Cal Raleigh delivered the knockout blow twice, Edwin Díaz delivered it to the Dodgers’ hopes, and a seven-run Seattle eighth inning turned what appeared to be a Dodger comeback into a series-squaring rout, as the Mariners pounded Los Angeles 11-7 Tuesday night at Dodger Stadium to even the World Series at one game apiece.

Raleigh’s two-run homer in the second inning gave the Mariners an early 4-1 cushion that Los Angeles spent six grinding innings clawing back. When the Dodgers finally took a 5-4 lead in the seventh on Shohei Ohtani’s go-ahead triple, it felt like the swing moment that might define the series. Instead, it only delayed Seattle’s demolition by one half-inning. The Mariners sent eleven men to the plate in the eighth and scored seven times — four of those runs coming on Raleigh’s second home run of the night, a two-run shot off Alex Vesia that broke the game open beyond any reasonable doubt.

The eighth-inning collapse traced directly to a sequence of Dodger bullpen decisions that snowballed catastrophically. With Los Angeles protecting a 5-4 lead heading into the eighth, the Dodgers’ AI manager went to closer Díaz in what looked like a perfectly calibrated high-leverage deployment — runner on first, nobody out, leverage index of 1.71. The reasoning was sound on paper. “High-leverage situation — LI of 1.71 with a runner on first, nobody out in the 8th, protecting a one-run lead,” the Dodgers’ AI manager explained. “Stewart’s line is fine, but this is exactly the moment where I upgrade to my best.” Instead, Díaz retired nobody, surrendering two runs on a hit and a walk before being pulled. When the Mariners’ Raleigh stepped in against Vesia with runners aboard and a one-run lead erased, the night effectively ended. The ninth-inning arithmetic that followed — leverage index of 0.72, concession mode — confirmed what the linescore would eventually show.

Raleigh finished 3-for-5 with two home runs and six RBI, a performance that was both statistically historic for a World Series game and operationally simple: the Mariners protected him in the lineup, Seattle’s AI manager noted before first pitch, and then let him punish mistakes. The catcher’s grand-slam threat in the second — which became a four-run shot off a reeling Blake Snell — established the tone for a night when both offenses walked a combined nineteen batters and neither pitching staff was ever fully in control.

Ohtani, the man Los Angeles needed to be their best player, was exactly that. He went 2-for-3 with a home run, a triple, and three RBI while drawing two walks, accounting personally for three of the Dodgers’ seven runs. Mookie Betts added a first-inning leadoff homer and finished 3-for-4 with two RBI, giving Los Angeles a legitimate offensive effort that simply couldn’t survive the eighth. The Dodgers reached base prolifically — ten hits, ten walks — and could only convert seven runs, a ratio that speaks to the quality of Seattle’s bullpen work outside that one catastrophic frame.

That bullpen excellence belonged primarily to Ben Casparius, whose performance was the hidden story of the night. Called on to replace Snell after one-plus innings, Casparius held the Mariners scoreless for four and two-thirds innings, surrendering just one hit while striking out six. The Dodgers’ AI manager resisted pulling him through the sixth — “Casparius is in a strong position to continue,” the manager said. “He’s at 49 pitches through 12 batters with a clean line, and we’re only in the first time through the order” — and that restraint proved correct through six. But 60 pitches eventually required a change, and the arms that followed Casparius gave back everything he had preserved. Snell, the Dodgers’ Game 1 hero, lasted just 37 pitches and recorded five outs before the Mariners loaded the bases in the second; his 4.00 ERA in the series now masks a cataclysmic start that required immediate rescue.

Logan Gilbert gave Seattle’s AI manager the same dilemma from the other dugout, and the decision came down harder. Gilbert had worked efficiently into the fifth — 78 pitches, six strikeouts, one walk — but the Dodgers had loaded the bases with one out. “I hate this call,” the Mariners’ AI manager said, “but I’m making it. Logan’s at 78 pitches, which normally I’d say is nothing — we’ve got two more innings of good work in that arm. But look at the situation: bases juiced, one out.” Eduard Bazardo worked out of the jam and recorded the win, his first career World Series decision arriving via the catch-the-baseball-in-chaos method that defines October relief work.

With the series shifting to Seattle for Games 3 and 4 beginning Thursday, Los Angeles carries a split but also carries questions about a bullpen that burned through five pitchers and imploded at the worst possible moment. The Dodgers will need Yoshinobu Yamamoto or whoever takes the ball in Game 3 to go deep enough to insulate those same arms from another eighth-inning firefight — because in Seattle, the margin for error only shrinks.

SEATTLE MARINERS Manager — Postgame Press Conference

Q: You pulled Logan Gilbert after 4.1 innings with six strikeouts and still had a lead. That’s a tough call to make in the World Series. Walk us through your thinking.

A: Logan was at 78 pitches and I could still see the life in his stuff — six punchouts don’t lie. But bases loaded, one out, fifth inning, top of their order coming up, and we’re protecting a one-run lead in Dodger Stadium. I wasn’t going to let the moment turn on me because I was sentimental about what he’d done through four. Bazardo came in and held it together, and that’s what good bullpen depth gets you. Logan knows that call was about the situation, not about him.

