Data sources
What we use today (truth layer) and what we still need (market + availability) to compute betting edges.
Truth layer (implemented)
- NBA Stats endpoints via
nba_api: schedule discovery, box scores, and play-by-play (PlayByPlayV3). - Local SQLite canonical store: reproducible pulls, join-ready tables, run logs.
These sources are generally free to access, but they can rate-limit or change structure; our runbook includes pacing (--sleep) and resume support.
Market layer (needed)
To advise bets, we must capture sportsbook markets (lines + prices) at specific times (opening, pre-game, closing, and optionally in-play). Candidate options are compared on the Decision Point page.
- Game lines: ML / spread / totals
- Player props: points / assists / rebounds / threes / combos
- Historical odds snapshots + line movement
Availability layer (high leverage)
Injuries, starting lineups, role changes, and minutes expectations can swing props dramatically. We want a provider (or bundle) that can supply:
- Injury reports and status changes (questionable/out/etc.)
- Projected starters / minutes proxies (or at least lineup availability)
- Team context: pace, offensive/defensive ratings, rest/back-to-back flags
If we cannot source projected minutes directly, we can infer them from historical rotations once we have enough season data, but injury feeds still matter.
Past releases
Release 2025-12-16 (bundle v2) — expand
Data sources
What we use now (NBA truth data), what we need next (odds/props markets), and what context sources improve accuracy.
Sources we’re using today
NBA stats via nba_api: game lists, box scores, and (optionally) play-by-play via stats endpoints.
pbpstats (optional): alternate play-by-play normalization and caching when needed.
Sources we need next (to support FanDuel-style betting)
To cover “all markets you can bet in the app,” we need a market/odds feed that includes:
- Pre-game lines: spreads, totals, moneylines
- Player props: points/assists/rebounds/3PM + combinations
- In-game markets (optional later): live spreads/totals/props
- Historical archive (for backtesting) if available
Practical note: the best path is usually a paid, normalized odds provider with clear terms. If we must scrape, we’ll add a dedicated ingestion tool with rate limits, caching, and schema drift detection.
Context sources (improves projections)
- Injuries / availability
- Starting lineups / rotation signals
- Team pace + offensive/defensive efficiency
- Schedule density (back-to-backs, travel)
These are “adjusters” to the baseline minutes × per-minute model.