Methods

Every number on this site can be reproduced from the published code and a free CFBD key. This page defines each one.

The point-in-time rule

A value computed "as of" a date uses only games that started before that date. For a committee release, the cutoff is midnight US Eastern at the start of the release day. A Tuesday-night game on a Tuesday release day therefore does not count, because its result was not known when the committee met. The code enforces this rule for every measure, and an automated test checks it on each build: adding future games, or flipping their results, must not change any value.

One measure reads the schedule beyond the cutoff, but never its results. Series balance needs to know whether a return game exists. Scheduling contracts are usually signed years in advance, so the existence of that return game was already known.

"As of" snapshots and ranks

For each season you can view the data as of any AP poll released before the first committee ranking, as of any committee ranking, or at the end of the season. Opponent ranks always come from the selected snapshot, not from the week the game was played. So with "Committee ranking #3" selected, every opponent shows its rank in that ranking. The rank at game time is shown as a separate column for reference. Before the first committee ranking of a season, ranks come from the AP poll and are labeled AP. Games after the snapshot's cutoff are listed as upcoming, without scores.

CFBD labels polls by week number, not by release date, so release dates are derived:

• Committee rankings come out on the first Tuesday after the week's games. The final ranking of the season comes out on the Sunday after conference championship weekend.
• AP polls come out on the Sunday after the week's games. The preseason poll counts as released the day before the season's first game.

Each derived committee date was checked against the published release dates for 2014–2025.

Schedule statistics

• Opponent record at game: the opponent's wins, losses and ties in games that started before this game, excluding games against the team being viewed.
• Opponent record (as of): the same, counting every game before the selected snapshot's cutoff. At the end of the season, that is the opponent's full-season record minus games against the team being viewed.
• Strength of schedule (SOS): the combined wins (ties count half) of every opponent already played, divided by their combined games, as of the snapshot. Games against the team being viewed are excluded, and each game counts once.
• Rest days: calendar days, US Eastern, since the team's previous game against an opponent in any division.
• Non-FBS opponent: an opponent that was not in FBS that season, meaning FCS, Division II, Division III or non-NCAA. A school that later moved up to FBS still counts as FCS for its earlier seasons.
• Games vs. ranked: games already played against teams ranked in the snapshot's poll.
• Opp rating: the opponent's team rating as of the snapshot.
• Difficulty: the chance a bubble team (see below: #4 through 2023, #12 from 2024) would lose this game, playing the same opponent at the same site, using ratings as of the snapshot. It is shown for upcoming games too. A win is worth exactly this much in wins above bubble.
• Still to play: games on the schedule after the snapshot, and their difficulty added up.
• Matchups nobody knew yet: bowl, playoff and conference championship games are left off a team's schedule until they are played, because nobody knew those matchups at the time of earlier snapshots. CFBD doesn't flag title games directly. They are found by their game notes, or, in seasons without notes, as a conference's only game in its last week, counted only in years that conference held a title game.

Matched schedules

When two teams are compared in the Schedule Explorer (and in Blind Résumé, with the opponents' names hidden), their games are lined up by difficulty: the chance a bubble team would lose that game, with the same opponent and site, from our ratings as of the selected date (see Team rating). First, as many pairs as possible are formed with one game from each team whose difficulties are within 8 points (5 or 12 if you choose), and among those pairings, the one with the smallest total gap. Sorting both schedules by difficulty and pairing in order is enough to find that best pairing exactly. The records in those look-alike games are the comparison: the same kind of games, and who won more of them. Only pairs where both games are played count toward the records.

The table lists every game once, toughest first: matched pairs side by side, and each game without a look-alike alone on its row, with the other side left empty. For the summary sentence, the games left over are paired hardest with hardest. Each of those pairings is labeled from the first team's side: tougher or easier when the two are more than the matching window apart, "much" when they are more than 25 points apart. A team with more games left over has extras with no counterpart. The Explorer's filters (site, opponents, non-conference) apply, so you can match only road games, for example. Blind Résumé uses played games only, from the date of that committee ranking, and shows each game's result, score, site and difficulty but not the opponent.

Click a matched pair (or tab to it and press Enter) to compare the two opponents. The popup shows the two opponents side by side (rating, where the game was played, record against everyone else, difficulty), because a game's difficulty comes from exactly two things: how good the opponent is and the site. Below that, the two opponents' own schedules are matched the same way, which shows where their ratings come from; pairs in there can be compared too. The popup uses the same date as the page and, in the Explorer, every game on the opponents' schedules (the filters apply only to the main view). When both teams played the same opponent, only the site differs. Opponents outside FBS have a rating but no schedule on the site, so the popup says so.

