vyndr/specs/propline-audit.md

PropLine + Box Score + Pipeline Audit — 2026-07-10 (Session 56)

Ground-truth inventory of what data we can access, what our pipeline handles, and the gaps. Live sections were fetched against real APIs (MLB Stats, ESPN, The Odds API /v4/sports). PropLine has no dev keys, so its inventory is derived from the authoritative code (proplineAdapter.MARKETS × oddsNormalizer.MARKET_MAP) — that IS the set our pipeline requests + normalizes. Raw capture: specs/audit-data/propline-audit-raw.json.

Baseline: 2276 tests passing at session start.

---

0. What is in-season RIGHT NOW (The Odds API /v4/sports, live 2026-07-10)

Sport	Active key(s)	Notes
MLB	baseball_mlb	✅ full season — the live product
WNBA	basketball_wnba	✅ in-season (3 games tonight per ESPN)
NBA	basketball_nba_summer_league only	⚠️ regular NBA off until Oct; summer league live
NFL	americanfootball_nfl, _nfl_preseason	⚠️ preseason lines starting
Soccer	40+ leagues (soccer_epl, soccer_usa_mls, soccer_fifa_world_cup, La Liga, Serie A, …)	✅ year-round via odds-api
NHL	(absent)	off-season — no lines

Takeaway: MLB + WNBA + soccer are the live opportunity today. NBA/NFL are pre-season shells; NHL is dark.

---

1. PropLine / odds inventory — what we REQUEST vs what we CAN normalize

proplineAdapter.MARKETS = the markets we actually ask for. MARKET_MAP = every market we can normalize (superset). A market we request but don't map → silent zero; a market we map but don't request → unused capacity (free to add — same call).

MLB

Requested market	→ stat_type	Grade whitelist?	Features (l5/l20)?	Outcome settle?
batter_hits	hits	✅	✅ hits	✅ hits
batter_home_runs	home_runs	✅	✅ homeRuns	✅ homeRuns
batter_total_bases	total_bases	✅	✅ totalBases	✅ totalBases
batter_rbis	rbis ❌	✅ (rbi)	✅ (rbi)	✅ (rbi)
batter_stolen_bases	stolen_bases	✅	✅ stolenBases	✅ stolenBases
pitcher_strikeouts	strikeouts	✅	✅ strikeOuts	✅ strikeOuts

Mapped but NOT requested (free capacity — all supported downstream already): batter_runs→runs, batter_walks→walks, pitcher_earned_runs→earned_runs, pitcher_hits_allowed→hits_allowed. Mapped, not requested, and needs feature/ settle wiring: batter_doubles→doubles, pitcher_outs→outs. Confirmed present in the real boxscore (§2).

WNBA / NBA

Requested: player_points, player_rebounds, player_assists, player_threes (+ NBA adds player_blocks, player_steals). All grade-whitelisted. Features: NBA/WNBA game logs come from the Python nba_api service (usually OFFLINE in prod) → l5/l20 often empty; ESPN fallback exists for season avgs only. Outcomes: NOT settled at all (outcomeService is MLB-only). player_turnovers/player_pra are mapped but not requested.

Soccer

PropLine has NO soccer entry (SPORT_KEYS/MARKETS omit it) — soccer props can only come from the odds-api backup. MARKET_MAP maps goals/shots/shots_on_ target/tackles/cards/corners/saves/passes/clean_sheet. Features: soccer extractor exists; outcomes: NOT settled.

NFL / NHL

NFL markets mapped + requested (pass/rush/rec yds, receptions, TDs) but the flow isn't graded end-to-end yet. NHL mapped/requested but off-season.

