feedback_active_tonight_gate — Findings

← All findingssource · memory/feedback_active_tonight_gate.md

STATUS: SHIPPED (REVISED). FIX 5 went through two iterations:

v1 (2026-05-21, commit dbfe7e9): joined snap.Injuries only. Set ActiveStatus for designated players (Out/D/Q/P/GTD), left healthy starters with the field absent (omitempty dropped it).
v2 (2026-06-09, commit bf97c5f on main): 4-step resolution: seed-unknown → lineup STARTING/ROTATION → live-boxscore Minutes>0 → injury overlay. omitempty removed. Field always present with one of 7 enumerated values: active|unknown|out|doubtful|questionable|probable|gtd.

Why the v1→v2 revision was needed: NYK@SAS Finals G3 (2026-06-08). Brunson/Hart/Castle/Champagnie were all STARTING in snap.Lineups and on the floor in snap.LiveBoxscore, but their active_status was field- absent because injury-only join had no row to write. sport-agents debate discarded all 4 with "insufficient prop_diagnostic data"; MCP enrich_pick returned [unflagged]. The wire couldn't distinguish "confirmed active" from "we don't know" — both serialized as field-absent. v2 closes that gap.

Hard-gate semantics preserved: Out/Doubtful/Questionable still overlay last in step 4 and still flip LineEdges.Quality="inactive_player" via the existing isHardGateStatus check in attachDiagnostic.

Known WNBA limitation (separate fix): WNBA snap.Lineups returns LineupType=PROJECTED from roster-fallback (lineup-service can't get real WNBA confirmed lineups), so step 2 doesn't promote anyone and every WNBA player stays "unknown" pre-tip. Fix lives in lineup-service WNBA ingestion, not data-hydrator.

On 2026-05-21 I delivered Fox OVER 14.5 points as a "lockdown-clean" pick. Fox was confirmed OUT (not Questionable — fully inactive) at tip. The snapshot DID flag him as Questionable in injuries.away_injuries 24h earlier — I just never read the injury section, only the 4-fix flags object. User caught it: "De'Aaron fox is not playing, how come u don't have up to date data?"

Why: the 4 methodology fixes (role-anchor, line-vs-mean, MIN-SRC, recent-vs-baseline) all live under prop_diagnostic.stats[stat].flags — a structured per-stat sub-object. The injury status lives under snapshot.injuries.{home,away}_injuries — totally separate path. The 4 flags can be "clean" while the player has Questionable/Doubtful/Out status. There is no FIX 5 = active-tonight gate.

How to apply: BEFORE recommending any prop:

Open snapshot.injuries.{home,away}_injuries and check the player's status.
Any status of Out / Doubtful / Questionable is a hard drop — do not surface as a pick, even if the 4 lockdown flags are clean. (Questionable converts to Out the majority of the time on game-day in playoffs.)
Cross-check with a live external source (RotoWire nba-lineups.php, official inactive list) — data-hydrator's snapshot.injuries is itself subject to the cache staleness in feedback_cache_ttl_bug.
Long-term: ship lockdown FIX 5 = active-tonight hard gate, surfaced as flags.active_status = "out" | "questionable" | "doubtful" | "probable" | "active" computed from the joined injuries section, with PASS-quality enforced when non-active. Then the existing quality field carries it through automatically.