NBA 2026-27 Wins Model

Last refresh 2026-10-28T15:40

Input files

Config files (edit these, then rerun build)

Rating sources in use

{
 "darko": 517,
 "darko_feb2026": 2,
 "undrafted_prior": 1
}

Team sources

{
 "monster": 518,
 "espn": 2
}

Model constants

player rating = 50% DARKO DPM + 50% DARKO market rating (closing-spread implied, per 100; MARKET_WEIGHT=0.5); DPM alone where there is no market rating
DARKO ratings through 2026-10-26 · market as of 2026-10-27 · market covers 99% of projected minutes
minutes (per game, game_model.py): role MPG = mean of ETR original MPG and Monster ROS m/g; Monster INJ / OUT / D players play none of the games before their
  est. return (Q half, X none); once healthy, chance per game = Monster ROS games / team games left from the return; next game capped by P/Q/D (0.85/0.5/0.2)
each game: expected minutes over 240 come off the bottom of the rotation, a shortfall is spread over who is available (+20% of role, max 40), the rest filled at -2.2
team net that game = 5 × Σ(rating × minutes) / 240; Net = average over remaining games, Next = next game, Full = a typical game once everyone currently hurt is back
P(home win) = Φ((that game's net_home − net_away + 2.3) / 12.5); wins = actual + Σ remaining probs + unscheduled games (end-of-season roster) vs average
next slate: Monster's daily projected minutes + status (props board's 10-min cache of RJ's Monster daily export) replace the model for every player it lists
Monster ROS comes from a browser dump (RJ's login): python3 monster_ros.py <dump>, script in scripts/monster_ros_dump.js; /status flags it after 3 days
Rookies: college BPM + age at draft + ln(pick) model (config/rookie_model.csv), slot prior for internationals
Minutes (Per game): ETR games x ETR MPG; MPG from ETR FPPG where ETR gives games only; minutes ETR leaves open go to its unprojected depth-chart players
Refresh: python3 refresh.py (DARKO, ESPN rosters, Hashtag, schedule+results, ETR minutes rebuild) then build

ETR per-game stats: how they are recovered

Per-game stat = pool mean + z × pool sd. ETR's points-league FPPG (Underdog scoring) is exactly linear in its 9-cat z-scores (R² 0.99999, rmse 0.02 FP), so that regression gives each sd (3PM from DraftKings best-ball PPG). ETR doesn't reveal pool means: those are Monster's means on 297 matched players, scaled ×1.049 to hit ETR's FPPG intercept. FG% and FT% z-scores are volume-weighted in ETR's ranks, so they stay z-scores. Blank stats = ETR publishes no z-scores for that player (outside its 9-cat top 300). Team stats = Σ(per-game × games) / 82 over players with ETR stats. Off/Def = 5 × Σ(DARKO O/D-DPM × minute share); rookies on the college model carry only a total DPM.