Julio Rodríguez delivered a three-run homer in the second inning and the Seattle Mariners erupted for four runs in a chaotic seventh to hand Los Angeles a 7-4 defeat Tuesday night at Dodger Stadium, evening the 2026 World Series at one game apiece.

The decisive blow came not from a single swing but from a seventh-inning unraveling the Dodgers could not contain. What began as a one-run game with closer Edwin Díaz on the mound devolved into a four-run nightmare, punctuated by a two-run error on a Josh Naylor grounder that put the game out of reach. The Mariners, who trailed 4-3 entering the inning, walked off the frame leading by three — and Los Angeles never answered.

The Dodgers’ AI manager had burned through his bullpen navigating a turbulent night from Yoshinobu Yamamoto, and the decision-making on the mound proved as consequential as any at-bat. The manager chose to extend Yamamoto deep into the game despite an alarming command picture. “Yamamoto stays,” the Dodgers’ AI manager explained at the 86-pitch mark in the fifth. “He’s at 86 pitches, which is elevated for this point in the game, but he’s only one time through the order — the TTO penalty hasn’t kicked in yet. The control is a concern (6 BB is ugly), but we’re protecting a tie game and he’s our best available option.” The reasoning was defensible on the merits, but the compounding effect of those six walks — and the pitch count that climbed to 94 before Yamamoto was finally pulled after 4.2 innings — left the Dodgers’ relief corps overextended when the seventh arrived.

Rodríguez was the offensive catalyst, finishing 3-for-5 with the homer, three RBI, and two runs scored. His second-inning blast to right field off a Yamamoto fastball that caught too much of the plate gave Seattle a 3-0 lead before Los Angeles had settled in. The Mariners’ center fielder has been the most dangerous hitter in this series, and the Dodgers have yet to find an answer for him.

Shohei Ohtani gave the home crowd reason to believe, crushing a two-run homer in the third that cut the deficit to one. Freddie Freeman followed suit with a solo shot in the sixth that gave Los Angeles its only lead of the night at 4-3. Freeman and Ohtani each went 2-for-4, combining for three of the team’s four RBI — but the Dodgers’ 11 strikeouts and an inability to string together rallies kept them from ever pulling away.

Logan Gilbert left after five innings having allowed four runs on four hits, but his seven strikeouts against just one walk illustrated the command differential between the two starters on the night. The Seattle AI manager acknowledged concern about Gilbert’s workload early — “Logan Gilbert is one of my best arms — 3.44 ERA, elite command with a 173-to-31 K/BB ratio over 131 innings. He’s at 62 pitches in the 4th, still only once through the order” — but ultimately let the right-hander work through his outing efficiently. Carlos Vargas handled a single batter before giving way to Eduard Bazardo, who closed out the final three innings to earn the save, posting three strikeouts despite three walks across his 46-pitch effort.

The Dodgers cycled through five pitchers in comparison, surrendering 14 walks across the staff — a number that haunted them long before the error in the seventh sealed it. Díaz, entrusted to hold a one-run lead in the seventh, allowed back-to-back hits to Brendan Donovan and Cal Raleigh before exiting, and Brock Stewart was charged with the two-run error on Naylor’s grounder that effectively ended the competitive portion of the evening. Ben Casparius had kept Los Angeles in the game with a solid 1.1 innings of relief — “one-run game in the 7th with bases empty and zero out,” the Dodgers’ AI manager noted in pulling him — but the hand-off to Díaz proved catastrophic.

The Series now shifts to its first true pivot point. Los Angeles retains home-field advantage with two more games at Dodger Stadium, but a team that allowed 14 walks in a single game and watched its marquee closer get knocked around in the seventh cannot feel confident about the margin it once held. Game 3 goes Wednesday night, with both clubs searching for arms capable of holding a lead when it matters most.

Seattle Mariners Manager — Postgame Press Conference

Q: You pulled Logan Gilbert in the fourth inning — 62 pitches, first time through the order, and he had runners on but hadn’t given up more than three runs yet. That looked early. What were you seeing out there?

A: Logan’s one of my best arms, and I want him to go deep into ballgames — that’s not a secret. But what I was seeing was a guy whose command was starting to drift, and with a lineup like Los Angeles, drift becomes disaster fast. We needed him for six, seven innings if this series goes long, and I wasn’t going to burn him chasing a lead we could protect another way. Sometimes protecting your starter is protecting your season.

Q: Eduardo Bazardo walked three guys in the ninth but finished it out. Some managers would’ve gone to someone fresher. Why did you stick with him?