Team rating, schedule difficulty, wins above bubble, quality wins and loss cost

"Strength of schedule" by opponents' win percentage treats a 5-3 team that lost only to top-10 teams the same as a 5-3 team that lost to bad ones. These numbers fix that by first rating every team.

• Team rating: points better (+) or worse (−) than an average FBS team on a neutral field. It is fit to the score of every game played so far, in every division, with each margin capped at 28 points so running up the score earns nothing. Home field is worth about 2 points, as measured from the games. Each team starts the season at last season's rating, pulled 40% of the way back toward its division's average, and that starting point counts like 3 games, so by midseason this season's results decide the rating.
• Bubble team: the last team into that season's playoff field, by rating at the same date: the #4 FBS team for the four-team playoff (2014–2023, and every season before it), the #12 team for the twelve-team playoff (2024 on). The benchmark matches the decision the committee actually faced that year: in 2023 the question was "is this a top-4 résumé?", not "is it a top-12 résumé?". The Schedule Explorer also shows wins above bubble against the other bubble (#12 before 2024, #4 from 2024), so you can see when a comparison depends on that choice. Until November 2026 the site used #12 for every season; that flipped two famous verdicts (2017 Alabama vs. Ohio State, 2023 Florida State), see the decision log.
• Schedule difficulty: how many of a team's games the bubble team would expect to lose, playing the same opponents at the same sites. Higher means harder. It describes the schedule alone: it is the same whether the team won those games or lost them.
• Wins above bubble: the team's actual wins minus the wins the bubble team would expect against the same schedule. Beating a team the bubble team beats 95% of the time earns 0.05. Winning a game the bubble team wins only 40% of the time earns 0.60. Losses cost the same way.
• Quality wins: the credit from wins alone (each win earns 1 minus the bubble team's chance in that game).
• Loss cost: the credit lost in losses alone, so a close loss to a top team costs little. (This is not the formula's "bad losses," which counts losses to weak teams.) Each loss costs the bubble team's chance of winning that game: losing a game the bubble team wins 95% of the time costs 0.95, losing one it wins 40% of the time costs 0.40. Ties count half of each. Wins above bubble = quality wins − loss cost.
• Record strength: the chance a bubble team does at least as well against the same schedule: plays the same opponents at the same sites and wins at least as many games. Smaller is more impressive. 5% means a bubble team matches the record only 1 time in 20. Where wins above bubble adds up credit game by game, record strength asks about the whole record at once, so it rewards an unbeaten or one-loss record against a solid schedule and doesn't keep growing with extra easy wins. The College Football Playoff adopted a record-strength measure of this kind for its own committee in 2025 (Footballscoop, AP); ours is our own version built on our rating, not a copy of theirs.

Too close to call. Wins above bubble rests on choices that are reasonable but not forced: exactly which team is the bubble, how spread out win chances are, and where margins are capped. At every committee ranking we recompute it six more ways, moving one choice at a time: the bubble a few spots either way (#3 and #5 around #4; #9 and #15 around #12), win chances spread by 10 or 14 points instead of 12, and margins capped at 21 or 35 instead of 28. Two teams are too close to call when those versions disagree about which one is ahead, or when the gap is under 0.05 of a game. It is a test of the pair, not of each team: two teams with similar schedules move together, so a small gap between them can be solid, while the same gap between teams with very different schedules can flip. Make My Case marks these with ≈.

Short seasons. Wins above bubble adds up games, so a team that played fewer games can't earn as much (2020 Ohio State played six). When two teams compared on Make My Case played two or more games apart, the page says so and shows wins above bubble per 12 games next to the total. Per game alone tracked the committee worse, so it doesn't replace the total.

Win chances come from the rating gap. A 12-point edge wins about 84% of the time, which is what fit 2015–2023 FBS games best when each was predicted from ratings known before kickoff. Nothing here is tuned to match the committee. Even so, on the final committee ranking of 2014–2025, ordering teams by wins above bubble agrees with the committee's order at about 0.88 (1.0 would be identical). The exception is 2020, when the committee leaned on team strength more than on résumés.

