RESOLVED: recall_search outage (feedback #137) — Meili chunks index restored + backfilled 19→11,694 docs via flower:reindex-from-vectors; recall_search verified returning rich hits. Safe for ops to close #137.

Follow-up closure for the recall_search total-outage (feedback #137 / route_feedback signal 149 / decision #71). Sequence: (1) diagnosed the running Herd Meili (:7700) was serving a stale April-20 data-dir with no `chunks` index; escalated as decision #71. (2) Operator fixed Meili during the window (chunks index reappeared on the correct instance — my fresh 02:29 reset segment became searchable + it accepted a write at 03:06 — but only 19 docs, i.e. freshly recreated + repopulating). (3) With the instance confirmed correct, ran `flower:reindex-from-vectors` on MAIN: indexed 11,694/12,086 chunks from MySQL-stored vectors, no LLM calls, 392 skipped (missing/stale vector pairs — the live embed pipeline will backfill those). (4) Verified: chunks index = 11,694 docs; recall_search('daemon reset make-before-break baton handoff') returns 4 high-score hits across docs/scratchpads. Withdrew decision #71 (moot). recall_search is fully restored. NOTE: the Solo timer-bloat (fb #138) is a SEPARATE, still-open issue — the operator fixed Meili without restarting Solo, so timer ops remain broken; tracked in #138 + decision #72.

{
    "command": "flower:reindex-from-vectors",
    "chunks_after": 11694,
    "chunks_before": 19,
    "related_signal": 149,
    "resolves_feedback": 137,
    "withdrew_decision": 71,
    "chunks_total_mysql": 12086
}

created

17h ago

triaged

—

resolved

—

resolved by

—