How we build independent win probability models

Why independent probabilities matter

Most sports analytics services derive "fair value" by devigging sportsbook lines — averaging Pinnacle, FanDuel, and DraftKings odds. This gives you a consensus probability that tracks the market by construction. It can be useful for line shopping, but it is not an independent model view — because the output already agrees with the market.

ZenHodl models are trained on game state and pregame priors: score differential, seconds remaining, period, Elo ratings, sport-specific features, and (in most sports) a pregame win-probability prior derived from ESPN's pregame number. ESPN's live in-game win probability is deliberately excluded, and no odds, lines, or market prices are used as inputs. This keeps our output independent from the market price you compare it against — useful for line shopping and auditing, but still subject to fees, slippage, and model error.

The tradeoff: our models trade some absolute Brier score for genuine independence — but the per-sport values reported below (NBA Brier 0.128, held-out ECE 0.053; NCAAMB Brier 0.146; NHL Brier 0.184) show the gap is competitive in practice. Independence is useful because it gives you a separate fair-value reference to compare with market prices, not a guaranteed edge. See /clv-evidence for the audit and retraction of our former outcome-conditioned CLV claim, /validation for the exported backtest snapshot, and /results for current live production performance.

Data pipeline

We scrape ESPN's play-by-play API across the core sports we model. Each game produces hundreds of snapshots — one per score change or significant event.

Sports: NBA, NCAAMB, NCAAWB, CFB, NFL, NHL, MLB

Data is stored as Apache Parquet files. One row = one game state (score, period, clock, ESPN WP, outcome label).

Module 1 of our course teaches you to build this exact scraper.

Feature engineering

Each game state is featurized with 13–16 variables. We deliberately keep the feature set small — overfitting to noise destroys trading value.

Model architecture

We use sport-specific models — no one-size-fits-all approach. The sections below distinguish the current live platform from adjacent research and course material.

Current live platform — supported sports

The models cover NBA, WNBA, NCAAMB, NCAAWB, CFB, NFL, NHL, MLB with sport-specific pipelines, post-hoc isotonic calibration (held-out ECE 1.4%–9.2% by sport), and a four-layer prediction pipeline: base model, team/player overlays, isotonic calibration, and live rolling recalibration.

Basketball & Football (NBA, NCAAMB, NCAAWB, CFB, NFL)

Split-Phase XGBoost with 16 features including team offensive/defensive ratings (ORtg, DRtg), pace, momentum (scoring runs over last 2 and 5 minutes), Elo ratings, and interaction terms (score_diff × time_fraction). NBA Brier: 0.128, held-out ECE: 0.053. Trained on 5,285 games (2021–2026) with walk-forward season-based splits. Post-hoc isotonic recalibration on the 2024–25 season.

Injury overlay: 58 NBA star players tracked in real-time via ESPN's injury API (10-minute cache). Each player has a pre-computed impact factor (e.g., Jokic: 10%, LeBron: 8%, Curry: 9%). When a player is OUT, the model subtracts their impact; QUESTIONABLE halves the adjustment. Total cap: ±15%.

Model selection uses a hybrid criterion: near-best Brier score AND best trading value (c/trade) on a held-out backtest window. This prevents selecting models that look good on calibration metrics but produce worse P&L.

Hockey (NHL)

XGBoost with 17 features including all basketball features plus hockey-specific metrics: power play %, penalty kill %, save %, faceoff win %, and penalty minutes differential. NHL Brier: 0.184 (improved ~10% from the 0.205 pre-calibration baseline), held-out ECE: 0.028. Trained on 4,225 games.

Injury overlay: 44 NHL skaters tracked via ESPN (goalies excluded — handled by the separate goalie quality adjustment). Star impact factors: McDavid 10%, Draisaitl 8%, MacKinnon 9%, etc.

Goalie & shot overlay: Starting goalie detection adjusts for goalie quality. Expected goals (xG), Corsi, and power play state provide real-time shot quality signals. Combined cap: ±12%.

Baseball (MLB)

Ensemble model with starting pitcher ERA, WHIP, K/9 as base features. MLB Brier: 0.152, held-out ECE: 0.014. Three post-model overlays stack on top of the base prediction:

• Run expectancy: Tango run-expectancy matrix (outs × runners × batting side) provides granular in-game state beyond the score.
• Park factors: 30-team FanGraphs 3-year rolling park factors (Coors Field = 1.25, highest).
• Bullpen quality: 30-team ERA, time-scaled from 0% (early innings) to 100% (late innings) as starters hand off to relievers. Capped at ±5%.

Soccer (Bundesliga, EPL, La Liga, Serie A, Ligue 1)

Live soccer trading is currently frozen (force-shadowed). The XGBoost model trained on 2,949 matches had silently vanished from the server in April 2026, degrading the live bot to a Poisson fallback that measures worse than its own entry prices. A retrained challenger failed its promotion gates, so the freeze stands, and a daily artifact-presence monitor pages if any model file disappears again. Pregame research continues on the Poisson + XGBoost stack.