Build Your Own Committee

Each measure you weight is first put on a common scale, measured from the FBS average at that date. Measures counted in games (wins above bubble, quality wins, loss cost, losses, ranked wins, schedule difficulty, FCS games and the other counts) share one unit: the spread (standard deviation) of wins above bubble across FBS teams at that date, usually about 2–3 games. So at the same weight, one ranked win counts as much as one game of wins above bubble; a measure that barely varies can't swing the ranking just because a one-game difference is large next to its own small spread. Percentages and team rating are scaled by their own spread. A team's score is the sum of weight × that value, and teams are ranked by score. Measures with no spread at a date (blank, or the same for every team) are hidden.

The page flags two kinds of weighting that tend to give odd rankings: putting weight on schedule difficulty, a schedule-only measure that credits a hard schedule even in losses, and weighting two measures that move almost in lockstep at that date (for example win % and losses), which counts one thing twice. Opponents' combined win % is not offered on this page for now: next to wins above bubble it swung rankings far more than any other measure, because it counts opponents' wins without asking how good those opponents were. It is still shown in the Schedule Explorer. "Order agreement" is the rank correlation between your order and the real one over the teams the real ranking includes. Your weights are saved in the page's link.

Head-to-head. Head-to-head isn't a measure a team has on its own, so it's applied after scoring. At setting g, a win over a team is worth g score points, but only when deciding the order of those two teams. The ranking shown is the order that maximizes the total, over every pair of teams, of the score margin of the team placed higher plus g when it beat the team below. A team can't jump a team it beat without passing everyone in between, and each pass costs that team's score margin, so sometimes the loser drops instead. We start from score order and move one team at a time while the total improves (a standard local search for this problem; it may stop short of the exact best order). Only FBS-vs-FBS games played by the selected date count; when two teams met more than once, the one with more wins counts, and a split series is ignored. "Head-to-head conflicts" lists every pair, both inside the top 25, where a team is ranked above a team that beat it, for the real ranking and for yours.

Why is one team above another? For any two teams, each weighted measure's edge is weight × (difference ÷ that measure's scale), so the edges add up to the gap between the two scores. Measures you haven't weighted are listed but not counted. When the lower team won their game, the page reruns your ranking at each head-to-head setting to report the smallest one that would put it ahead. Team links accept names, for example ?a=tennessee&b=alabama.

One-score losses. A loss costs a team the bubble team's chance of winning that game (1 − its difficulty), whatever the score. At setting c (0 to 100%), a loss by 8 points or fewer costs (1 − c) of that. The forgiven amount comes off loss cost and is added back to wins above bubble, so it changes your ranking only through those two measures. Win %, losses and head-to-head don't change.

Conference title-game loss. A team that reaches its conference title game plays one more strong opponent than a team that doesn't, and a loss there costs it wins above bubble that the team at home never risked (2024 SMU, 2020 Notre Dame). This setting counts that loss in full (the default), half, or not at all, in wins above bubble and loss cost only. The amount dropped is the loss's cost, the bubble team's chance of winning that game. Win %, losses and head-to-head don't change, and a title-game win always counts. Title games are found the same way as in the Schedule Explorer.

Projected playoff field. From 2024: the five highest-ranked conference champions plus the next seven (2026 is assumed to keep this format until the CFP says otherwise). 2014–2023: the top four. None before 2014. Until the title games are played, each conference's highest-ranked team stands in for its champion; independents can't take an automatic bid. Byes go to the four highest-ranked champions in 2024 and to the top four seeds from 2025 (straight seeding). The field is drawn for your ranking (seeds in the "Your field" column, a dashed line under the last team in on its own rank) and for the real ranking, using only its ranked teams. This is the same rule the Committee Formula page uses, except that page knows the actual champions.

Your team and what-ifs. Pick a team (it defaults to the favorite team chosen in the header) to highlight it, keep it in the table wherever it ranks, and see its rank, playoff status, best win (the hardest game won), worst loss (the easiest game lost) and wins above bubble in words. You can also pick each remaining game as a win or loss. A picked game is scored like a real one: a win adds the game's difficulty to wins above bubble and quality wins, a loss adds 1 − difficulty to loss cost, and the opponent gets the other side, including a ranked win if your team is ranked. It also counts as a head-to-head result. Team rating doesn't move, because it needs a score, and every other team's remaining games are left unplayed, so the result is the most the picks could move your team, not a forecast. Picks are saved in the link (&team=tennessee&picks=WWL).

