Notre Dame · 10-2
67.2 / 100 · 35% weight
78.4 / 100 · 25% weight
85.2 / 100 · 15% weight
67.9 / 100 · 15% weight
79.5 / 100 · 7.5% weight
0.0 / 100 · 2.5% weight
Account
Worked examples · FBS · NRR v0.57
Follow the evidence.
Explain the order.
Two teams can share a record and have different résumés. A team can also have stronger wins and finish lower. Follow the contributions, then decide which assumptions you would challenge.
A short reading route
Résumé first. Forecasting is a separate question.
- Read the result: this model weighs the season already played. It does not settle who would win tomorrow.
- Find the tradeoff: compare the largest contributions, including evidence favoring the lower-ranked team.
- Question the assumptions: the scenarios below show how changed priorities can reverse an order. They are not new adopted weights.
Every example uses the 2025 Selection Sunday archive, cutoff December 7, 2025 at 12:00 UTC (7 a.m. Eastern). These are historical v0.57 calculations, not today’s rankings. An Index gap of 3.295 is not a prediction of a 3.295-point game.
Comparison 1 · Same frozen checkpoint
Stronger wins do not erase damaging losses.
73.570#9 Notre Dame · 10-2
70.276#12 Alabama · 10-3
+3.295Index gap favoring Notre Dame
Alabama’s Quality-Win Depth score is higher: 89.357 versus Notre Dame’s 67.154. Its win portfolio includes Georgia, Vanderbilt and Missouri. Notre Dame’s includes Southern California and Pittsburgh. Those wins give Alabama the larger Quality-Win Depth contribution.
Notre Dame still finishes ahead. Its two losses were 40–41 to Texas A&M and 24–27 to Miami. Alabama lost 17–31 to Florida State, 21–23 to Oklahoma and 7–28 to Georgia. The model’s Loss Integrity score is higher when losses do less damage: Notre Dame’s 67.941 versus Alabama’s 18.992. Notre Dame also leads in performance, control and foundation. The table shows their combined effect; no single score alone determines the order.
Compare the evidence on the same scale
Each bar shows a component score from 0 to 100; higher is more favorable under this model. Higher Loss Integrity means less damaging losses. A bar is not a win probability, a percentile or a prediction. Weight sets importance; the separate Index contribution includes that weight and this team’s foundation.
67.2 / 100 · 35% weight
78.4 / 100 · 25% weight
85.2 / 100 · 15% weight
67.9 / 100 · 15% weight
79.5 / 100 · 7.5% weight
0.0 / 100 · 2.5% weight
Alabama · 10-3
89.4 / 100 · 35% weight
77.7 / 100 · 25% weight
72.4 / 100 · 15% weight
19.0 / 100 · 15% weight
62.4 / 100 · 7.5% weight
0.0 / 100 · 2.5% weight
The longer bar need not have the larger impact on the total: weights and foundation also matter. Open the arithmetic below to reconcile every contribution. Both teams share the same historical checkpoint and published default weights.
Inspect the arithmetic: scores, weights and Index contributions
A component score is on a 0–100 scale. Multiply it by its weight ÷ 100 to obtain weighted composite points, then by this team’s foundation and 0.85 to obtain the Index contribution shown below. Higher Loss Integrity means less damaging losses. All six component scores run in the higher-is-better direction.
| Component · weight | Notre Dame score | Notre Dame Index contribution | Alabama score | Alabama Index contribution |
|---|---|---|---|---|
| Quality-Win Depth · 35% | 67.154 | 17.482 | 89.357 | 22.761 |
| Schedule-Adjusted Record · 25% | 78.375 | 14.574 | 77.663 | 14.130 |
| Team Performance · 15% | 85.159 | 9.501 | 72.412 | 7.905 |
| Loss Integrity · 15% | 67.941 | 7.580 | 18.992 | 2.073 |
| Win Control · 7.5% | 79.468 | 4.433 | 62.394 | 3.406 |
| Conference title · 2.5% | 0.000 | 0.000 | 0.000 | 0.000 |
Foundation multipliers: Notre Dame 0.875057; Alabama 0.856209. Head-to-head credits: both 0. Add the six contributions above to the common baseline of 20 to reproduce each Index. Rounded displayed cells can differ by a few thousandths when summed.
