The policy in brief
What this part would establish
- Preseason polls have no formal role in selection, schedule strength, tiebreakers, or quality-win labels.
- Opponent strength is recalculated throughout the season using the team each opponent became; quality-win credit is awarded only for opponents actually defeated.
- A National Résumé Rating measures accomplishment and completed-game performance; a separate Team Strength Rating estimates future quality and availability.
- The public NRR begins with the first official CFP ranking and updates at every committee release; each checkpoint uses only games completed through the preceding Saturday.
- A committee placement more than three positions outside the objective résumé range requires a written explanation.
Test the policy
Open National Résumé Rating v0.48
Inspect every component, the schedule-calibrated record share, national schedule-validation curve, cumulative Top-25 opponent tiers, exact nonlinear win evidence, additive loss attribution, opponent résumé cards, game line scores, explicit conference-championship labels, consistent title-game protection, and the narrow peer head-to-head tiebreak; audit a model with no team, conference, score floor, or manual override; and compare twelve weekly CFP checkpoints across 2024 and 2025.
How it works
The National Résumé Rating
Version 0.48 gives quality-win depth 30%, schedule-adjusted record 30%, opponent-adjusted performance 15%, loss integrity 20%, and win control 5%. Conference names, brands, recruiting, and preseason expectations never enter the score. Final Opponent Value blends 60% official CFP checkpoint standing, 20% schedule-connected performance, 10% schedule caliber, and 10% schedule-adjusted record. The official standing curve distinguishes truly elite opponents from the back of the Top 25; unranked teams receive a formula-derived baseline capped below the No. 25 floor. This removes the prior circular risk of using one team’s NRR as the main evidence for another team’s NRR. The performance input moves two points for each point schedule caliber sits above or below 50, limiting soft closed-slate inflation without a conference label. Each win’s directly comparable Game Value blends 75% opponent-and-venue value with 25% sustained performance using final margin and Q4, 6:00, and 2:00 checkpoints. Opponent quality then enters one continuous, unbounded diminishing-return evidence curve and game control supplies only a bounded multiplier. Every stronger opponent still adds more evidence; there is no hard evidence cap or tier cutoff. Evidence units from every distinct defeated opponent are added before one smooth curve converts the total to the 0–100 Quality-Win Depth component. There is no first-, fourth-, or fifth-win coefficient: equivalent wins receive equivalent evidence. The nonlinear opponent curve allows one truly elite victory to outweigh two fringe Top-25 wins, while four Top-20 wins still receive four complete contributions. Road context is worth up to six points and neutral context up to three, scaled continuously by opponent quality. Every opponent is ranked nationally at the exact weekly checkpoint; cumulative Top 5, Top 10, Top 15, Top 20, and Top 25 win counts appear on the résumé card, lesser support is disclosed separately, and every evidence card can open the opponent’s horizontal résumé and the archived quarter-by-quarter game line. Schedule-adjusted record gives 40% to achieved record, 30% to wins above a fixed benchmark expectation, and 30% to schedule-challenge caliber. Ranked opponents draw 90% of their schedule value from the selected official checkpoint and 10% from national play-level strength; unranked opponents retain play-level strength. The schedule term blends 30% of the full slate with 70% of one smooth challenge-weighted average, then uses one continuous 12-point logistic contrast curve so elite slates separate from ordinary ones without a tier line. Every opponent counts, with no tier cutoff or counted-game cap. Scored opponent-adjusted performance begins with the national ridge estimate and subtracts one continuous, deduction-only soft-schedule credibility term. The same shortfall lightly scales Win Control, keeping an unbeaten elite résumé near the top of the scale without granting soft schedules a free pass. Opponent-adjusted performance fits every offense and defense together from meaningful scrimmage-play EPA across the national schedule. Loss integrity stays on an independent first-pass opponent graph; its continuous 1.2-rate exponential decay applies to the 0.45-power weakest-loss shortfall below Opponent Value 60, while no loss earns quality credit. A smooth two-point lower-tail calibration preserves separation among damaged résumés without imposing a score floor: zero remains zero and 100 remains 100. Quality-Win Depth receives a smaller 1.5-point version of the same endpoint-anchored calibration. The multiplicative additional-loss frequency rate is 25%. Every conference-title loss receives one-third ordinary game severity and half ordinary frequency treatment; repeat-opponent penalties remain, so a rematch still carries distinct résumé cost while reaching the title game is not punished like an ordinary extra game. When two teams have the same record and sit within one point in the frozen provisional résumé Index, a direct win earns a small 0.35 raw-composite peer tiebreak. That rule is capped, formula-defined, and visible—it is not a committee reorder. The five weighted measures are first multiplied by one universal record foundation—40% floor plus achieved winning percentage curved to 1.15, followed by a continuous soft-schedule deduction that rises with loss rate to the 1.5 power—making the actual record the majority input to résumé credibility. Scored performance keeps the full slate as its schedule baseline and uses one continuous gate to validate it with the two strongest national schedule connections. The same formula applies to every team, with no school, conference, or manual adjustment. The published NRR Index then remains exactly 20 + 0.85 × raw composite, and the unrounded Index determines rank. There is no school override, conference multiplier, poll reconciliation, or manual adjustment. Readers may test alternative weights privately without changing the proposal default. The public NRR begins with the first official CFP ranking and is recalculated at every committee release using only games completed through the preceding Saturday.
A distinct strength estimate
The Team Strength Rating asks how good a team is likely to be next and may weigh current availability, recent performance, and validated predictive efficiency. NRR’s opponent-adjusted performance is different: it describes completed plays only. An injury does not erase completed wins in the résumé rating, but it can affect the separate strength estimate. The committee must explain how it balanced the two.
Venue-adjusted opponent quality
The same opponent should not carry the same difficulty label everywhere. The proposal’s initial Q1 bands are NRR 1–15 at home, 1–20 on a neutral field, and 1–25 on the road. The public bands aid communication while the underlying model stays continuous.
Publish the dashboard and the disagreement
Every team page would show CFP rank, résumé rating, strength rating, schedule strength, strength of record, quality results, road performance, best wins, losses, and conference results. The public archive preserves every release, movement from the prior week, the same-date CFP comparison, model code, data definitions, recusal rules, and an annual validation report.
Safeguards
What keeps the rule honest
- The committee retains judgment rather than outsourcing selection to a formula.
- Scoring margin uses a continuous diminishing curve, so every additional point changes the result by less than the point before it.
- Model weights and data quality are reviewed annually and tested for bias and instability.
Still to resolve
Questions testing must answer
- Which historical outcomes should determine whether the initial weights are well calibrated?
- How much predictive information, if any, should influence selection rather than matchup analysis?
- What independent body should audit model code, input accuracy, and committee explanations?
Discuss this rule
The proposal ends.
The thread starts here.
Open a focused discussion without leaving the policy chapter. Threads and replies are moderated before publication and also appear in the main community forum.