---

2. MLB box score field map (REAL — Braves @ Pirates, 2026-07-09)

Batting fields present: runs, doubles, triples, homeRuns, strikeOuts, baseOnBalls, hits, atBats, stolenBases, totalBases, rbi, sacFlies, … Pitching fields present: strikeOuts, baseOnBalls, hits, inningsPitched, earnedRuns, outs, battersFaced, …

OUTCOME_MAP (MLB) — stat_type → game-log/boxscore field:

hits           → hits            total_bases → totalBases    home_runs → homeRuns
rbi            → rbi             runs        → runs           walks     → baseOnBalls
stolen_bases   → stolenBases     strikeouts  → strikeOuts     earned_runs → earnedRuns
hits_allowed   → hits (pitching) innings_pitched → inningsPitched
doubles        → doubles  ⟵ NEW  triples → triples ⟵ NEW      outs → outs ⟵ NEW
batter_strikeouts → strikeOuts ⟵ (batter Ks — roadmap)

Every field above is real in the boxscore/game log → all are settleable today.

2b. ESPN box score (WNBA/soccer — the non-MLB settle path)

WNBA settlement source = ESPN summary boxscore.players[].statistics[] (labels/names arrays: PTS/REB/AST/3PM/STL/BLK/TO). No completed game at fetch time (all 3 tonight were Scheduled), so labels weren't captured — wiring the ESPN settle path is a roadmap item (espnStatsAdapter already parses this shape for season avgs).

---

3. Pipeline flow + gaps

Current flow: cron (SNAPSHOT_CRON=1, UTC 14,19,22,1,3) → per sport: settleAllOutcomes() (S55, MLB only) → snapshotService.runSnapshot → PropLine (3-key rotation) / odds-api backup → gradeAndCacheSlate → per-player archetype + features → Redis (snapshot:{sport}:latest, grades:{sport}) → ticker.

Gaps found:

1. RBI silent failure — batter_rbis → rbis mismatch (§1). FIX this session.
2. Under-requesting MLB markets — only 6 of ~12 supported; runs/walks/doubles/ earned_runs/hits_allowed/outs are free in the same call. EXPAND this session.
3. Outcome settlement is MLB-only — WNBA/NBA/soccer grades NEVER settle, so accuracy is MLB-only. Biggest roadmap item (needs the ESPN settle path).
4. No pipeline alerting — a failed/empty/missed cron run is silent. ADD ntfy this session.
5. No retry — one PropLine blip marks a sport unavailable for the whole cycle. ADD retry-once this session.
6. No missed-cron detector — nothing notices if a slot didn't fire. ADD.
7. NBA/WNBA features depend on an offline Python service — l5/l20 often empty; ESPN gives season avgs only. Roadmap: an ESPN game-log feature path.
8. Soccer has no PropLine source — odds-api backup only; document + monitor.

---

4. Sport coverage status (verified 2026-07-10)

Sport	Odds props	Grade pipeline	Features	Outcomes	Status
MLB	✅ PropLine + odds-api	✅	✅ real (statsapi)	✅ (S55)	LIVE
WNBA	✅ odds-api/PropLine	✅	⚠️ Python offline → thin	❌ not settled	PARTIAL
NBA	⚠️ summer league only	✅	⚠️ offline	❌	OFF-SEASON
Soccer	✅ odds-api (no PropLine)	⚠️ extractor exists	⚠️	❌	PARTIAL
NFL	⚠️ preseason	⚠️ mapped, not e2e	❌	❌	PRE-SEASON
NHL	❌ off-season	⚠️ mapped	❌	❌	DARK

---

5. Fixes shipped this session (Phase 4/6)

• batter_rbis → rbi (unblocks a market we already request).
• PropLine MLB MARKETS expanded: +batter_runs, +batter_walks, +batter_doubles, +pitcher_earned_runs, +pitcher_hits_allowed, +pitcher_outs.
• doubles + outs wired into featureCache + outcomeService + all three grade whitelists (analyze.js / scan.js / validation.py).
• Pipeline alerting (ntfy) + PropLine retry-once + missed-cron detector.

See specs/vyndr-roadmap.md for the priority-ordered plan that follows from this.