For example, Notre Dame Quality-Win Depth: 67.154 × 0.35 = 23.504 weighted composite points; × 0.875057 × 0.85 = 17.482 Index points. The foundation is a multiplier already included in every contribution, not another amount to add.
What could change this order?
As an illustrative policy change, move 10 percentage points from Loss Integrity to Quality-Win Depth: 45% wins, 5% losses; keep all other weights and inputs fixed. The resulting pair is Notre Dame 73.512, Alabama 75.397: the order reverses.
This is a fixed-component rescore, not a full refit or a recommendation. Corrected game inputs or opponent values would require a fresh full calculation. None of these examples establishes optimal weights or predicts a matchup.
Comparison 2 · Same frozen checkpoint
The same record does not mean the same résumé.
75.270#8 Oklahoma · 10-2
66.200#15 Utah · 10-2
+9.069Index gap favoring Oklahoma
Oklahoma and Utah both finished 10–2 at this checkpoint. Oklahoma’s recorded wins include Alabama (23–21, road), Michigan (24–13, home) and Tennessee (33–27, road). Utah’s win evidence includes Arizona State (42–10, home), Cincinnati (45–14, home) and Kansas (31–21, road).
Oklahoma leads in Quality-Win Depth and Schedule-Adjusted Record. Utah leads in performance, control and slightly in Loss Integrity. The schedule-caliber diagnostic is 56.429 for Oklahoma and 52.428 for Utah. That diagnostic has zero separate weight; schedule information already feeds other calculations and foundation. This comparison does not isolate a causal schedule effect.
Compare the evidence on the same scale
Each bar shows a component score from 0 to 100; higher is more favorable under this model. Higher Loss Integrity means less damaging losses. A bar is not a win probability, a percentile or a prediction. Weight sets importance; the separate Index contribution includes that weight and this team’s foundation.
Oklahoma · 10-2
86.0 / 100 · 35% weight
81.2 / 100 · 25% weight
70.4 / 100 · 15% weight
58.1 / 100 · 15% weight
61.1 / 100 · 7.5% weight
0.0 / 100 · 2.5% weight
Utah · 10-2
50.6 / 100 · 35% weight
77.7 / 100 · 25% weight
76.4 / 100 · 15% weight
59.4 / 100 · 15% weight
75.3 / 100 · 7.5% weight
0.0 / 100 · 2.5% weight
The longer bar need not have the larger impact on the total: weights and foundation also matter. Open the arithmetic below to reconcile every contribution. Both teams share the same historical checkpoint and published default weights.
Inspect the arithmetic: scores, weights and Index contributions
A component score is on a 0–100 scale. Multiply it by its weight ÷ 100 to obtain weighted composite points, then by this team’s foundation and 0.85 to obtain the Index contribution shown below. Higher Loss Integrity means less damaging losses. All six component scores run in the higher-is-better direction.
| Component · weight | Oklahoma score | Oklahoma Index contribution | Utah score | Utah Index contribution |
|---|---|---|---|---|
| Quality-Win Depth · 35% | 86.015 | 22.405 | 50.615 | 12.957 |
| Schedule-Adjusted Record · 25% | 81.236 | 15.115 | 77.714 | 14.210 |
| Team Performance · 15% | 70.408 | 7.860 | 76.427 | 8.385 |
| Loss Integrity · 15% | 58.050 | 6.480 | 59.425 | 6.519 |
| Win Control · 7.5% | 61.073 | 3.409 | 75.276 | 4.129 |
| Conference title · 2.5% | 0.000 | 0.000 | 0.000 | 0.000 |
Foundation multipliers: Oklahoma 0.875573; Utah 0.860462. Head-to-head credits: both 0. Add the six contributions above to the common baseline of 20 to reproduce each Index. Rounded displayed cells can differ by a few thousandths when summed.
For example, Oklahoma Quality-Win Depth: 86.015 × 0.35 = 30.105 weighted composite points; × 0.875573 × 0.85 = 22.405 Index points. The foundation is a multiplier already included in every contribution, not another amount to add.
What could change this order?
As a deliberately large policy change, move 30 percentage points from Quality-Win Depth to Team Performance: 5% wins, 45% performance; keep all other weights and inputs fixed. The resulting pair is Oklahoma 71.785, Utah 71.864: the order reverses.
