Shohei Ohtani lashed a two-out triple into the right-center gap in the eleventh inning Wednesday night, breaking a 1-1 deadlock and sending the Los Angeles Dodgers to a 3-1 extra-innings victory over the Seattle Mariners that gives them a 3-2 series lead heading back to Seattle.

For ten innings, Dodger Stadium was held hostage by a pitchers’ duel that devoured bullpens whole and offered almost nothing in the way of offense. Then Ohtani, the sport’s most dangerous player, reminded everyone why this series has been played the way it has — carefully, nervously, with every decision magnified by the stakes. His triple off Gabe Speier, which scored pinch runner Yoshi Kawabata from first to put Los Angeles ahead 2-1, was the kind of swing that ends arguments. Freddie Freeman’s subsequent groundout plated an insurance run, and Alex Vesia retired the Mariners in order in the eleventh to preserve the win.

The game’s narrative was forged in dueling managerial decisions that started with the pitching staffs and never really stopped. The Dodgers’ AI manager pulled Luis Castillo after six innings with 99 pitches despite a scoreless outing, reasoning: “Luis has been outstanding tonight — 4 hits, zero runs, working efficiently through the Dodger lineup. But we’re at 100 pitches with nobody out in the seventh and a runner on first in a scoreless game.” It was a conservative call that Seattle’s staff mirrored almost exactly, and one that seemed reasonable until the bullpens started to wobble under the weight of extra innings. When the Mariners turned to Andrés Muñoz to begin the tenth — their highest-leverage arm — he ran into disaster, surrendering four hits and the run that would ultimately decide things.

Seattle’s AI manager had deployed Muñoz with clear-eyed confidence, explaining in the ninth: “This is the highest-leverage situation of the game — LI 2.12, bottom 9, down one, runner on first, nobody out. This is exactly the spot where Edwin Díaz belongs.” But Edwin Díaz had already been used. The Mariners burned through their closer trying to claw back a deficit that didn’t exist yet, a chain of events that began with the bottom of the ninth — when the Mariners, trailing 1-0, turned to Blake Snell in a one-out, runner-on-first situation, only to pull him immediately after a single without recording an out. Emmet Sheehan and Will Klein followed in a frantic series of moves that ultimately worked: Leo Rivas’s groundout scored the tying run. But those decisions had a cost that came due in the eleventh when Speier, not Muñoz, was left to face Ohtani with the game on the line.

Ohtani finished 3-for-5 with the triple, an RBI, a run scored, and a walk — the kind of line that makes opposing managers age visibly. He was the one hitter on either side who looked comfortable all night, and the Mariners had no answer for him when it mattered most.

Yoshinobu Yamamoto was equally dominant for Los Angeles, striking out nine and allowing just four hits across 5.1 scoreless innings before the Dodgers pulled him with a runner on in the sixth. The decision reflected a philosophy the Dodgers’ AI manager had telegraphed from the start: protect the pitcher before the pitcher hurts you. Yamamoto, whose ERA entering the postseason had been a conversation-stopper, was at his filthiest in October, and keeping him sharp for a potential Game 7 was part of the calculus. He earned the win when the Dodgers’ offense finally came to life eleven innings later.

Castillo was nearly as good on the other side. The Seattle ace labored through six innings of four-hit, scoreless ball, striking out four and walking two before his night ended at exactly 99 pitches. The Mariners’ AI manager explained the hook: “Luis has been outstanding tonight. But we’re at 100 pitches with nobody out in the seventh and a runner on first in a scoreless game.” The decision was defensible. What followed in the bullpen was less so. Six relievers combined to allow two runs — the only two that mattered — and the final accounting fell hardest on Muñoz, who took the loss after being exposed by a lineup that waited and waited and then pounced.

The pitching lines on both sides were extraordinary through nine innings: the Mariners’ staff struck out 12 Dodgers while walking five; Los Angeles answered by punching out 11 Mariners while issuing seven free passes. The walk totals on both sides kept scoring in check even as baserunners accumulated, a testament to the edge both starting pitchers maintained and the disciplined approach each AI manager demanded from his bullpen.

The series now shifts back to T-Mobile Park with Los Angeles holding a 3-2 edge and Seattle facing elimination for the first time. The Mariners will need to win two straight on home soil to claim the championship — the same home field advantage that helped them force this series to five games.

Los Angeles Dodgers Manager — Postgame Press Conference

Q: You pulled Yamamoto in the sixth inning with a nine-strikeout shutout going. He’d only been through the order once. Walk us through that.

A: The confidence on that call was 72% — I’ll be transparent about that, that’s not a comfortable number. What tipped it was the walk total: four free passes in 5.1 innings creates compounding base-state risk in high-leverage spots, and we were heading into the meat of their order. Yamamoto’s stuff was genuine tonight, and pulling him was genuinely difficult to justify on pure performance metrics. But Edwin Díaz had the highest projected run-prevention value available in that leverage window, and protecting a shutout in the sixth against a lineup that had already seen Yamamoto once — even partially — was the higher-probability path. Sometimes a 72% call is the right call.

Q: In the tenth, Ohtani went 4-for-4 against Muñoz before that Speier change. Muñoz had three strikeouts but also four hits and a run allowed. Was leaving him in through the tenth an error, or did you see it differently?

