project_news_speed_first_light

First data-bearing run of sports-scout-service/scripts/validate_news_speed.py (2026-07-28, events since 2026-07-09 when injury_status_history began logging). The 07-09 build ran on 0 events; this is the real first light.

Numbers: 503 status-change events → only 17 scorable prop-checks; 486/503 = "no odds coverage", which in this script means no pre-news line_before existed (or no odds at all for the player/game). Of the 11 "moved" cases: ALL latencies 25-30h, mostly threes, clustered on two games (1022600201/202) with injury-report churn — that's the books' daily posting/refresh cadence showing up, not reaction to the news event. Zero moves under 6h; zero observable "line lagging news" windows.

Why: WNBA prop lines post late (near game day), injury news breaks earlier. So the exploitable surface is NOT intraday movement after news — it's whether the FIRST POSTED line after news is stale. Current instrument cannot distinguish "slow reaction" from "posted later at the correct level".

How to apply: (1) Re-run monthly as events accrue — 3 weeks bought 17 scorable checks, so power grows slowly. (2) The decisive next instrument is opener-staleness: join status-change → first quote posted AFTER the news (odds-service has first_seen_at) → score that opener vs outcome/close. (3) Any Brier claim from news speed must be scored at bet-time prices (openers), not the close — the brier_board scores vs close by construction. Related: project_brier_vs_vegas_closed, project_wnba_backlog, project_decision_layer_audit.

SUPERSEDED 2026-08-06 — both premises measured FALSE. Lines are NOT scarce at news time (median first post 42.2h pre-tip; 93.3% >4h out; 85% of transitioning propped players already had a quote) — the "17/503 scorable" was a player-COVERAGE problem, only 98/355 transitions map to any prop. And props are NOT soft at release: blind over at the opener is −1.3% ROI (n=5,470) vs −3.4% at close, i.e. the opener is ~1.1pp better but still inside the vig — nothing like +6.21%. Late news buys nothing either: post-21:00Z scratch teammates 53.9% vs pre-21:00Z 53.6% (z=+0.12), vanishing entirely for ≥20-mpg scratches. See project_minutes_lever_measured. Do not build a T-60 run on this basis.