You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 676eeb1
Browse filesBrowse the repository at this point in the historyBrowse files
Hand a revoked review's reports back to the community gate
The first revoke rebuilt a reviewed row's community reports as one
unbucketed event on the row's last_seen, which included the review date.
A later report on the original day then counted as a day apart, bucketed
reporters lost the three-reporter rule, a row the community had published
on its own was unpublished, and the review's placeholder count of 1
carried into the ledger.
Now the ledger keeps the real events: approving a pending number no longer
pops its state, the quorum recheck keeps the state of a reviewed row whose
only other support is community reports, and reports that arrive while
the review holds a row up are recorded too. A reviewed row is never
demoted by a bucket recheck. On revoke, a row with a ledger state goes
back to the gate (count and dates from its events when the review had
published it), and one without is unpublished. Production data: a no-op
merge stays at v52, revoking #27's number gives v53, a rerun stays put.
Tests: assert_revoked_review_hands_reports_back_to_the_gate covers each
case above. The approval test asserted the ledger forgot the report once
the review published it; that loss is the bug, so it now asserts the
report is kept unpublished. The revoke test's settled row now stays
published, a legacy reviewed row covers the no-ledger path, and an
FCC-only revoke step makes the version bump depend on the revocation
alone (later version numbers shift by one).
0 commit comments