Committee Formula

For each season since 2014 we fit one formula to all of that season's committee rankings. At every ranking, each pair of ranked teams is one example ("A was ranked above B"), and so is each pair of a team ranked #16–25 with one of the ten strongest unranked teams (by wins above bubble), so the formula also learns where the line at #25 falls. The model is a regularized logistic regression on the differences between the two teams' measures (a Bradley–Terry model): it learns one weight per measure, the same for every ranking that season. Measures are standardized over the season, so a weight is "how much one standard deviation of this measure moved a team up". Head-to-head is the exception: +1 if a team beat the other, per meeting.

The measures: win %, losses, ranked wins, bad losses (to teams outside our top 50 FBS ratings, or to any non-FBS team), opponents' win %, schedule difficulty, conference champion, power-conference membership, average margin (each game capped at ±21), head-to-head, and two probes: last week's rank and the preseason AP rank. The probes aren't things the committee says it uses; weight on them shows how much rankings lean on earlier opinion.

Point in time: every measure for a ranking uses only games that started before it came out. "Ranked wins" counts opponents ranked in the previous committee ranking (the last AP poll before the first one), because the ranking being explained wasn't known yet. The page also compares two other ways of counting an opponent's rank: when the game was played, and this week's ranking.

How well it fits: the share of pairs the formula orders the same way the committee did, both on the rankings it learned from and with each ranking predicted from the season's other rankings (the honest number). Also Kendall's tau over each top 25, and the playoff field the formula's order would produce, with the field rules applied as rules (top 4 through 2023; from 2024, the five highest-ranked conference champions plus the next seven). Baselines on the same pairs: the AP poll released two days before, last week's ranking, our rating alone and win % alone. The 90% intervals come from re-fitting 200 times on resampled teams; a weight counts as changed from one season to the next only when the two intervals don't overlap. The penalty strength was chosen by cross-validation across 2014–2025 (accuracy was nearly flat; we picked a value at the stable end).

Limits: measures overlap (win % and losses move together), so read weights together rather than one at a time. A season in progress has only a ranking or two and its weights are rough. The Improved Linear Model is listed as a baseline; see below.

The Improved Linear Model baseline

John A. Trono's Improved Linear Model (Saint Michael's College) is a published formula built to match the committee. It adds five numbers from the power rating in The Hidden Game of Football (Carroll, Palmer and Thorn): a team's average point margin and its schedule strength (the average rating of its opponents), once with full margins in games between FBS teams and once counting only wins and losses in every game (all non-FBS opponents treated as one team), then subtracts a penalty per loss. We use his weights exactly as published: the original set (fit to the 2014–2017 final top fours) and the 2025 set his second paper favors for the 12-team field (fit to the 2014–2023 final top twelves, 2020 excluded). So for those seasons it is being scored partly on data it was fit to. Like everything here, it sees only games before each ranking.

Our reproduction matches the paper's published 2024 scores to within a constant shift of about 2.5 points (where ratings are centered, which never changes the order) for 11 of the 12 teams we checked; LSU is about 10 points off for a reason we couldn't find (our LSU games match the official record). Over 2014–2019 our version picks 21 of the committee's 24 top-four teams (14 in the exact spot) where the paper reports 22 (15); it also matches the paper's 2020, 2023 and 2024 results. It differs only in two near-ties: in 2018 Georgia edges Oklahoma for #4 by 0.4 points (out of about 89), and in 2021 Notre Dame edges Cincinnati by 0.2; the paper had Oklahoma and Cincinnati. Small differences in the game data are enough to flip ties that close.

Win credit by rank or by strength

"Ranked wins" gives full credit for beating #25 and none for beating #26, and the same credit for #1 as for #25. The page can switch every part of the formula to strength-based credit instead: "quality wins" adds up, over a team's wins, the chance that a bubble team would have lost that same game (same opponent, same site, opponent's rating as of the ranking). Beating a team about as good as #1 earns nearly a full win; beating a weak team earns almost nothing. Everything else in the formula stays the same. Across 2014–2025 the two versions predict held-out rankings about equally well (92.2% rank-based, 92.3% strength-based); strength-based is ahead in 2015, 2017–2022 and 2025.

The strength ladder places the committee's top 25 by our team rating (points better than an average FBS team on a neutral field) instead of evenly spaced, so you can see where #5 and #6 are nearly the same team and where a real gap sits. A gap of 1.5 rating points or more (about a 5-point swing in neutral-field win chance) is drawn as a tier break.

Preseason influence

Three checks on whether the preseason AP poll still steers the committee, built on the same point-in-time measures. Each uses the season's formula refit with the nine résumé measures plus preseason rank, and without last week's rank (last week's rank already carries the preseason poll forward, so including it would hide the effect we're looking for). Its résumé part (everything but preseason rank) is the "résumé" below.