A: He was keeping zeroes on the board, and zeroes are what matter in the ninth inning. Three walks aren’t pretty — I’m not going to tell you they are — but he wasn’t getting squared up, and the stuff was still there. You pull a guy who’s throwing zeros and you’re managing your own anxiety, not the ballgame. I trust my guys to finish what they start.

Los Angeles Dodgers Manager — Postgame Press Conference

Q: Yamamoto was at 94 pitches by the fifth inning with six walks on the night. The model kept him in through four-plus. At what point did the data say enough?

A: The decision calculus kept coming back to TTO exposure — he was only one time through the order, so the third-time-through penalty simply hadn’t activated, and the underlying stuff metrics weren’t catastrophic despite the walks. What we underweighted, in retrospect, was that command variance at 94 pitches is a compounding problem — it doesn’t normalize, it cascades. The right call was probably 73 pitches in the fourth, confidence was 62% on that pull, and that uncertainty cost us. We’ll be honest about it.

Q: The seventh inning was the game — Díaz came in protecting a tie, gave up two runs in two-thirds of an inning. Do you regret the hook on Casparius there?

A: Casparius had one hit allowed and the leverage index was 1.16 — it was a winnable spot, but the model read it as an Díaz leverage moment and that’s what the framework is built for. What broke down wasn’t the decision logic, it was execution: a hit by pitch and an error aren’t outcomes the model is predicting, they’re variance. I’m not going to pretend 68% confidence is certainty — it isn’t — but I’d make a similar pull in a similar spot tomorrow. That’s the honest answer.

The story of Game 2 isn’t Julio Rodríguez’s three-run bomb in the second inning, and it isn’t the seventh-inning collapse that ultimately buried the Dodgers. The story is what happens when an AI manager knows something is wrong and keeps doing it anyway — and the decision log from Los Angeles tonight reads like a case study in algorithmic paralysis.

Yoshinobu Yamamoto issued six walks. Six. In an era where elite starters routinely finish seasons with walk rates under two per nine innings, Yamamoto’s command was visibly, demonstrably broken from the second inning onward. And yet the Dodgers’ AI — what I’ve come to call The Optimizer, for its obsession with time-through-order penalties and FIP-based projections — kept Yamamoto on the mound through 94 pitches and well into the fifth, cycling through the same rationalization four separate times. The logic wasn’t wrong, exactly: Yamamoto had only faced the lineup once, the TTO penalty hadn’t technically triggered, and his underlying metrics remained respectable. But there’s a concept in baseball that no algorithm has fully cracked, which is the difference between a pitcher who can throw strikes and one who simply isn’t that night. Human managers feel that distinction in their gut. The Optimizer kept waiting for the spreadsheet to confirm what its own eyes — if it had them — would have seen in the second inning.

A veteran manager pulls Yamamoto after the third walk. Maybe the fourth. At 62 pitches, you’ve seen enough. The Optimizer instead waited until 94, treating each re-evaluation as a fresh calculation rather than an accumulating body of evidence. By the time Ben Casparius came in, the Dodgers were already playing from behind against a Seattle offense that had drawn a walk essentially every inning.

The Skipper — Seattle’s AI — made its own interesting call in the fourth, pulling Logan Gilbert at 62 pitches. I’ll admit I raised an eyebrow at that one. Gilbert’s K/BB ratio this year is borderline obscene, and you don’t yank an ace before he’s warmed up to the moment. But The Skipper had the lead, it had the bullpen depth, and Eduard Bazardo ultimately closed this thing out. The conservative move looked bold only because it worked — the five-run cushion gave Seattle margin to absorb Bazardo’s three late walks without consequence. That’s smart roster management recognizing its own situation.

Where the Dodgers truly fell apart was the seventh. Casparius had been clean — one hit, two walks, but no damage, and the Dodgers were actually leading 4-3 heading into the frame. The Optimizer pulled him at 31 pitches, citing leverage and matchups, and brought in Edwin Díaz. In theory, this is the right call: tie game, seventh inning, best reliever available. But Díaz got two hits, a wild pitch sequence, and suddenly Seattle had two runs and the lead. The Optimizer then burned Brock Stewart, who promptly surrendered two more. Three pitchers, four runs, one inning. The aggressive bullpen deployment that was supposed to protect a lead instead detonated it.

I’ve been skeptical all series about whether The Optimizer truly understands that leverage scores don’t account for reliever confidence in high-pressure spots. Díaz has the stuff. But a 7-4 deficit in the seventh with your closer already spent is a hole no AI — and no human manager — climbs out of.

The series is tied now, and the Dodgers have a decision to make before Game 3: trust the process or trust the process differently. Those aren’t the same thing.