too-large | 12 | The web agent's 'same' would join outlets past the size cap. |
| premises-too-large | 1 | One licence file with more records than the premises cap. |