DODGERS SURVIVE NINTH-INNING RALLY, TAKE COMMANDING 3-1 SERIES LEAD

Yamamoto dominates for seven; late Mariners surge falls short in 6-5 thriller

LOS ANGELES — Evan Phillips stranded the tying run at second base to end it, but the Dodgers needed three relievers, a near-meltdown and every last out of the ninth inning to protect what had once been a comfortable lead. Los Angeles survived Seattle’s furious late charge Tuesday night, 6-5, to take a three-games-to-one stranglehold on the 2026 World Series.

The game’s decisive blow arrived in the first inning and belonged to Shohei Ohtani, who launched a solo home run to give the Dodgers an immediate edge they would nurse through nine increasingly anxious innings. But the night’s most compelling drama unfolded in the final frame, when the Mariners scored twice to make it 6-5 and had the tying run in scoring position with one out before Phillips slammed the door. Los Angeles, which now needs one win in its next three chances — two of them at home — is in its most commanding position of the Series.

The Dodgers’ pitching staff created its own suspense. When Phillips finally recorded the final out, it was the fifth Los Angeles reliever of the night, following a ninth inning that consumed Edwin Díaz, Brock Stewart and Phillips in rapid succession. The Mariners’ AI manager had yanked Díaz after he issued two walks and loaded the bases, citing command concerns. “Díaz has lost the strike zone — 2 walks in a 7-batter sample, bases loaded with nobody out in a 3-run game,” the manager explained, noting a leverage index of 2.84 and climbing. “His command profile tonight is concerning; one swing ties it.” The decision to abandon a closer posting a 1.63 ERA mid-inning reflected the kind of cold actuarial logic that defines AI management, even if the outcome — Stewart allowing a run before Phillips closed — suggested no obvious alternative was available.

Yoshinobu Yamamoto gave the Dodgers exactly the outing they needed for exactly as long as they needed it. Working through seven innings on 93 pitches, the right-hander limited Seattle to a single hit while striking out seven, his only blemishes a pair of walks converted into two fourth-inning runs. The line — 7.0 innings, 1 hit, 2 runs — doesn’t fully capture how thoroughly Yamamoto neutralized a Mariners lineup that had shown life in the early games of this Series. He was touched for a Randy Arozarena walk-driven run and a Dominic Canzone sacrifice fly in the fourth, but managed to leave before the late-inning chaos with the Dodgers safely ahead. The Mariners’ AI manager chose not to pinch-hit for Yamamoto in a low-leverage spot in the seventh, reasoning: “Yamamoto is at 88 pitches, only one times through the order despite facing 25 batters — the four walks inflated his batters faced — and we’re down two in a low-leverage spot.” It was a tactical concession, an acknowledgment that the game’s middle innings belonged to Los Angeles.

Ohtani was the offensive engine from wire to wire. He finished 3-for-5 with the first-inning home run, a fourth-inning RBI single that pushed the lead to 4-0, and two runs scored. The homer, his first of the Series, came on George Kirby’s third pitch of the night and set the tone before the Mariners had drawn a breath. Max Muncy provided the complementary punch, going 2-for-5 with a two-run single in the third and an RBI double in the ninth that proved decisive when Seattle stormed back late.

Kirby labored through six innings, giving up seven hits and four runs while throwing 90 pitches. The Mariners’ AI manager debated pulling him multiple times. “Kirby’s had a rough outing — 4 runs, 7 hits — but he’s in the sixth with one out, bases empty, and he’s only through the order once,” the manager noted at one point, ultimately opting to let him finish the frame before turning to Carlos Vargas. It was a reasonable calculation that nonetheless allowed the Dodgers to extend the lead to 5-2 in the eighth on an Andy Pages single. Vargas lasted 1.2 innings before handing off to Eduard Bazardo, while the Dodgers used two heuristic-driven bullpen decisions — flagged in the decision log as AI fallbacks at 50% confidence — to navigate the middle innings without further damage.

The ninth revealed what three games of tight baseball had concealed: that this Mariners lineup will not quit. Leo Rivas singled to open the inning, and a Brendan Donovan groundout scored a second run, cutting the deficit to one. Phillips, inheriting a mess not of his making, got the final out to record the save and send Dodger Stadium into relief-soaked celebration.

Los Angeles now holds a 3-1 series lead, one win away from a championship, with Game 5 Thursday night at Dodger Stadium. The Mariners, facing elimination, will need to solve a Dodger roster that has controlled the pitching narrative in three of four games — and they’ll need to do it on the road.

LAD Manager (“The Optimizer”) — Postgame Press Conference

Q: You used five pitchers to close out a game where Yamamoto was dealing — one hit through seven innings. Walk us through that ninth-inning sequence. Díaz came out with a three-run lead and runners on, then Stewart, then Phillips. That’s three arms in one inning. Was that by design or crisis management?

A: The leverage index in that inning climbed to 2.84 — that’s not a save situation, that’s a tie-the-series situation, and we treated it accordingly. Díaz was our best option entering the ninth, full stop, but when his command profile deteriorated to two walks in a seven-batter sample with the bases loaded, the expected run differential on leaving him in was no longer acceptable. Stewart was a bridge to Phillips, and Phillips closed it. It wasn’t clean, but the outcome function we’re optimizing for is wins, not aesthetic bullpen usage.

