Anytime TD โ Model Evaluation
Frozen eligible pregame snapshots ยท rate-based Poisson baseline ยท prospective collection
Insufficient sample (0 eligible predictions; minimum 20 for calibration). Do not interpret calibration or accuracy from this few picks โ prospective collection is in progress.
Predictions
0
Distinct Games
0
Model Version
labwork-nfl-anytime-td-v1
Date Range
โ
As-Of Cutoff
none
Model Evaluation
Avg Predicted Prob
โ
Actual Scoring Rate
โ
Brier Score
โ
Log Loss
โ
ฮต=0.0001
vs. Simple Historical Baseline
Static position league-prior anytime-TD rate (pre-specified, no training, no lookahead).
Baseline Brier
โ
Baseline Log Loss
โ
Model Brier ฮ
โ
negative = model better
Calibration โ Predicted vs Observed Scoring Rate
Green = observed scoring rate, steel = model's avg predicted probability. Well-calibrated bins have the two bars near equal. Sample size per bin shown below.
By Position
| Pos | N | Wins | Avg Pred | Actual Rate | Brier |
|---|---|---|---|---|---|
| No data | |||||
By Probability Range
| Range | N | Avg Pred | Actual Rate |
|---|---|---|---|
| 0-10% | 0 | โ | โ |
| 10-20% | 0 | โ | โ |
| 20-30% | 0 | โ | โ |
| 30-40% | 0 | โ | โ |
| 40-50% | 0 | โ | โ |
| 50-60% | 0 | โ | โ |
| 60-70% | 0 | โ | โ |
| 70-85% | 0 | โ | โ |
Betting Performance (separate from model evaluation)
Every frozen eligible pregame snapshot with a recorded pregame YES price; flat 1-unit stake on YES.
Bets
0
Units Net
0.00
ROI
โ
Avg Odds
โ
1 unit per bet, YES side only, recorded pregame price.
Excluded / Missing Records (never treated as losses)
Void
0
Needs Review
0
Pending (no stats yet)
90
Frozen, ungraded
0
- โขEvaluation uses frozen eligible pregame snapshots only; every refresh is not a separate prediction.
- โขPassing TDs excluded; rushing + receiving TDs only (matches projection scope).
- โขVOID/NEEDS_REVIEW/PENDING excluded from calibration and counted separately โ never treated as losses.
- โขReturn/defensive TDs are not in nflverse weekly stats; other_qualifying_tds is unverifiable and left 0.
- โขSample below 20: do not interpret calibration or accuracy from this few picks.