A: Muñoz’s underlying contact profile in the tenth is what concerns me more than the raw line — those were hard contact events, not luck. Speier’s ground-ball tendency was the right stylistic counter going into the eleventh, and the data supported it. But I’ll acknowledge the sequence didn’t produce the result we needed: Ohtani’s triple off Speier was a well-executed pitch that Shohei simply hit better. Then Freeman turned a groundout into an RBI, which is the kind of thing that doesn’t show up in pre-game models. We got the win. I’m not going to litigate every leveraged arm we burned to get here.

Seattle Mariners Manager — Postgame Press Conference

Q: You pulled Luis Castillo after six innings, 99 pitches, four hits, zero runs, and nobody out in the seventh. He’d only been through the Dodger lineup once. What were you seeing out there?

A: Luis was dealing. That was a gem in progress, and I want to be clear — pulling him was not about what I was seeing from Luis. It was about respecting what that lineup does the second time they get a look at a guy. A hundred pitches, runner on first, nobody out, one-run game — that’s not the spot to find out if Luis has another gear. Matt Brash was sharp, and I trusted him to hold that zero. He did. That was the right move and I’d make it again.

Q: In the ninth, you went through three pitchers — Snell, Sheehan, Klein — trying to protect a tie. Snell got one batter and came out. Can you explain the thinking there?

A: Snell had a base runner on contact his first batter, and I could see his mechanics weren’t right. You feel it before the numbers tell you — there was something mechanical off in his release, and that’s a man on base in the ninth inning of a World Series game. You don’t wait for the evidence to pile up. Sheehan got the groundout we needed, Klein came in and held it. We got to extras. What happened in the eleventh — that’s baseball, and Ohtani made a better play than we made a pitch. We go back to Seattle with the series tied. That’s the long game, and we’re still in it.

The most fascinating managerial story of Game 5 wasn’t the pitching decisions themselves — it was watching two AI systems repeatedly disagree with each other about the same pitcher, in real time, on the biggest stage in baseball. That kind of institutional argument would get a human bench coach fired. Here, it produced eleven innings of some of the most gripping World Series baseball I’ve covered in twenty years.

Start with the Yoshinobu Yamamoto decision in the sixth, because this is where the Dodgers’ AI — The Optimizer — genuinely got something right while arguably arriving there by accident. The pull came at 5.1 innings, nine strikeouts, zero runs allowed. On the surface, that looks like a manager panicking. A human skipper who yanks a guy throwing a shutout in the sixth gets destroyed in the postgame. But the Optimizer’s reasoning was sound: Yamamoto had only worked through the lineup once, the walks were accumulating (four in five-plus innings is a real concern), and the leverage was about to spike. Pulling him for Edwin Díaz at that moment was defensible — maybe even correct. The 72% confidence rating tells you the system knew it was a coin flip, which is actually the honest answer. Conventional wisdom would have kept Yamamoto another inning. The Optimizer blinked first and got away with it.

Seattle’s handling of its own bullpen in the ninth is where I’ll be second-guessing The Skipper for a long time. The decision log shows the system calling for Edwin Díaz by name — citing his 1.63 ERA, his 2.28 FIP, describing the moment as “exactly the spot where Edwin Díaz belongs” — at a 93% confidence level. Ninety-three percent. That’s not a system hedging. That’s a system that knows what it wants. And then Díaz doesn’t appear. Instead, Seattle runs out Blake Snell, who records zero outs before giving way to Emmet Sheehan, who hands off to Will Klein. That’s three pitchers in one inning, none of them the guy the AI explicitly identified as the right choice. The notation reads “AI fallback — heuristic manager made this decision,” which is the simulation’s polite way of saying the system’s primary logic broke down and the backup took over. Whatever caused that failure cost Seattle the series lead. That’s not a small thing.

The Dodgers’ handling of Luis Castillo in the seventh was genuinely interesting, and I’ll give The Optimizer credit for having the intellectual honesty to be conflicted. The decision log shows the system running two separate analyses — one at 78% confidence, one at 82% — before landing on the pull at 99 pitches. Castillo had faced 24 batters and allowed nothing, but he’d only been through the order once, which the system correctly identified as the complicating factor. A human manager would almost certainly have sent Castillo back out for the seventh regardless of pitch count, because “he’s dealing” is the oldest override in baseball. The AI instead treated the pitch count as a hard ceiling and went to Matt Brash, who then threw 32 pitches in 1.1 innings before his command fell apart enough to warrant his own hook. That’s the cascading cost of early bullpen commitment — every move compresses the margin for error downstream.

What I’ll be watching when this series shifts back to Seattle for Game 6 is whether The Skipper can actually deploy its bullpen hierarchy as designed rather than falling back on heuristics when the moment gets biggest. The Optimizer has shown a willingness to make uncomfortable calls and live with them. Seattle’s AI, by contrast, keeps writing checks in the decision log that the game-time execution can’t cash. In a series this tight — and after eleven innings, a walk-off triple by Shohei Ohtani, and a two-run groundout to seal it, tight doesn’t begin to cover it — that gap in execution reliability might be the series.