Before the season this site published the unbeaten curve of 2024: 134 perfect records becoming 96 after one week, 60 after two, 43 after three, and one by late November. It was offered as the shape a season takes, and a shape is only worth anything if the next season follows it. Three weeks of 2026 are now in the books, and 2026 has not followed it. Through Monday, September 21, 34 of 136 FBS teams are unbeaten. At the same point of 2024 there were 43 of 134. As a share of the field that is 25.0% against 32.1%, and 2026 has been behind at every checkpoint.
The gap is not spread evenly. Almost all of it opened in the first ten days, and a surprising part of it came from a direction the 2024 season barely used at all.
What was pulled, and the rule for who counts
data_layer/build_weeks_0_3_2026_results.py walks ESPN's public college-football scoreboard one UTC day at a time, August 28 through September 22, and keeps every game involving an FBS team whose Eastern kickoff date falls between August 28 and Monday, September 21. That is 257 games, all final, none tied, retrieved the morning of September 23. The builder refuses to write if a kept game is unfinished. Week two holds 86 of them, the same count the site's week-two results file froze on September 12, which is the check that the two pulls agree.
Who is FBS is the house rule every 2026 piece here has used: a team ESPN tags with one of the eleven FBS conference ids, except North Dakota State and Sacramento State, which ESPN tags that way and the site treats as FCS programs. That gives 136 FBS teams. The exception matters below, and the piece shows the numbers both ways.
The 2024 figures are not retyped. They are imported from the 2024 piece's own ledger, which re-derives them from the 2024 cache every time it runs. Snapshots use that piece's convention, through each Monday, so week one includes the Labor Day game in both years.
charts/chart_unbeaten_2026_check.py — 34 checks.The two curves, side by side
| Through | 2026 unbeaten | Share | 2024 unbeaten | Share | 2026 behind by |
|---|---|---|---|---|---|
| Week 0 | 129 | 94.9% | 131 | 97.8% | 2 |
| Week 1 | 84 | 61.8% | 96 | 71.6% | 12 |
| Week 2 | 51 | 37.5% | 60 | 44.8% | 9 |
| Week 3 | 34 | 25.0% | 43 | 32.1% | 9 |
The preseason piece's most useful number was its null: if every game were a coin flip, how many teams would still be unbeaten? In 2024, one week in, the answer was 66.0 and the real season kept 96, 1.45 times as many, which the preseason piece traced to early games against weaker opponents. 2026's null after week one is 65.5, and 2026 kept 84: 1.28 times. The early-season cushion that makes the unbeaten count stick is thinner this year. By week three 2026 has 34 unbeatens against a null of 17.19, so the ratio has widened to 1.98 as the easy September games pile up behind the survivors, but the survivors started from a lower base.
The mirror count is mixed: 2026 had 45 winless teams after week one against 2024's 37, then 13 and 7 against 18 and 12.
Where the extra losses came from
Every first loss is one of three things. A collision is a loss to a team that was also unbeaten at kickoff, which is how most early deaths happen because most teams are unbeaten early. An ambush is a loss to an FBS team that already had one. And then there is the non-FBS loss.
| Week | 2026: collision / ambush / non-FBS | 2026 total | 2024 total |
|---|---|---|---|
| 0 | 6 / 0 / 1 | 7 | 3 |
| 1 | 41 / 1 / 3 | 45 | 35 |
| 2 | 23 / 8 / 2 | 33 | 36 |
| 3 | 12 / 5 / 0 | 17 | 17 |
Through three weeks 2026 has produced 102 first losses against 2024's 91. The whole difference, and more, is in weeks zero and one: 52 against 38, fourteen extra. Weeks two and three produced 50 against 53, three fewer. After the opening ten days the two seasons are losing unbeatens at almost exactly the same rate. The ledger balances both ways: 102 first losses plus 34 survivors is 136, and 91 plus 43 is 134.
Week zero itself is part of the story. Fourteen FBS teams played in 2026's week zero against six in 2024's, so more teams had the chance to lose before September. Seven did.
The direction 2024 barely used
The 2024 piece counted every first loss of the season and found exactly one that came from a non-FBS opponent. 2026 has had six in three weeks.
| Date | Lost its unbeaten record | To | Score |
|---|---|---|---|
| Aug 29 | Jacksonville State | North Dakota State | 7–33 |
| Sep 5 | Bowling Green | Tarleton State | 13–20 |
| Sep 5 | Charlotte | The Citadel | 41–43 |
| Sep 5 | Utah State | Idaho State | 17–29 |
| Sep 12 | Air Force | North Dakota State | 32–38 |
| Sep 12 | Nevada | Montana State | 7–20 |
Two of the six are North Dakota State, and that is exactly where the counting rule bites. ESPN tags North Dakota State with an FBS conference id; the site treats it as an FCS program, as every 2026 piece here has. Count it ESPN's way and the six become four. Either way it is several times 2024's full-season total, in three weeks.
The same comparison in the other direction: FBS teams played 70 games against non-FBS opposition while still unbeaten and won 64, losing 8.57% of them. The 2024 season played 78 such games and won 77, losing 1.28%. Two more FBS teams, Northern Illinois and UL Monroe, lost to FCS opponents after they had already been beaten, so across all 105 FBS-versus-non-FBS games in the window the FBS side went 97–8. The buy-game piece measured how lopsided these games usually are; this September has been less so.
What this does and does not say
It says the 2024 curve was a poor template for the first ten days of 2026, and a good one after that. From week two on the two seasons shed survivors at similar rates: 2026 lost 39.3% of its remaining unbeatens in week two and 33.3% in week three, where 2024 lost 37.5% and 28.3%. The level is lower because 2026 started faster.
It does not say 2026 is a more upset-prone season in general. Six non-FBS upsets, or four, is a small count, and small counts move a lot from year to year; one season against another is not a trend. The FBS rule also changes the answer: the two seasons are compared under the rules each piece used, the 2024 cache's appearance-count proxy and the 2026 conference-id rule, which could classify a borderline program differently. And the 2024 figures are from this site's cache of that season, not an official record, as the 2024 piece said at the time.
The one number worth watching is the survivor count at the end of September. 2024 had 19 unbeatens through September 30. At 2026's current pace it will have fewer. Whether the gap narrows once conference play starts, when collisions become rarer and ambushes more common, is the part of the curve the preseason piece got right, and the part 2026 has not yet reached.
Reproducibility: python data_layer/build_weeks_0_3_2026_results.py rebuilds the 2026 file from ESPN, and python charts/chart_unbeaten_2026_check.py recomputes every figure on this page and redraws the chart, failing if any of its 34 checks does not hold. Written September 23, 2026, before week four.