lisa/control-suggestion-state-api

[ccf-lisa]

automated implementation by lisa.

[gusfcarvalho]

Findings inline (1 Medium, 2 Low — non-blocking). Solid change overall: cross-cell aggregation is order-independent and well-tested, latest-per-control supersession is correct, and writes are transaction-gated. Headline: outcome is derived from newly-inserted rows rather than from whether the LLM matched the control, so capped/excluded/rerun-deduped controls record no_match despite a real match — worth confirming the UI cross-references pending suggestions. Two redundant indexes noted.

[gusfcarvalho]

All three findings from the prior round are addressed in fbe957b and verified:

• Medium (outcome derivation) — recordControlResultsTx now decides matched from matchedControlKeySet(validation.Mappings) (LLM-matched controls) rather than insert success, while SuggestionCount stays on newly-inserted counts. A control whose mappings were capped / excluded / rerun-deduped now correctly records outcome: matched, suggestion_count: 0, with a new test covering the run-cap case. No-match upserts still DO NOTHING, so a later empty cell cannot downgrade an earlier match.
• Low — redundant index dropped from both RunID and Outcome gorm tags; the composite unique index (run_id, control_catalog_id, control_id) is intact for the upsert conflict target.

Verified locally: go build ./... clean, worker + suggestions packages pass. CI green. LGTM.

[ccf-lisa]

PR approved. Marking as ready for e2e.