• Anchoring. The formula fit to each committee ranking on its own, with the weight on preseason rank turned into a plain comparison: two teams with the same résumé, one preseason #5 and one preseason unranked; how often is the #5 team ranked ahead? 50% means no pull. Seasons are pooled by ranking number (first ranking, second, …), weighting each season by how precise its estimate is, to see whether the pull fades.
• Twin résumés. Two teams at the same ranking whose résumés the formula calls a coin flip (either ahead 45–55% of the time), with preseason AP ranks 10 or more spots apart, at least one of them ranked, and no game between them. Each pair counts once a season, at the ranking where their résumés were closest. Without a preseason pull, the preseason favorite would be ahead about half the time. Inspired by critics' complaints about 2024 Arizona State, picked last in the Big 12 preseason.
• Stickiness. Week to week, how far a ranked team's rank moved for each spot its résumé rank moved (a team that drops out counts as #26). Because the résumé formula is imperfect, its noise alone makes this number fall below 1 for anyone, so the AP poll over the same weeks is the yardstick, not 1.

Limits: the résumé formula is itself fit to the committee's rankings, so whatever preseason pull it absorbs through correlated measures (a preseason favorite tends to have a better margin, say) isn't counted here. These numbers are more likely to understate the pull than overstate it.

Time Machine

The Time Machine re-ranks one committee ranking with another season's formula. Each formula keeps its own season's scale: a measure is standardized with that season's average and spread, so one more win or a harder schedule counts for what it counted that year, not for one standard deviation of this year's spread. The ranking itself is built the same way as the Committee Formula's: the linear score, then, among the top 40, the expected wins in pairwise contests, so head-to-head counts.

Résumé only (the default) uses each season's formula refit without last week's rank and the preseason poll. The full version keeps them, but last week's rank then comes from the season being re-ranked, which pulls every formula toward what this season's committee already said. Only finished seasons' formulas are offered, plus the season's own; formulas from 2014–2023 were fit to a four-team field and five power conferences, and "power conference" always means the re-ranked season's power conferences.

Committee Drift and Why It Changed

The weights come from the Committee Formula fits, one per season, with 90% bootstrap intervals. A weight changed only if its interval doesn't overlap the season before's. Each rule or metric change has a hand-written entry in layers: what the CFP said (Official), what was reported, what others argued, and our read, each with links to its sources.

Verdicts. Where a change should move a measure (say, "games against strong opponents count for more" → schedule difficulty up), the entry names that measure and the direction. The verdict compares the change's season with the last finished season before it, using the same overlap rule: Supported if every named measure moved the expected way, Partly supported if some did, Not supported if none did (overlap or the wrong way), and Too early while the season is unfinished. Format rules (bids, byes, seeding) are applied as rules, so they get no verdict. Our measures are proxies for the CFP's own: "schedule difficulty" is not the CFP's schedule-strength formula, which isn't published. A "Not supported" verdict means the rankings didn't move clearly on our measures, not that the committee ignored the change.

Alternate Rankings

Every column ranks the same teams on the same date, from games before the ranking's cutoff: our formula (full and résumé only), the last AP poll before the ranking, the Improved Linear Model with both of Trono's weight sets, our team rating (how good a team is), and wins above bubble (what it accomplished against its schedule). Agreement with the committee is Kendall's tau-b over the committee's top 25, with teams missing from a column tied last, and the overlap between the two top-n groups (n = playoff field size). Other people's open rankings are linked, not copied.

Blind Résumé

Pairs are two teams from one committee ranking (by default within four spots of each other), shown without names, conferences, last week's rank or the preseason poll. After you vote, the page shows who they were, which one the committee ranked higher, and the season formula's chance that the first is ranked higher (the logistic of the score gap, leaving out head-to-head). Votes are stored only in your browser; there's no shared vote collection yet.

Make My Case

The case on the numbers alone compares the team with each neighbor on measures that aren't fit to the committee: wins above bubble, record strength, quality wins, loss cost, head-to-head, conference title, team rating and schedule difficulty. Each is marked as a point for the team (✔), for the neighbor (✘), or too close to call (≈). Wins above bubble uses the test above; the others call a gap too close under 2 percentage points of record strength, 1 point of rating, or 0.1 of a game. Notes flag a different number of games played, a conference title-game loss, a comparison that flips with the other bubble, and facts the numbers can't see (an injury, what the committee chair said), with sources.

