Alabama takes over No. 1 after Week 10

What the power order says beyond the record

LSU is No. 6 at 6-3, ahead of No. 9 UCLA at 9-0, despite a lower winning percentage. Their power ratings are +19.0 and +18.1. This is a comparison of estimated scoring strength, not a claim that the higher-ranked team has accomplished more.

An opponent-adjusted power rating can value a competitive loss differently from a heavy loss, while both add one loss to the record. To evaluate accomplishment, use wins above benchmark. This report does not insert the current season's résumé ranks into an earlier week's history; current opponent evaluations would bring later information into that comparison.

• LSU results
• UCLA results

Movement, sample size, and close comparisons

Oklahoma State moved from No. 30 to No. 17, while its rating changed by +2.3 points. Places and points measure different things: other teams' changes can move a rank even when a team's own rating barely changes. Schedule-wide re-estimation means movement cannot be attributed to one game alone.

Alabama leads Georgia by 0.44 model points. 0 of the 138 ranked FBS teams have two or fewer completed games in this snapshot. Close rating gaps and thin samples call for caution; the model does not provide a rank confidence interval, and the precision of the display is not evidence of certainty.

• Schedule connectivity and cross-division limits

Where the projections missed

Across 315 non-tied games with usable projections, the mean absolute margin error was 12.9 points. This measures how far the predicted score difference was from the final difference; picking the winner correctly can still leave a large margin error.

The largest listed FBS-involving margin miss was Tulsa at Tulane: projected designated-home margin +0.9, actual designated-home margin −50.0, an absolute error of 50.9 points. These are scoring differences, not predicted or final scorelines. We do not infer an injury, coaching decision, or other cause from that error alone.

A winner given 35% or less is labeled an upset by this site's convention. That threshold is descriptive, not a claim that such wins should never happen. A single surprising result cannot establish probability calibration; that requires many comparable, honestly evaluated forecasts.

• Tulsa game log
• Tulane game log

What this report can verify

Of the 315 non-tied projections in the aggregate above, 10 have eligible recorded pregame publication evidence, 305 are eligible walk-forward reconstructions, and 0 lack the metadata needed for either verified cohort. The published and reconstructed records are kept separate; the aggregate is provided only to describe the available legacy report inputs.

Walk-forward calculations use earlier-period information for ratings and probability calibration. They reconstruct what the model would have said; they are not evidence that those odds were on the website before kickoff. Only a validated successful-publication record can support that claim. Legacy projections with missing provenance are unverified.

Ranks and records come from the solve through this reporting period. Later source corrections or model changes can revise a regenerated archive. Linked team profiles show the current publication, and the next-period matchups are available projections rather than an immutable copy of an earlier preview. This automated report has no personal byline or human sign-off unless an explicit review is recorded.

• Publishing and correction standards
• Model methodology