Texas Tech finishes ahead of Texas and misses the playoff behind them

Texas Tech projects to nearly ten regular-season wins and Texas to eight and a half. Those are means, and the two win distributions overlap heavily, so on any given draw either can finish ahead. Across all 50,000 seasons Texas Tech is 46.7% to make the playoff and Texas is 48.4%, so the team projected to the better record is the one on the outside.

Now put both at three losses. Texas is 80%. Texas Tech is 2%.

That's not a rounding difference or a quirk of one simulation. It's the whole argument about what the committee rewards, run 50,000 times.

Texas plays three games it isn't favored to win: Ohio State at home at 45.4%, Texas A&M on the road at 48.2%, LSU on the road at 49.3%. Texas Tech's hardest game all season is Arizona at home, and it wins that one 68.6% of the time. A 9-3 Texas arrives in December having beaten teams the committee ranks. A 10-2 Texas Tech arrives having beaten nobody it can name.

Part of that 80-against-2 is simply that a Texas which survives that schedule at 9-3 is a better team than a Texas Tech which fails to at 10-2, and the conditionals below can't separate the two. The part that does separate cleanly is the regression further down: holding record and rating and schedule equal, a win over a top-25 team is worth 0.69 of a loss. Texas Tech has nobody to collect one from.

Where each floor gives way

Texas is the only team on the board that survives a third loss comfortably, and Ohio State is the only other one that survives it at all, at 46%. Notre Dame falls furthest, 79% to 4%, because a third loss on a twelve-game independent schedule leaves nothing to point at and no title game to win.

Sit with Texas Tech's row. It's 45% at two losses, worse than every other contender here is at two, and it projects to finish with the second-best record of the ten.

What the committee is actually weighing

Take every FBS team from 2014 through 2025, drop the covid year, keep the ones that finished with five losses or fewer, and fit final ranking place on the resume. Unranked teams are censored in at 30th rather than dropped. That's 898 team-seasons, and every coefficient below is conditional on those two choices. A loss costs 3.46 ranking places, so everything else divides through by that and becomes a price in losses.

The bottom two rows are the arguments the sport has been having since 2014, and neither survives. An SEC or Big Ten badge is worth 0.13 losses against an ACC or Big 12 one at t -0.97, which is nothing. Playing a harder schedule, once you already count who you beat, is worth +0.01 losses at t +0.31, which is also nothing. Strength of schedule doesn't pay. Wins over ranked teams pay, and strength of schedule is only the opportunity to collect them.

Nobody argues about the top row. Teams that open ranked No. 1 finish about 1.59 losses better than unranked teams matching them on record, rating, schedule and quality wins. Read that as an association. A preseason ballot encodes returning production, recruiting and the portal, and one rating control leaves some of that in the residual, so the term may be measuring what August voters already knew rather than what December voters anchor on. That's one linear term, 0.082 losses per ranking place, so the ladder is the fit extrapolated rather than five separate results: No. 10 worth 0.85, crossing zero near No. 20, and opening 25th costing you 0.38 against going unranked. At t -7.22 the per-place term is solid. Both ends are the imposed straight line rather than estimated results, and I'd want another decade of ballots before leaning on either.

Second on the board is which conference you're in, at 1.09 losses, settled in a realignment meeting. Being a brand is third, at 1.04. A win over a top-25 team, the only one of the four a team can go out and earn, is fourth at 0.69.

Of the seasons where each team makes the field, this is the share arriving with a conference title in hand: Miami 81%, Texas Tech 63%, BYU 59%, Ohio State 49%, Georgia 45%, Indiana 41%, Oregon 31%, Penn State 31%, Texas 26%, Notre Dame 0%.

Miami is playing a knockout. Four in five of its berths come with the ACC trophy, so its 64.4% is largely a bet on winning one game in Charlotte. Ohio State and Georgia win their leagues in fewer than half the seasons they make the field, because they don't need to. Notre Dame sits at zero by construction and carries the highest number on the board anyway, 75.7%, which is the trade an independent makes: no title game to lose, and none to save you.

The rule nobody has priced

The 2026 format guarantees all four power champions a place regardless of ranking, gives a fifth automatic bid to the highest-ranked Group of Six team, and lets Notre Dame in automatically at top 12. Across 50,000 simulated seasons it changes the field 12.8% of the time.

That guarantee binds on one league in particular. An ACC champion finishes outside the top 12 in 10.5% of seasons. Big 12 winners do it 6.8% of the time. For the Big Ten it's 0.3%, and for the SEC 0.1%. A rule written to protect four leagues equally is, in practice, about the ACC.

Which is a strange thing to have written in a year after the ACC produced a runner-up and an unranked champion in the same December.

Ratings are CFBD's published preseason SP+, chosen over our own because Kalshi's traded win-total market sits closer to it, 0.52 games against 0.89 across 53 teams. Games already played are held at their real results: 98 were final when this ran, and 790 were simulated. Committee model fit on 2014-2025 and walked forward, recovering 11 of the committee's 12 per season. The simulation draws each team's true rating from its projection error, but it uses the committee coefficients at their point estimates, so every percentage here carries Monte Carlo and rating uncertainty and none of the uncertainty in the committee fit itself. Two of that fit's terms cannot be signed at all, which is why the tenths should not be read as precision. The model sees results and ratings and schedules, and nothing else. It has no view on quarterbacks, coordinators or injuries, so nothing here prices a personnel change. Reproducible from cfp_sim.py, cfp_committee.py, cfp_record_check.py, cfp_narratives.py, cfp_analysis.py.