The one totals angle that cleared our gate, and then failed its live test

Most of what this desk has tested does not work. That is not a complaint, it is the finding: our public graveyard holds more than forty angles that looked real in a spreadsheet and died against closing lines, from wind thresholds to altitude to returning production. We publish the kills because they are what make the survivors worth anything.

This article was written about a survivor. It is now about what happened when the survivor met games it had never seen. The short version: the 2026 forward test closed on September 20 and the angle failed it, at 47.3 percent unders on 131 graded games against a gate of 52.38 percent. The rule, the evidence, and the reasoning below are left as they were published before the season, because the point of writing them down first was that they could not be tidied up afterward.

The rule, stated before kickoff

FBS regular season, weeks 1 through 3. Closing total of 47 or higher. The side is the under, measured against the closing no-vig total.

That is the whole rule. No team filters, no weather overlay, no discretion. It was pre-registered before the 2026 season kicked off, which means the rule text, the qualifying threshold, and the grading method were written down and frozen while the result was still unknowable. Whatever happened in weeks 1 through 3 this season, we could not move the goalposts afterward.

The evidence behind it

Across 2018 through 2025, qualifying games went under 54.75 percent of the time, on a sample of 884 games. Break-even at standard pricing is 52.4 percent.

The mechanism is boring, which is a point in its favor. Early season totals leans on offenses that have not been seen yet: returning skill players are visible in August, rebuilt defenses are not, and the market prices the hype side of that asymmetry. High totals concentrate the effect, because a 47 plus number already assumes two functioning offenses in week 2.

We treat the history as reason to test, not proof. Plenty of dead angles had eight good seasons behind them, and a market that reads the same public studies we do can reprice any of this. Which is why the rule now has to pass a forward gate it cannot see coming.

The final result, graded in public

For 2026, the pre-registered gate was 52.38 percent unders or better on qualifying games across weeks 1 through 3. The count started with the first week 1 final and every qualifying game was graded automatically against its closing no-vig total, win or lose. The window closed with the last week 3 game on September 20, and the final count is 62 under, 69 over, 2 push, on 131 graded games: 47.3 percent unders. That is below the gate. The angle failed.

Week by week it went 19 under, 21 over and 1 push in week 1 (47.5 percent, 40 graded games), 24 under and 20 over in week 2 (54.5 percent, 44 graded games), and 19 under, 28 over and 1 push in week 3 (40.4 percent, 47 graded games). In all, 133 games qualified at a closing total of 47 or higher, and 8 of the 157 completed FBS games in the window had no captured pre-kickoff closing line and were skipped, not graded. One correction is on the record: the first two weeks of the count were graded off a lines table that kept only the latest capture per book, and 62 of 100 games had that capture overwritten after kickoff and were skipped. The grader was repointed at the full capture history on September 17, which is why the sample jumped from 34 to 85 games that day. Every week in the final count is graded from the full history.

What does it mean? At 131 games, 47.3 percent is not distinguishable from a coin flip, and a coin flip loses money at standard pricing. One three week window cannot prove that the historical 54.75 percent on 884 games was never real, and it does not have to. The deal was that the angle had to clear the gate on games it had never seen, and it did not. The plainest reading is that the backtest overstated the edge, helped along by a 47 point threshold that was chosen after looking at the data, or that the market has since closed whatever was there. The 2025 season had already slipped to 52.1 percent on 119 games, and we said so before kickoff.

So the angle goes where we said a failure would go. It is logged as rejected in the testing log at The Lab, and that entry cites this article. During weeks 1 through 3 it fed the weekly game board as a totals signal on qualifying games, labeled with its sample like every other signal there; it no longer does. The frozen rule, the season by season history, and the final count stay up on the early totals page. Nothing on that page was editable after the fact: qualifying games were determined by the closing number, grades came from final scores, and the rule was fixed before the first snap.

The limits, before anyone asks

This section was written before the season and is left as it was: three weeks is a small window, and at typical volume the 2026 test was expected to grade something like 100 to 150 qualifying games, enough to fail loudly but not enough to prove anything alone. It graded 131 and it failed. A passing grade would have extended the evidence without ending the question, and a failing grade works the same way in reverse: it does not prove the history was noise, it only means this desk will not act on it. The angle was always descriptive history, not a guarantee.

That is the deal this desk offers on every number: the rule in advance, the count in public, and the same treatment for a miss as a hit. This one was a miss.