Receipts, including the ugly ones.

Week 7 · experiment-only

All fourteen current Week 7 atlas-research-v2 reports were graded against verified final scores and all seventy player outcomes.

The week contained both strong volume reads and meaningful misses, so the published error metrics should be treated as a small sample rather than evidence of calibration.

Availability uncertainty and low-total game structure deserve separate evaluation because they can affect winner confidence and player volume in different directions.

Experiments proposed

Test a separate availability-risk adjustment against held-out weeks instead of folding every injury gate into one narrative confidence score.

Compare position-specific yardage threshold bands on held-out weeks before changing the current house-threshold method.

Evaluate a low-total favorite shrinkage experiment without promoting it until the sample is large enough to compare Brier error and yardage MAE.

Week 6 · experiment-only

Week 6 is fully settled from official final scores and five verified player statistics for each of the 14 latest reports.

Four player outcomes were voided from documented inactive, injured-reserve, or wrong-game roster conditions; the remaining 66 outcomes were graded.

The A.J. Brown wrong-game roster forecast exposes an identity-validation failure that must be fixed before it can be treated as a forecasting error.

Brier error, yardage error, hit rates, and sample sizes are descriptive for one completed week and do not establish calibration or profitability.

Experiments proposed

Test a mandatory game-roster identity check before publication, including current team and game participation evidence for every selected player.

Evaluate whether offensive-line and quarterback availability gates should widen projected score bands and reduce winner probabilities on held-out completed weeks.

Compare role-weighted yardage thresholds against the current method on later completed weeks without changing active probabilities yet.

Week 5 · experiment-only

Week 5 is fully settled from official final scores and five verified player statistics for each of the 15 latest reports.

The slate included a tie and several large player-volume misses, so the review keeps score and role uncertainty visible instead of forcing a cleaner story.

Brier error, yardage error, hit rates, and sample sizes are descriptive for this completed week and do not establish calibration or profitability.

Experiments proposed

Test a larger uncertainty penalty for concentrated passing games when a primary receiver carries a late availability designation.

Evaluate whether quarterback-change and late offensive-line scenarios deserve wider score bands before changing published probabilities.

Hold every proposed adjustment out of the active method until it is compared against later completed weeks.