How the committee's pattern explains this rank uses the season's formula, which is fit to the committee and includes last week's rank, so it explains the committee rather than arguing with it. A team's formula score is the sum, over measures, of weight × (value − average) ÷ spread, with the average and spread taken over the teams the season's formula compares (the ranked teams plus the ten strongest unranked teams at each ranking). Each bar is one of those terms, so zero means "average for that group", not "average FBS team". Against each neighbor, the chance is the logistic of the score gap (head-to-head left out), and the edges listed are the measures whose terms differ most.

Rank score

A rank says who is ahead; a rank score also says by how much. The team at #1 gets 1.0, the team at #25 gets 25.0, and every other team sits where its score puts it: 1 + 24 × (S₁ − S) ÷ (S₁ − S₂₅), where S is the score behind the ranking and S₁, S₂₅ are the scores of the teams at #1 and #25. Past #25 the numbers keep going. With fewer than 25 teams, the last one is the anchor. On the 2025 selection-day ranking, the season formula's #1 and #2 are 1.0 and 1.7, there's a break after #5 (5.2, then 8.3), and #8 through #11 sit within one point of each other, right at the playoff line.

Each page uses its own score: Build Your Own Committee uses your weights, the Committee Formula page the season's formula. Make My Case and Case Files use the season's résumé-only formula (without last week's rank or the preseason poll), the same base as Stretch, so an old opinion can't hold a team's number up. (The full formula may become an option later.) A gap of 2.0 or more between neighbors, twice the average gap, starts a new tier: the gap strip shades it and tables draw a line. Head-to-head can rank a team just above one with a slightly better score; the rank score follows the score, so those two numbers sit out of order and the row is tagged H2H.

Case Files

A case file is a claim (“this team belongs at #N or better at this committee ranking”) plus evidence cards. A card is saved from Build Your Own Committee, Make My Case, the Time Machine or the Schedule Explorer: it keeps that tool's link (so Open live rebuilds the exact view with today's data) and a small copy of what the fan saw. A card can also be a point in the fan's own words. Case files are kept in your browser. A share link carries the whole case after the #, compressed; that part of a link is never sent to any server.

Every claim is scored the same way, whatever the cards say:

• Stretch (lower is better): how far the season's résumé-only formula (the formula refit without last week's rank and the preseason poll) has to change to put the team at #N. Each weight's change is divided by how much that weight has moved from one committee season to the next (its spread across the finished seasons, at least 0.1), and the changes are combined (square root of the sum of squares). So 1 means a change about as big as the committee typically makes in a year. 0 means the committee's own pattern already puts the team there. The search passes the teams ahead that are cheapest to pass first, nudging the weights a little at a time and re-ranking with head-to-head included, then trims the result back to the smallest share that still gets there; no weight may change sign. It's an upper bound on the smallest possible bend, so it's shown as “about”. Small: under 0.5; moderate: under 1.5; large: under 3; past 6 the claim is marked out of reach.
• Past committees that agree: the ranking's teams re-ranked by each finished season's résumé-only formula (the Time Machine's orders) and how many put the team at #N or better.
• Counterpoints: against the team now at #N, the team just above, and the next one up, the points from Make My Case's numbers-alone comparison that go the other way (head-to-head first), plus what's too close to call.
• Cards from Build Your Own Committee list what the fan changed from the Résumé starting point (wins above bubble alone) and flag settings that bend the ranking in known ways: a schedule-only measure weighted up, a negative weight, nothing measuring results, a heavy head-to-head setting.

Old cards keep working. Each card type has a version; when a tool's link changes, an upgrade rewrites old cards' links, and a retired tool's cards show their saved copy with the date and the reason. Saved cases and frozen share links are part of the site's tests, so a change that would break them fails before it ships.

Consistency Checker

The protocol tells the committee to separate comparable teams by head-to-head, results against common opponents and (through 2023) conference titles. At every ranking we list each pair in the top 25 where the order goes against one of those: A above B although B beat A (net of rematches), or although B beat a team that beat A. Comparable uses the season's résumé-only formula (no last week's rank, no preseason poll, no head-to-head): its chance that A belongs above B. From 35% to 65% the résumés are close, so the tiebreaker should have decided: flagged. Under 35%, B had the better résumé and the tiebreaker: also flagged, marked "lower team's résumé better". Over 65%, A's résumé was clearly better and the tiebreaker doesn't apply. Conference titles are checked at the last ranking of 2014–2023, for non-champions up to six spots above a champion. Pairs where wins above bubble is also too close to call (see Too close to call) are marked "WAB ≈", a second test of comparable that owes nothing to the committee. Margin of victory: the protocol says not to reward it, so each season shows whether the formula's weight on capped average margin is clearly above zero.