Esports (CS2, LoL)

Both esports run models on the platform; live order flow varies with each sport's current risk status. CS2 has been sizing-paused on realized CLV; LoL was reverted to its v1 analytical model in August 2026 after the v2 XGBoost override lost two independent head-to-head evaluations (see below).

CS2: 15 features including round differential, Elo, map-specific win rates (per team per map), recent form, head-to-head record, and Elo momentum. Economy data from bo3.gg and HLTV scorebot provides equipment value and consecutive-loss tracking as post-model overlays. Entry restricted to underdogs (20–42c) with 20c+ minimum edge after a live P&L audit found the model overperforms on underdog lines and underperforms on favorites.

LoL: v1 analytical model — Elo ratings plus a binomial series model for BO1/BO3/BO5. The v2 XGBoost override (20 features: gold differential, kills, towers, dragons, barons, inhibitor advantage, and more) was reverted in August 2026 after two independent evaluations on 110 fire-anchored matches both favored v1 (Brier 0.1692 vs 0.2077). The revert stands; v2 remains a shadow challenger.

Tennis (ATP, WTA)

Hierarchical analytical model: player-specific serve win rates (by surface) → point probability → game probability → set probability → match probability. Serve rates are computed from 13,174 ATP matches (2020–2024) using the Sackmann dataset, giving us player-specific rates for hard, clay, and grass surfaces instead of tour averages. Elo ratings are surface-aware (4,606 players). The live recalibrator auto-activates after 50 resolved trades.

Post-model overlay stack (all sports)

Every prediction passes through two correction layers after the base model, and the system wraps them in three measurement-and-risk controls:

1. Post-hoc isotonic calibration: Fitted on held-out calibration season. On the calibration season it is fitted to, the residual ECE is near zero — but the honest out-of-sample figure, measured on held-out test seasons the calibrator never saw, is 1.4%–9.2% by sport (best MLB 0.014, worst NFL 0.092). When a well-calibrated model says 70%, teams win about 70% of the time.
2. Live rolling recalibration: Automatically refits an isotonic correction every 25 resolved trades per sport, using a 500-sample rolling buffer persisted to disk. Catches calibration drift from rule changes, roster turnover, or seasonal patterns without manual intervention.
3. Closing-line value (CLV) audit: Admitted live fills are measured against our captured closing price where one exists; the scorecard publishes the current measured count and coverage rather than freezing an old percentage into this methodology page. The winner/loser split we used to cite here was retracted 2026-07-21 as a measurement artifact; the honest unconditioned aggregate (share of fills beating the close and mean CLV per trade) is published per sport at /clv. The four-failure-mode taxonomy (bad model / bad blend / bad timing / bad CLV label) is published at /clv/repair alongside the explicit reactivation criteria for any sport currently paused.
4. Dynamic position sizing: Rolling 50-trade performance per sport determines a size multiplier (0.25x–1.5x). Hot sports (>55% WR + positive P&L) get sized up; cold sports (<40% WR) get sized down. Recomputed every 10 minutes.
5. Meta-model edge filter (research, shadow mode only): A Gradient Boosting classifier trained on resolved trades was proposed to separate real edges from noise. It has never been trusted to block live trades — it runs in shadow mode as a research instrument. The "filters 28% of losing trades" figure we once cited was not reproducible and has been removed.

Spread/Total

Spread and total models require a different setup from moneyline markets: regression on remaining margin or total, then a distributional layer to convert that forecast into cover/over probabilities.

Backtesting methodology

Our backtests are designed to avoid the common mistakes that inflate results.

• Time-split validation: Train on season N−1, test on season N. No future data leaks.
• Real market prices where available: Where direct market snapshot history exists, backtests use recorded market prices. Where coverage is incomplete, we call out any proxy assumptions explicitly on the validation page.
• Deduplication: One entry per (game_id, side, score, period). No duplicate signals from the same game state.
• Subsampling where needed: Historical replay backtests can downsample state rows to better match how the live system would realistically observe and act on the game.
• Hold to settlement: All trades held to game end. No exit timing or stop-loss assumptions.
• Net of fees: Results are reported net of the fee and slippage assumptions used by the active backtest export, with the exact assumptions disclosed on the validation page.
• Sport-specific entry filters: Filters like score differential, game state, or pregame-consensus agreement are applied only where validation shows they improve robustness.
• Execution windows: Entry timing constraints vary by sport and market. We only keep them when out-of-sample testing shows they improve execution quality.

What we tried that doesn't work

We believe in showing failures alongside successes.

• Sell-the-top / fade spikes: Negative EV. Information-driven price moves don't mean-revert.
• Compression sniping: Negative on historical data, adverse selection live.
• Martingale / double-down on dips: Catastrophic — doubles position into adverse moves.
• Early NBA taker configs: Earlier versions underperformed until pacing/context features and stricter entry filters were added. Old model versions are not treated as current proof.
• Spread/Total mean-reversion exits: Spreads reprice permanently from score changes. Fix: hold to settlement instead.
• ESPN live WP as model input: Tested and rejected — creates circular dependency with market. ESPN pre-game WP works as a fixed prior.
• Conformal prediction intervals: Tested and rejected — added complexity without improving trading value.