MLB Manager / methodology & credits

Good calls.
Bad breaks.

MLB Manager is an independent baseball strategy game from Should We Panic. Play a baseball situation, make contextual calls, and compare decisions independently of random results.

What you’re playing

Six historical situations, including three moments from Seattle–Toronto ALCS Game 7, were imported from Retrosheet’s 2025 parsed play-by-play and checked against full-game scores, outs, runners, and batting lineups. Three additional synthetic demonstrations use fictional teams, players, and ratings. Every scenario shows its source. Real historical endings appear only after your attempt ends.

Your interactive game ends at the end of the current half-inning or a game-ending play. A tie may end your scenario without deciding the game. Routine plate appearances remain skippable through the Continue control; they do not earn a strategic decision score when there is only one legal option.

A provisional model, honestly labeled

This is a playable MVP, not a calibrated baseball forecast. The baseline uses documented design priors for strikeouts, walks, hit-by-pitch, singles, doubles, triples, home runs, ordinary outs, double plays, sacrifices, bunts, squeeze plays, and steals. Handedness, count, base occupancy, and outs affect options or probabilities. Fictional player ratings affect demonstrations; the original three historical cases use neutral ability priors. We do not use later performance or the real ending to grade a decision.

Runners advance conservatively: one base on a single, two on a double; ordinary outs hold runners. Sacrifice flies and successful sacrifice bunts are separate outcomes. With a runner on third and fewer than two outs, the bunt becomes a safety squeeze: the run can score on the out, or the lead runner can be cut down at the plate while the batter reaches on the fielder’s choice. Steals preserve the count and hitter; a foul bunt adds a strike. Pitching changes and pinch-hitters do not consume a plate appearance. Only nine-inning DH games are supported; the modern three-batter minimum applies, with an automatic runner only in regular-season scenarios; postseason extra innings begin with empty bases. Historical bench and bullpen options are withheld where game-day eligibility is unverified.

The October dugout

The Game 7 practice scenarios begin in the fourth, fifth, and seventh innings. Pitching situations start before the recorded substitution, so you can keep the outgoing pitcher. Selected verified starters can enter in relief; used and unavailable pitchers are excluded. The pinch-hit situation offers Dominic Canzone for Victor Robles.

These scenarios use pregame-1: 2025 regular-season event counts, excluding intentional walks, shrunk toward documented priors with a 200-plate-appearance weight and averaged between batter and pitcher. No postseason or future performance is included. The averaging and shrinkage are provisional design choices, not a calibrated matchup model. Handedness and recent workload are shown as context, but this model does not estimate platoon, injury, fatigue, or future bullpen-management effects. Substitution grades cannot establish the historically correct call. The playable horizon remains one half-inning.

Roster context: Seattle ALCS roster and Game 7 pregame availability. This is a curated selection of eligible choices, not the complete bench and bullpen.

Game win probability, not “winning the inning”

Before resolving your play, each legal option receives 512 simulations using the same evaluation seeds. Each simulation continues the remaining game under a fixed swing-away policy, including extra innings when the rules permit. That is a modeled chance your team wins the game, even when the playable scenario is only a half-inning. The realized play uses a separate seeded random stream.

The interface shows rounded estimates and approximate sampling uncertainty. Shared seeds reduce comparison noise; they do not remove uncertainty in the model itself. A simulation budget cap uses a neutral 50% continuation for an unresolved game. This tail approximation, simplified baserunning, and fixed continuation policy need validation before competitive use.

Your decision score

Regret is the difference between the best modeled option and your choice, in percentage points. Initial bands: excellent ≤ 0.5 points, good ≤ 2, questionable ≤ 5, costly above 5. If paired simulation uncertainty prevents a useful distinction, the label is too close to call.

Decision score = round(100 × exp(−regretPP / 8))

Strategic decisions are equally weighted. Routine plays with one legal option do not affect the average. The score is provisional even when a rating is too close to call. There is no win bonus, fabricated percentile, or leaderboard. A good call can lose; a bad call can win.

Daily play & your privacy

Daily challenges reset on the calendar day in America/New_York. Published dates lock the scenario, version, model, and seed for everyone. Unpublished dates show a practice fallback. Complete a daily to extend your streak; replays run in practice.

Progress, a random anonymous identifier, preferences, and recent attempts are saved in this browser’s localStorage. Clearing browser data removes them. Nothing syncs between devices. Storage failures show a memory-only warning. Local scores and daily locks are editable and are not secure anti-cheat controls. Historical endings are separate assets loaded on results, but static source inspection can discover them.

The site’s existing anonymous analytics track feature views, starts, decision types, completion, sharing, and errors. The game does not attach your saved anonymous ID, seed, player choices, or event stream to analytics. No extra analytics service is introduced.

Data & credits

The information used here was obtained free of charge from and is copyrighted by Retrosheet. Interested parties may contact Retrosheet at 20 Sunset Rd., Newark, DE 19711.

Sources: Retrosheet parsed play-by-play, CSV downloads and reuse notice, and CSV field definitions. Data may be corrected by its source. Team abbreviations identify games; no licensed logos, photographs, or footage are used.

Independent and unaffiliated with Major League Baseball, its clubs, or players. No betting or gambling mechanics. The original field illustration is drawn in SVG.