The bet that needed Świątek to lose

Two cards ago the rules killed a bet family on schedule. Last card they kept one. This card they did both at once — and then the autopsy on the dead one found something better than a verdict.

The table

Same canonical cut as every audit: every graded bet where the model claimed a 10-point-plus edge over one book's pick'em market, by family. No blending — the blend is how a losing family hides inside a winning average.

Family	bets	model said	market priced	actually hit	verdict
Batter — hits	627	65%	50%	55% ✓	kept again 8/15
Batter — strikeouts	203	69%	50%	53% ✓	kept 8/15 (early)
Pitcher — strikeouts (post-kill)	123	70%	51%	54% ✓	re-entered at half stakes
Tennis props (pooled)	35	72%	51%	37% ✗	tripwire fired → benched

The tripwire, finally

Three report cards in a row said the same thing: tennis looks awful, the sample is short of the pre-committed floor, so nothing happens. This morning the pooled sample crossed 30 — 35 bets, 14 points worse than the market, while the model claimed 72% — and the tripwire I wrote in June did what it was written to do. All three tennis prop families went report-only: still predicted, still graded, not stakeable, and flagged that way anywhere the model surfaces them.

That part was mechanical. The interesting part is why.

The autopsy

The written suspects were wrong. I'd guessed Wimbledon's best-of-five format, or grass. The graded bets say otherwise: cut the sample by side and it splits clean in half. Tennis "over" bets on games won: 13 bets, 13 points better than the market — better than my batter props. Tennis "under" bets on games won: 16 bets, 34 points worse than the market. And the more confident the model got on an under, the worse it did — bets it priced at 75%-plus hit 29%.

Here's the mechanism, and it's embarrassingly simple once you see it. A tennis match winner never wins fewer than 12 games — winning two sets means winning at least 12, even 6-0 6-0. The prop lines sit at 10.5 to 13.5. So "under 11.5 games" isn't really a stat bet at all: it's a disguised bet that the player loses the match, and probably loses it badly.

My model doesn't know about that cliff. It predicts a smooth count of expected games, and whenever that estimate drifts a little below 12, the smooth math sprays probability across outcomes that mostly require a loss. The result: the model kept issuing supremely confident unders on the best players in the world. It had Pegula at 78% to win 11 games or fewer. It faded Świątek, Sabalenka, Fritz, Rybakina. Of the confident unders I could match to results, five of the six players won the match outright — which meant the bet was dead before the second set ended. The model wasn't unlucky. It was betting against a floor it couldn't see, and the market priced the floor correctly every time.

The fix

Same day, that bet shape stopped existing: the model can no longer build a games-won "under" at all, at any claimed edge. Not down-weighted — removed, the way you'd remove a bet you know is malformed rather than merely unlucky. The surviving side (overs, plus aces — whose losses trace mostly to a surface-labeling bug fixed back in June) is what the tennis families' re-entry case will be built on: 100 fresh graded bets at 2-plus points over market, counted only from the fix. With the American hard-court swing and the US Open ahead, that sample accrues now, not next season. Teaching the model about the floor properly — modeling the match outcome and the game count together — is offseason work, and the bench is where the families wait for it.

The corpse climbs the ladder

Meanwhile the family the rules killed in July kept arguing from beyond the grave, and this morning the argument cleared the written bar: 123 unstaked post-kill bets at 3.1 points over the market. The re-entry rule — written before the kill, precisely so this moment wouldn't be a judgment call — says 100-plus at 2-plus earns back half stakes. So pitcher strikeouts is live again, at half weight, with every ticket it touches sized down accordingly.

Worth saying plainly: the resurrection got less impressive as it grew — +4.0 at 90 bets, +3.1 at 123. That decay is exactly why the ladder's next rung demands a fresh hundred at 3-plus for full weight, and why a slide back to zero re-kills it. Half stakes is not a vote of confidence. It's the correct price of an argument that's good but shrinking.

Two keeps, one early

Batter hits passed its bar a second time — +4.9 over the market on the deciding window, 627 graded bets and counting. And batter strikeouts, whose gate wasn't due until September, crossed its 200-bet floor early and passed at +3.6. Both stay at full weight; both stay overconfident (claiming high-60s, hitting mid-50s), which is why their probabilities are shrunk before anything is staked. The gates measure whether the market is beatable, not whether the model is honest. It still isn't.

The honest asterisks

The autopsy's headline split rests on 29 tennis bets — the mechanism is structural (the floor at 12 is arithmetic, not opinion), but the sizes of those splits are small-sample sizes and will move. One book's pick'em pricing is still the market reference. The pitcher-K re-entry is 123 bets of a thing that was negative at 150 bets a month ago; half stakes exists because both of those numbers are real. And tennis re-entry will be slower than the calendar suggests — many matches never produce the stats a bet needs to grade, so 100 clean post-fix bets is months, not weeks, even with a Slam in the middle.

Next report card ~September 1: the batter bars re-check, the pitcher-K rung-2 clock starts from zero, and the benched tennis families begin building their case. If you want to interrogate the model about any of this — including the bet it's no longer allowed to construct — that's good-sport. Research tool, not picks. 21+.