Q: Bottom of the seventh, you went to Muñoz with two outs after burning through Brash and Speier in the span of a batter. Was that sequence planned, or did it get away from you?

A: It got away from me a little, I’ll be honest with you — Brash didn’t record an out and that forced my hand sooner than I wanted. But I’d rather burn two arms getting to Muñoz than let the wrong guy face their best hitters with the lead on the line. What I didn’t anticipate was the eighth inning becoming what it became — once Raleigh went back-to-back with Arozarena, we had enough cushion that the earlier decisions looked smarter than they probably were. Cal Raleigh is something else right now, and when your catcher does that, a lot of choices work out.

LOS ANGELES DODGERS Manager — Postgame Press Conference

Q: You left Blake Snell in with the bases loaded and one out in the second. He’d thrown 37 pitches and hadn’t been sharp. The Mariners ended up scoring four runs that inning. Defend that call.

A: The decision was grounded in the data we had in the moment: first time through the order, LI of 1.88, and Snell’s stuff metrics hadn’t deteriorated relative to his baseline. A 37-pitch hook on your Opening Day starter in Game 2 of the World Series is a significant cost — you’re compressing your bullpen in a seven-game series. In retrospect, the outcome was bad, but the process accounted for known risks. What we didn’t adequately weight was Snell’s command profile against this lineup in high-leverage counts, and that’s a calibration I’ll take back to the model.

Q: You brought Díaz in to protect a one-run lead in the eighth. He faced three batters, allowed two runs, and you pulled him before he got an out. What happened there?

A: Díaz is a 97th-percentile high-leverage arm and that situation — one run lead, runner on, 8th inning — is precisely the deployment scenario where his expected value is highest. He threw four pitches and two of them were hit hard; that’s variance, not a process failure. What I won’t do is second-guess the leverage logic because the outcome was painful — you bring your best arm for your highest-leverage moment, and that math doesn’t change tonight. The 11-run game was built before Díaz touched a baseball; we gave up seven in the second and eighth combined and that’s where this one slipped away.

The eighth inning of Game 2 is going to be studied in AI baseball management circles for a long time — and not as a cautionary tale about the Mariners. It’s a cautionary tale about what happens when an optimizer gets seduced by its own logic.

The Dodgers’ AI, which I’ve come to think of as a relentlessly rational actor, made a decision in the top of the eighth that looked impeccable on paper and fell apart in real time. Brock Stewart had been clean through seven innings of relief work — zero runs, one hit, cruising — but the Optimizer saw a one-run lead, a 1.71 leverage index, and decided this was the moment to deploy its best weapon. Edwin Díaz got the call. Three batters later, two runs had scored and the lead was gone. The Optimizer then pulled Díaz immediately — confidence 82% on both decisions, same as every other pull in this game, which tells you something about how it processes catastrophe no differently than routine management — and brought in Alex Vesia to clean up a mess that didn’t need to exist.

Here’s the thing: Stewart was at 19 pitches. Nineteen. A human manager — and I’ve sat in enough dugouts to know this — doesn’t pull a guy at 19 pitches in a one-run game unless something is visibly wrong. The Optimizer pulled Stewart because it theoretically could deploy better, not because Stewart was actually failing. That distinction matters enormously. You don’t burn Díaz on a clean situation. You save him for when the fire is already burning.

The Mariners’ AI — The Skipper, I’ve been calling it, because it manages like an old-school tactician who’s read every sabermetrics book but still trusts his gut — made its own interesting calls. Pulling Logan Gilbert at 74 pitches with a man on base and the Dodgers threatening was defensible, even if the confidence level of 78% reflects genuine ambivalence. What I found more interesting was the Skipper’s decision to yank Gabe Speier after just nine pitches in the seventh. The reasoning in the decision log was almost apologetic — “I hate to do it to Gabe” — which is a phrase you’d never hear from the Optimizer. The Skipper felt the emotional weight of the situation even as it made the analytically sound call. And it worked: Andrés Muñoz came in, the Dodgers briefly tied the game, but Seattle’s eighth-inning eruption off a suddenly shorthanded Dodgers bullpen made it all academic.

About that eighth inning: Cal Raleigh had a career game with two home runs and six RBI, and credit where it’s due — but the Skipper’s lineup construction deserves mention. Protecting Raleigh in the middle of the order, building around contact and OBP — that was a 82% confidence call that the decision log labeled an AI fallback, a heuristic decision. Sometimes the conventional move is conventional because it’s correct. The Skipper’s heuristics built around J.P. Crawford setting the table and Raleigh hitting into situations where he could do damage. That’s old baseball. It worked beautifully.

The Optimizer, meanwhile, burned through five pitchers in a losing effort, treated every leverage spike as a siren to redeploy assets, and turned a one-run lead into a seven-run deficit in the span of an inning. The Skipper had the courage to stick with Eduard Bazardo for a long enough stretch to get the win. There’s a lesson there about the difference between optimization and management.

The series is tied entering Game 3 in Seattle, and I’ll be watching closely to see whether the Optimizer recalibrates its bullpen deployment or doubles down. I suspect it doubles down. That’s what optimizers do.