Repair official collection and retire exhausted material tasks
Development CI / check-and-build (push) Successful in 2m59s
GitHub Git mirror / mirror (push) Successful in 6s
Development CI / deploy-development (push) Successful in 5s

This commit is contained in:
eezd committed 2026-11-05 13:30:05 +08:00
1 parent d5d350b7ec
commit d0c2beb9b3
23 files changed
+399 -35

No files matched your search

+18
View File
@@ -413,3 +413,21 @@ callback chain finished, including explicit partial packages; it does not mean
full source quality passed. Exhausted candidate/request budgets remain incomplete.
Mocked tests and real captured responses are separate evidence; neither proves
complete 2025 coverage or authorizes production activation.
### Bounded collection failures (November 2026)
SEC, DOJ and Congress use one durable retry layer (three failed attempts, at least
one hour apart). EIA, DOL, Michigan and ADP retain the same budget per calendar
release in `collection_failures` / `collection_retired_at`. Cooling does not spend
an attempt. Retirement retains evidence and values; new releases are independent.
A source is disabled only after three failed discovery/schedule checks, not for
one document gap. Explicit official photo captions can exclude illustrations from
text extraction. Missing official dates are never replaced by update timestamps.
Migration `db2345678901` is additive and must precede activation. With the crawler
stopped under the deployment lock, preview `python -m
alphapulse.crawler.repair_collection`; append `--execute` once to establish the
repair budget for existing news gaps. It preserves the previous retry evidence
and records a durable initialization marker. Repeating it cannot renew the budget.
Set the authorized SEC contact in the private instance environment before enabling
SEC; do not put the address or credentials in tracked configuration.