You'll remember Caminero's home run — but the play that turned the Athletics' 96% game was a soft single two batters later

You saw Junior Caminero's three-run homer cut the Athletics' lead to one, and that is the play you will remember. We read both win-probability feeds end to end. In MLB's, the play that turned the game is not the homer but the next batter but one: Jonny DeLuca's single moved the Athletics' odds 19.6 points, more than any other play of the night, and marked the exact moment they fell below 50% for the first time — the homer ranks third ^mmodeled. The two feeds agree on the shape of the collapse but weight individual plays differently, so that leverage ranking is MLB's. Here is the record behind that. On 10 August 2026 the Tampa Bay Rays beat the Athletics 10-6 at the Athletics' home park (MLB gamePk 824969; ESPN event 401816477). For seven innings the Athletics led: they scored twice in the first, once in the fifth, and three times in the seventh to go up 6-2. Entering the top of the eighth they were six outs from a win, and two in-game model feeds put them there near certainty. The Athletics' win probability peaked at 96.9% (MLB) and 97.0% (ESPN) late in the bottom of the seventh, after Jacob Wilson's two-out double, while the Athletics were still batting; the seventh's final out then left them at 96.4% (MLB) and 96.5% (ESPN) with six outs to go ^mmodeled. Yandy Díaz opened the eighth with a single, and the figures fell to 93.8% (MLB) and 93.4% (ESPN) ^mmodeled. Then Tampa Bay batted around. The first eight Rays hitters of the inning reached base — five singles, a three-run homer, and two walks, no outs — and the lead changed hands before the Athletics recorded a single out. Caminero's three-run home run, his 34th, cut it to 6-5; Chandler Simpson's single tied it 6-6; Taylor Walls's two-run single made it 8-6, all with the bases still uncleared by an out. Tampa Bay scored eight runs in the frame on seven hits (six singles and Caminero's homer) and two walks, adding a sacrifice fly and another RBI single after the first out. The Athletics did not score again. Tampa Bay out-hit them 12 to 10 for the game, but 7 of those 12 hits came in the eighth. Final: Rays 10, Athletics 6, one error each.

The following is analysis, not fact. Read the box score the way a standings page does and this is an ordinary result: a road team wins by four, out-hits the home team 12-10, moves on. Read the win-probability series the way only a whole read allows — end to end, both feeds — and the same game is a near-decided one that reversed before the Athletics recorded a single out ^mmodeled. The two descriptions are of one event. What reconciles them is that the entire margin — all eight Tampa Bay runs and the lead change — is compressed into a single half-inning, and the swing is steep: from 96.4% (MLB) with six outs to go to a loss, a drop of roughly 96 points, having peaked near 97% late in the seventh ^mmodeled. A final score cannot carry the shape inside the inning. The Athletics did not surrender the lead gradually and then fail to recover it; they recorded no out at all until they were already behind. Eight consecutive batters reached, six runs crossed, and an 8-6 deficit existed before the first out was made. The model's per-at-bat drop was not even across those eight batters. By MLB's feed the Athletics' figure fell 96.4 → 93.8 → 89.0 → 74.5 → 65.7 → 46.1 → 33.6 → 21.9 → 7.2 — steps of about 2.6, 4.8, 14.5, 8.8, 19.6, 12.5, 11.7, and 14.7 points ^mmodeled. This is why the play a human remembers is not the play the ledger records as the turn. The largest single step was DeLuca's single that put the go-ahead run in scoring position (19.6 points), followed by Walls's two-run single (14.7) and Caminero's three-run homer (14.5) — not the leadoff hit, and not the homer; the biggest moves fell in the middle and back of the sequence, not at its front ^mmodeled. Note also the texture of the damage: of the seven hits, six were singles. This was not a barrage of extra-base contact but a line that would not stop — the kind of inning that a run-differential summary or an out-of-town score flattens into the word "eight."

A 96% figure is a model estimate, not a verdict. It states the model's implied chance of a home win at that instant; a leading team carrying such a figure is expected to lose some fraction of the time, and a single game that ends in a loss is consistent with the estimate rather than proof the model failed ^mmodeled. We do not have a calibration study for either feed, so we do not attach an observed loss rate to the printed number. One half-inning is also the smallest meaningful sample in baseball; a bloop that is caught, one more strike to any of the first eight hitters, and the collapse never compounds ^sspeculative. We also do not have the internal methodology of either feed — the inputs each model uses, and how it is fit — so we do not claim what context it does or does not price in; we read its published output and no more ^mmodeled. Nor should the concentration of hits be read as a quality judgment: the free feeds we pulled carry the play descriptions and the model outputs but not the batted-ball or expected-value data, so we cannot say whether Tampa Bay's six singles were struck hard or found grass, nor whether the Athletics' relief was unlucky or simply hittable ^sspeculative. What is not in dispute is the closed record: a 6-2 lead entering the eighth, eight Rays runs in the frame, and the first out of the inning recorded only after the Athletics were already losing — a sequence over which both feeds ran from a 96.9%-97.0% peak to a home loss ^mmodeled.

The 96.9% figure was published by MLB as a per-at-bat estimate of the game state; we did not compute it ^mmodeled. What we did was read the full series after the fact, whole, as a single object — and from that vantage the game looks the way we are describing it, which no single in-moment reading supplies. A spectator in the park on the seventh-inning stretch saw a comfortable four-run lead, not a 96.9%; the figure existed on a scoreboard feed, but the shape of the whole descent is visible only laid end to end ^mmodeled. We were also never in the dugout or the bullpen; we can read that eight Rays reached before an out, but not what any pitching decision was meant to accomplish, and we do not guess. The vantage is the finding: the same eight runs that a box score files under "eighth inning" a whole-series read files under "the game."