Why a rating moved

Our rating is a least-squares fit, so each team's rating meets its own equation: rating = (3 × preseason prior + the sum, over games, of capped margin + opponent's rating − home edge) ÷ (games + 3). Between two snapshots the change therefore splits exactly into two parts. This week's games: for each new game, how much the result beat or missed what the old rating expected, divided by games + 3. Past opponents: how much the ratings of teams already played changed, divided by games + 3; beating a team that later looks better lifts you. What's left ("other", typically under a tenth of a point) is the league re-centering on the average FBS team, the home edge being refit, and rounding.

What Moved

Each committee ranking against the one before (ranking #1 against the AP poll before it). Games are those that kicked off after the previous ranking's cutoff and before this one's. The formula column is the change in the team's score in that season's formula, with the three terms that changed most; terms change from new results and also from opponents' results changing what earlier games were worth.

Résumé vs. Power

Résumé is wins above bubble (wins beyond what a bubble team would expect against the same schedule); power is our team rating. Both use only games before the ranking. "Résumé ahead of power" ranks teams by the gap between the two, each standardized over the ranked teams.

Team History and Conferences

Team History shows a team's committee, formula and AP rank at every ranking since 2014, its record, rating and wins above bubble at the season's last ranking, and whether it made the committee's playoff field with the field rules applied. Conferences use each season's latest snapshot: average and median member rating, ranked members, non-conference record against FBS teams from other conferences (also just against power conferences), games against non-FBS teams, and average schedule difficulty so far.

Predict the Reveal

Order 25 teams before a ranking comes out. Points: 3 for the exact spot, 2 within two spots, 1 for a team that's in the top 25 anywhere. Three yardsticks get the same score: "no change from last ranking", the AP poll released before the ranking, and our forecast. Practice mode hides a past ranking and shows the games since the one before. Predictions are stored only in your browser.

What the Chair Said

Entries are researched by hand from news coverage of the chair's remarks after each ranking. We paraphrase rather than quote and link the coverage. Each gets our numbers as of that ranking and a verdict: Backed when our measures point the same way as the stated reason, Mixed when some do and some don't, Not backed when they point the other way. Our measures aren't the committee's own, so a verdict is a comparison, not a ruling.

Next Ranking Forecast

Two orders, averaged. The first is the season's own committee formula (see Committee Formula), including last week's rank, refit on only the rankings already released and applied to every game played before the next ranking's cutoff. The second is the latest AP poll. Each team's forecast spot is the average of its spot in the two (outside the AP top 25 counts as 26th), with the formula breaking ties. Before averaging, we scored both alone: from 2014 to 2025, each ranking from #2 on forecast only from what was known before it, the formula alone missed by 1.7 spots on average, the AP poll by 1.5, "no change" by 2.0, and the average of the two by 1.3. We tried the plain average only, with no tuned weights, so that number isn't flattered by fitting. The track record on the page is the same backtest. Ranking #1 isn't forecast: there's no earlier ranking to fit to. The playoff field in the forecast uses only conference champions already crowned.

What-If

Every game left that involves a team in the committee's top 25, the formula's top 40 or a conference-title contender starts with our rating's favorite winning. Each pick updates the teams' résumé numbers the way a real result would: record and win share, losses, ranked wins (against the latest committee top 25), losses to teams outside our top 50, schedule difficulty and wins above bubble (each game's credit is computed from today's ratings, same opponent and site). Average margin moves as if the game were decided by 14 points, about the typical capped FBS margin. Some things stay as they are today: team ratings, opponents' records and the top-25 and top-50 lines. The ranking uses a finished season's résumé-only formula, because a projection several weeks ahead can't know the rankings in between, and last week's rank would only echo today. Head-to-head counts the way it does on the formula page. Conference title games count as games: one already on the schedule is picked like any other game; for one not set yet, you choose the two teams (they start as the likeliest pair from Title Paths) and the winner, and it is played on a neutral field with the same credit rules. Its winner is the conference champion. The field takes the five best-ranked champions plus the next seven, and is seeded by the season's rule (see Selection Day).

Title Paths

For each FBS conference of eight or more teams, every conference game left is played out 20,000 times with our rating's win chances (home edge included). The two best conference records reach the title game (the Sun Belt, the one conference with divisions: each division's best record), and the title game is a neutral-site game between them. Army–Navy doesn't count in the American Athletic standings, as the conference rules say.