This is a fixed-component rescore, not a full refit or a recommendation. The reversal requires a large change and the resulting gap is only 0.079; it does not show this pair flips under small adjustments. Corrected game inputs or opponent values would require a fresh full calculation. None of these examples establishes optimal weights or predicts a matchup.
Undefeated is an achievement. Certainty is a different claim.
Winning every game is evidence already earned. It does not establish certainty against opponents that have not been played. NRR summarizes completed results under published rules; its Index is not a win probability.
- What the bars show
- Component scores under the displayed rules and checkpoint.
- What the policy examples show
- How changing priorities can change the order, with the components fixed. Those ranges are not confidence or credible intervals.
- What a forecast would require
- A separate probabilistic model, declared assumptions and evaluation on unseen games. No predictive uncertainty band is estimated on this page.
Why uncertainty needs a separate model
Some unregularized pairwise maximum-likelihood models can produce unbounded strength estimates when outcomes separate teams. That is a model-specific inference problem, not evidence that the finite NRR composite here has an infinite score. Bayesian or other uncertainty methods need an explicit model and assumptions; no method has been adopted merely to add a band to this display.
Ready to test · No reader results yet
Can five readers explain it back?
This page now adds component bars before the arithmetic. The presentation is ready to test; it has not been shown to improve comprehension. Use this as a small usability exercise: recruit five willing adults, ideally three unfamiliar with the proposal and two familiar with college football. Record anonymous IDs R01–R05 and prior familiarity, not names or contact details. Record whether readers first saw the bars or the text-only route, and preserve their exact answers; do not call five sessions an effectiveness trial. Five readers can expose confusing wording; they cannot validate NRR or represent all fans.
- Give each reader the short route and both examples. Allow up to ten minutes. Do not explain the answer while they read; record where they pause or ask for help.
- Ask the five questions below before revealing the answer guide. Capture their own words and whether they needed a prompt. A disagreement with the weights is not a comprehension failure.
- Classify each response as correct unprompted, correct after a prompt, incorrect or unclear. Log the confusing label and location. Revise any recurring confusion seen in at least two readers; treat any confusion between Index units and predicted game points as a priority.
- Keep the first observations, version the revised wording, and retest affected questions with actual readers. Report the observed counts and limitations rather than declaring success from a completed worksheet.
Five questions to ask
- Why does Notre Dame lead Alabama here even though Alabama has the higher Quality-Win Depth score?
- Why can Oklahoma and Utah have the same record and different Index values? Which evidence favors Utah?
- What is the difference between a bar’s component score, its weight and its Index contribution? Is a score of 80 an 80% chance of winning?
- Does an Index gap predict the game’s point margin or identify the strongest team tomorrow? Are the policy scenarios statistical confidence intervals?
- Name an assumption or corrected piece of evidence that could change the ordering. What stays fixed in the displayed policy scenarios?
Facilitator answer guide — open after the questions
- Notre Dame’s loss, performance, control and foundation advantages outweigh Alabama’s larger win contribution under these weights.
- Records omit opponent quality, venue and performance context. Utah leads in performance, control and slightly in Loss Integrity; Oklahoma’s win and adjusted-record contributions dominate.
- The score is 0–100; weight is the share of the composite; Index contribution includes weight, team foundation and the 0.85 scale. The shared 20 baseline is added once. A component score of 80 is not an 80% win probability.
- No. These are résumé calculations, and the policy scenarios are not confidence intervals. Matchup prediction needs a separately evaluated predictive model.
- The displayed weight transfers reverse the pair with components, opponents, foundation and head-to-head fixed. Corrected inputs need a full recalculation; disagreement with priorities can be reasonable.
Download the blank five-reader worksheet and protocol (TXT). No responses are collected by this page. Reader recruitment, observations and subsequent revisions remain pending.
Reproduce the comparison
Frozen inputs. Visible limits.
Prepared October 15, 2026. The examples reconstruct the canonical v0.57 Index from the archive’s six-decimal components and foundation. Last-digit differences from an unrounded full refit are possible. Pair selection is purposeful teaching material drawn from historical development data; it is not a holdout, independent validation or proof that the higher team deserves its place.
Run node scripts/build-nrr-reader-examples.mjs --check in the matching source. The check verifies the frozen model/archive hashes, unique game IDs and record counts for the four selected teams, component reconciliation and both displayed policy reversals. It does not audit every archive game or reproduce raw-data opponent estimates.