NBA predictions as JSON

Tonight's player predictions

/data/projections.json: slate_date (the US date), model_version and rows, one per player and market:

• player_id, player, team, opp
• stat: PTS, REB, AST, 3PM, STL, BLK, or a combined market PRA, PR, PA, RA
• proj: our prediction; floor and ceiling: the range the result lands inside 80% of the time
• minutes: predicted minutes
• line, price, book: the bookmaker line, its decimal price for our side, and the bookmaker
• side and edge: the side our probability favours and by how much (our probability minus the bookmaker's, margin removed)
• l10_over: how often he went over this line in his last 10 games
• mu, mu_rank: how much the opponent gives up to his position against the league, and its rank (1 = softest)
• prior, move_reason: the earlier prediction and why it moved, after team news

Tonight's game predictions

/data/games.json: games, one per game: start (UTC), home, away, pred_home, pred_away, total with total_lo–total_hi (80% range), p_home (home win chance), line_home (our spread for the home side, negative when favoured), and the bookmakers' book_line_home, book_total and book_p_home; out lists who's ruled out on each side.

Just the Tip

/data/tips.json: today (tonight's tip, or null), summary (the regular-season record: average CLV, beat-the-close rate, won–lost) and tips, every tip posted, with the price at posting, the closing price and CLV, and the result.

Track record

/data/track_record.json: calibration.markets, per market and window (7 days, 30 days, season): n, mae (average miss), bias (prediction minus result), coverage (share inside the 80% range), below and above.

Splits

/data/splits.json: one row per team per game this season: score, rest days for both sides, the closing spread (from that team's side) and total, and whether a top-three-minutes player sat.

Fields may be added; anything removed or renamed changes model_version first. Decimal odds throughout.