Model versions and technical record
nflfastR / nflfastR-6.0.0-fastrmodels-2.1.0-nonspread-regulationTraining window: Not established; retrospective evaluation is not claimed to be leakage-free.
Published training excludes ties/OT; pinned artifact training cutoff unverified.
nfl4th-adapted / under-review-nfl4th-2026-v1Training window: Not established; retrospective evaluation is not claimed to be leakage-free.
Rule-adapted experimental scenarios; recorded observed-action diagnostics do not identify unchosen counterfactual outcomes.
fumble / under-review-contextual-baselines-v1Training window: 2015,2016,2017,2018,2019,2020,2021,2022
No future games, recovery locations, eventual yards gained, or kick outcomes in predictors.
fg / under-review-contextual-baselines-v1Training window: 2015,2016,2017,2018,2019,2020,2021,2022
No future games, recovery locations, eventual yards gained, or kick outcomes in predictors.
xp / under-review-contextual-baselines-v1Training window: 2015,2016,2017,2018,2019,2020,2021,2022
No future games, recovery locations, eventual yards gained, or kick outcomes in predictors.
penalty / under-review-contextual-baselines-v1Training window: 2015,2016,2017,2018,2019,2020,2021,2022
No future games, recovery locations, eventual yards gained, or kick outcomes in predictors.
overtime-empirical / under-review-overtime-empirical-v1Training window: 2017–2025; comparisons use strictly prior seasons
Experimental empirical overtime estimates for supported possession phases; not calibrated coaching or officiating costs. Points describe pre-play win/loss/tie probabilities, except END GAME is an observed outcome. Sparse, unsupported and ambiguous states remain unavailable.
game-profile-audit / under-review-game-audit-v5Training window: 1999–2025; target comparisons use prior seasons only
Descriptive fixed-pattern historical comparisons and play review triggers; no intent or misconduct inference.
spread-reference / under-review-spread-v1Training window: 1999–2025; strictly prior target seasons
Descriptive absolute margin error versus recorded closing spread; no misconduct probability; market-only rating capped at 3/5.
game-expectations / under-review-expectations-v1Training window: Five completed seasons before the game; calibration uses three earlier seasons, each predicted from its own prior five seasons.
Team/opponent/referee penalty expectations and retrospective yardage/turnover outcome expectation. Separate family tails; referee/team win records are context only.
Analysis input hash: 8d091317a38fe5c15c9aef6d65e43751068c25988b13641f3adeea3907f15bef