Methodology

Postgame Outlet publishes McCabe's human ratings, PGO's experimental model and a Fantasy preview. Each has its own inputs and purpose. Check the edition and saved dates before comparing them.

McCabe's ratings

Sean McCabe combines quarterback, non-quarterback offense and defense assessments. The component values add to his team rating. These are human judgments, kept separate from the PGO model.

The current PGO model

PGO's weekly model combines regular-season and playoff history. Inputs describe recent results, passing and rushing efficiency, sacks, turnovers and quarterback history. Some describe the same games, so they are influences within one formula, not separate proof of team strength. Earlier preseason and corrected boards remain available as dated archives.

After a full week has verified final scores and the required team and player statistics, PGO updates its inputs and publishes the next rankings. The fitted weights stay fixed. Missing or conflicting information keeps the last verified board visible with a delay notice.

How ratings become a score

A team's rating measures its model strength relative to the average of the 32 teams. Subtract the opponent's rating to get the projected neutral-site margin at equal rest. Add the venue and rest adjustments to get the game margin. Each saved matchup explains those steps.

Combined points currently use frozen 2025 regular-season and playoff scoring averages. Split that total in half, then move half the favorite's projected winning margin from the underdog to the favorite. For a 46-point total and a 4-point winning margin, that gives 25 points to the favorite and 21 to the underdog. The scores are averages, not promises of an exact NFL score. Nearly equal averages indicate a close game.

Winner picks, spreads and confidence

Straight-up W/L/T grades who won the game. The separate ATS record compares PGO's suggestion with a saved sportsbook handicap. We also show whether the straight-up winner pick covered that sportsbook line. Pushes, missing lines and no-edge selections are separate.

PGO's own projected line is shown alongside the sportsbook line. Exceeding PGO's projection is not the same as being accurate: margin error measures how close the forecast was to the result.

Expected pool points = fixed confidence points × selected team's estimated win probability. For example, 10 points at a 60% win chance gives 6 expected pool points; a win earns all 10 allocated points. These are not NFL scoreboard points. Late confidence entries remain labeled and do not count as pregame probability evidence.

Injuries and prediction locks

PGO checks official availability before the prediction cutoff, currently 60 minutes before kickoff. If the expected quarterback is confirmed unavailable, the conditional pick is withheld; a verified quarterback update may revise an unlocked forecast. Later inactive information is displayed separately and does not rewrite the locked pick.

Current non-QB injuries and replacement depth are context, not fitted point deductions. Prior playing time does not establish backup quality. Missing reports, rookies and unresolved identities remain unknown; an active roster listing does not prove health. Saved source dates show what was known and when.

Fantasy and evidence

The public Fantasy board is a Week 1 preview with scoring controls and dated injury annotations. New notes do not recalculate its saved projections or make the preview gradeable. Check the source date and official availability before using it.

PGO remains experimental. Accuracy is reported from saved forecasts and verified finals, with separate eligible counts for winners, margins, totals and probabilities. Penalty and alternative-weight tests have not demonstrated improvement under their fixed screens. Updating totals remains a separate experiment. More inputs or a passing software test do not establish a better model.

Open the ratings and records · Forecast Lab, methods and evidence · Corrections and accountability