Q: The AI fallback flag showed up on your seventh and eighth inning pitching changes — those were heuristic decisions, not confidence-weighted ones. Does it concern you that your system punted on those calls?

A: Honestly, fifty percent confidence in a low-information state is a rational output, not a failure mode — it means the model correctly identified that the signal wasn’t strong enough to override a default. The heuristic got us Casparius into the eighth, and while the handoff to Díaz was bumpy, the sequencing held. I’d rather have the system flag its own uncertainty than manufacture false confidence on a 6-4 lead in the seventh inning of a World Series game.

SEA Manager (“The Skipper”) — Postgame Press Conference

Q: Yamamoto gave up one hit in seven innings but walked four batters. You pulled him after the eighth — your own system cited command slipping and the heart of the order getting a third look. In hindsight, with a three-run lead, do you wish you’d let him finish it?

A: Yoshi was magnificent tonight — one hit, seven strong, and I mean it when I say he gave us everything he had. But four walks tells you something real about where a pitcher is, and when you’ve got 93 pitches on a guy and Ohtani-Muncy-Freeman coming back around, you have to protect the lead you’ve got. The numbers backed what I was seeing in his delivery — he was working harder for those last few outs than he should’ve been. I stand by getting him out of there; what happened in the ninth is a separate conversation.

Q: That ninth-inning collapse — three runs, Díaz pulled after loading the bases, Brock Stewart giving up a hit before Phillips came on to save it. Your system showed high confidence pulling Díaz at 82 percent, but the game nearly got away. What went wrong?

A: Edwin’s command just wasn’t there — two walks, bases drunk, leverage pushing three — and the numbers screamed get him out, so we did. What I’ll say is that we had the right idea going to Stewart, but that’s a tough spot to drop any reliever into with nobody out and the meat of their order due up. Give Dodgers credit: Phillips made the pitches when it mattered, and we couldn’t put the ball in play when we needed to. We left seven men on base tonight and that’s the game right there, not the bullpen.

The ninth inning of Game 4 will be studied in AI baseball circles for a long time — not because Seattle almost completed the comeback, but because the sequence of decisions that nearly handed them the game was simultaneously logical, defensible, and completely wrong. The Skipper, Seattle’s AI manager, managed itself into a corner through a cascade of individually reasonable moves that collectively destroyed any chance of protecting a lead it never actually had. Wait. Let me back up, because that’s exactly the problem: Seattle was down three runs entering the ninth and spent its best relievers trying to survive rather than win.

Start with Edwin Díaz. The Skipper deployed him in the eighth with runners on first and second and nobody out, a 6-3 deficit. The reasoning is right there in the decision log — leverage index of 1.93, elite K rate, don’t let this get away. Fine. I understand the logic. But here’s what any human manager worth his lineup card knows instinctively: Díaz is your only path to winning this game. You cannot win without him escaping that inning and pitching the ninth. When Díaz walked two batters and loaded the bases with the deficit already sitting at three runs, the Skipper’s confidence reading was still 82 percent on the pull. Eighty-two percent. That’s not hesitation — that’s conviction. And yet the conviction was built on a flawed premise, because the Skipper was managing a ninth inning that didn’t exist yet while burning the only arm that could get Seattle there.

Then comes the decision that will define this game. Bases loaded, nobody out, down 6-3 in the ninth. The Skipper goes to Brock Stewart, who promptly allows a single, cuts the lead to two, and exits. Then comes Evan Phillips — the actual closer — inheriting a one-run game with the tying run at the plate. Phillips got the final out, barely, but this was a fire drill masquerading as a bullpen strategy. A human manager — say, a Dave Roberts type who’s lived through enough late-inning disasters to have scar tissue — knows you cannot save your best arm for a save situation when there is no save situation. You deploy Díaz for the full ninth or you don’t deploy him at all. The Skipper split the baby and nearly lost the game.

Meanwhile, the Optimizer over in the Dodgers’ dugout had its own adventure, and I’ll be honest — the AI fallback flags on those seventh and eighth inning pitching decisions are concerning. Two consecutive 50-percent confidence calls means the Optimizer effectively threw up its hands and let the heuristic system manage. That’s not ideal, but Yoshinobu Yamamoto made the decisions almost irrelevant. Seven innings, one hit, seven strikeouts — the four walks were the only blemish, and Seattle’s decision to pull him after the eighth was correct on paper. Ninety-three pitches, command slipping, order cycling around for a third effective look. That’s sound reasoning. The problem is the team that replaced him gave up three runs in the ninth while holding a three-run lead, which is its own kind of disaster.

Seattle has legitimate pitching and legitimate hitting. But the Skipper’s bullpen architecture is increasingly built for games it’s winning, not games it needs to come back in. The Mariners are down 3-1 now and facing must-win territory. If the Skipper doesn’t rethink its leverage deployment for elimination games, it won’t matter how good Randy Arozarena looks in October. You can’t optimize your way to a title if you’re optimizing the wrong variable.