WINFORGE
ACCURACY LEDGER

Before live 2026 picks are graded here, WinForge reports results from an independent 2025 test (trained on 2022–2024) using games kept out of training. The model was retrained 2026-08-14. Results: WR receiving yards model-fit score 0.59, QB passing yards model-fit score 0.45, RB carries model-fit score 0.61, WR targets model-fit score 0.80. A model-fit score shows how closely projections tracked real differences in results; it is not a hit rate or percent accuracy. See the methodology page for how the model is evaluated. Updates weekly once the 2026 season begins.

Graded picks: 0. Hit rate: —. WinForge will not publish a back-filled, simulated, or cherry-picked record as live performance.

Each row shows player, market, sportsbook or pick'em line when available, model projection, pick direction, edge, locked timestamp, lock hash, and once graded: official actual result plus HIT/MISS/PUSH. The row structure is designed so a future reader can audit what was known before kickoff.

A projection only enters the ledger if it is committed before the game starts. That protects the record from post-game editing and clearly separates a locked projection record from retrospective analysis.

Rows are graded after official stat results are available. Pushes are separated from wins and losses, and the independent test on games kept out of training is kept separate from future live hit-rate reporting.

The ledger is the public record of projections WinForge stands behind. A projection only counts if it was committed before kickoff and graded after official results.

The ledger is a transparency tool, not betting advice. Past statistical performance does not guarantee future outcomes or profit, and WinForge is not a sportsbook.