Unified-model historical results

Historical model results at averaged bookmaker odds, separate from original recorded picks.

PredictionsPredictionsMonitoringMonitoringPlayer propsPropsFantasyFantasyBenchmarksBenchmarks

From 15 August 2026 · all retained odds · exact lines and Over/Under evaluated separately. Each outcome uses the average of available bookmaker odds, including single-bookmaker offers.

Reconstructed backtestAvailable matches: 15 Aug 2026 – 18 Oct 2026. These are rebuilt forecasts, separate from original recorded picks.Retained prices include quotes with unknown or post-match capture times. Choose recorded pre-match quotes for the stricter quote cohort.
Reset
All-time unified results
3,007Distinct matches
584,208Observed outcomes
508,650 / 75,558Priced / missing
123,857EV bets
−7470.40uEV profit · −6.0% ROI
0.1796Brier score · 351,657 binary outcomes

EV bets have positive estimated return at the averaged odds. Every observed outcome stays in the inventory; missing model prices never enter model returns. All priced returns also include zero and negative EV. No odds range or profitable-cell filter is applied.

‹

October 2026

›

Month: 57,083 EV bets · −2906.75u · −5.1% ROI

Historical reconstruction · one unit per exact-line outcome with positive estimated EV. Days use UTC−6, like the original calendar.

MonTueWedThuFriSatSun1−7.1%1388 bets2+6.8%1881 bets3+11.7%1678 bets4−8.6%1957 bets5−13.0%1627 bets6−9.3%1967 bets7−15.2%337 bets8−14.2%407 bets9−14.1%1978 bets10−11.0%8323 bets11−3.4%5831 bets12+2.6%1909 bets13−13.8%2310 bets14+10.8%3074 bets15−14.1%5929 bets16−9.5%2343 bets17−9.0%8754 bets18+13.3%5390 bets19No replay yet20No replay yet21No replay yet

25 Oct 2026

This date is beyond the frozen replay coverage. Original picks remain available in their separate history.

All observed outcomesPositive EV onlyCell benchmarks →
Match / cellAverage oddsModel probabilityEstimated EVActual / return
Backtest method and coverage

Each exact line and direction is one virtual unit. Bookmaker offers are averaged, not counted as independent bets. Integer pushes and quarter-line half wins/losses are settled proportionally. W/L count positive/negative returns; P counts zero returns.

The model logic was selected using earlier results only. The report is historical research, not proof of profitability or a replay of every current live adjustment. Brier scores use half-point binary outcomes only. Overlapping lines and directions are correlated.

  • Forecasts were reconstructed, not saved before the historical matches.
  • Retained prices may have missing or post-match capture times; a separate recorded pre-match quote cohort is available.
  • Result availability in fitting uses kickoff plus three hours as a proxy.
  • This retained replay does not reproduce every live lineup, tempo or division transformation.
  • Unavailable model prices and unverified card units remain explicitly missing.

Historical model: unified-20261021-v5 · Snapshot: history-20261021-v4 · Input cutoff: 2026-10-21T12:56:16+00:00

Drag to move this button. Its position is remembered.