Ties go through the tiebreakers the conferences publish, in the order most of them share: (1) head-to-head among the tied teams (for three or more, the combined record among them if they all played each other, or a team that beat all the others); (2) record against common conference opponents; (3) outside the four power conferences, record against the highest-placed common opponent, then the next; (4) the combined conference win share of each team's conference opponents; (5) a ranking. The conferences use the CFP ranking or a computer composite at step 5; we use our own rating. The Mountain West and American Athletic go from head-to-head straight to rankings, and with three or more teams tied that didn't all meet, take the best-ranked teams outright. When a step separates some of the tied teams but leaves two or more still level, those go back to step 1. On the final standings of 2022–2025 (every conference without divisions, plus the Sun Belt), these steps pick the two teams that actually played in 25 of 26 title games; the old shortcut (record, then our rating) picked 18. The miss is the 2025 Mountain West, where the computer composite ranked UNLV above San Diego State and our rating has it the other way. "Needs a tiebreaker" shows how often a team was level on record with the second-place team. "Clinched a spot by record" and "Can't reach it by record" are worked out from records alone, without tiebreakers, and assume equal conference schedules. Once a title game is on the schedule, its two teams are used as they are.

Conferences

FBS teams only, grouped by the conference they played in that season, as of the latest snapshot. Strength is our team rating (see Team rating): the conference's average and median, and its best team. Non-conference record counts games against FBS teams from other conferences; "against power conferences" keeps only the SEC, Big Ten, Big 12 and ACC (plus the Pac-12 through 2023). Games against lower divisions count every game against a non-FBS team, and how many of the conference's teams played one. Schedule difficulty is the average, over the conference's teams, of the games a bubble team would expect to lose against each team's schedule.

Selection Day

The field and seeds follow the CFP's format for each season. 2014–2023: four teams, seeded by ranking (semifinals 1 v 4 and 2 v 3). 2024: twelve teams, the five best-ranked conference champions plus the next seven; the four best-ranked champions got seeds 1–4 and the first-round byes, and the other eight were seeded 5–12 by ranking. Since 2025: the same twelve, seeded straight by ranking, so the top four get the byes whether or not they won their conference. We assume the 2025 rules hold for 2026 until the CFP says otherwise. First round: 5 v 12, 6 v 11, 7 v 10 and 8 v 9 at the higher seed's campus; the winners meet 4, 3, 2 and 1. These rules reproduce the real 2024 and 2025 brackets seed for seed.

"Our formula" is the season's committee formula (see Committee Formula) on the final ranking's date, given the same conference champions. Before selection day the page uses the latest ranking, and the best-ranked team in each conference stands in as its champion, as in the CFP's own weekly "if the season ended today" bracket. Playoff games are the postseason games between two teams in the committee's field. For teams one side picked and the other left out, we list their postseason games, which are a hint at most: a bowl is a different game against a different kind of opponent.

Scheduling measures (modules)

Each measure is a self-contained module with a neutral name, a definition, a default weight and a credit line. The list below is generated from the same registry that the Build Your Own Committee page uses.

Data corrections

The raw CFBD data is never edited. These fixes are applied at build time:

• 2023 final committee ranking: CFBD lists Georgia and Florida State tied at #5. The committee ranked Florida State #5 and Georgia #6 (source).
• Postseason "committee" polls: CFBD lists a postseason committee ranking for 2014, 2020 and 2023, but the committee publishes nothing after the bowls.
  • 2014: a duplicate of the final ranking, dropped.
  • 2023: a copy of the AP post-bowl poll, dropped.
  • 2020: the real Dec. 20 selection-day ranking, kept as that season's final release.
• Duplicate games: 16 games from 2004 and 2008 appear twice under different IDs. Only one copy of each is kept.
• Conference title games without a note: five title games carry no note and aren't the conference's only game that week (2018 Pac-12, 2020 SEC, Big Ten and Pac-12, 2021 Pac-12). They're listed by game so conference champions and pre-game snapshots treat them correctly.
• Missing 2019 ranking: CFBD has no committee ranking for Nov. 26, 2019. It is added by hand from the published top 25 (source).

The data coverage report lists every season's releases and fixes, and is regenerated with each data refresh.

Divisions before 1978

FBS (originally Division I-A) began in 1978. For earlier seasons, CFBD tags the major college programs of the time as FBS, and this site follows that tagging.