EDHbots
Calibration

Simulate your Commander deck.

Stop guessing. Run thousands of reproducible games to measure win rate, speed, consistency, and matchups — and analyze a single card's impact before you sleeve it.

Free · no account · runs in your browser. Seats 2–4 come from the pinned bracket-3 field.
Heroes in a Half Shell (exact precon) · optimized against the bracket-3 field
current build
16.1%±2.3%
optimized build
19.7%±2.5%
confirmed difference +3.6pp±3.4measured at a fresh seed the search never saw

cut Coin of Mastery · add Purphoros, God of the Forge — the only swap that cleared zero; the search stopped there.

1,000-game confirmation at fresh seed 40402 · search seed 40401, 70 pairs named → 12 runnable · engine rev57 · simulated win rate — model-relative, not a claim about your table

The same precon we took to a real table. See the calibration record · Optimize your deck · All precons

Commander isn't solitaire. Neither is our simulator.

Four seats, turn order, an interaction layer, archenemy targeting, commander recasts — thousands of games, and the output is a win probability with a confidence interval, not a made-up power score.

Atraxa, Praetors' Voice — art
seat 1
Atraxa, Praetors' Voice
43.7%±3.1%
Krenko, Mob Boss — art
seat 2
Krenko, Mob Boss
7.6%±1.6%
Edgar Markov — art
seat 3
Edgar Markov
21.0%±2.5%
Talrand, Sky Summoner — art
seat 4
Talrand, Sky Summoner
27.7%±2.8%

a real pod: 1,000 games · seed 20260831 · engine rev57 · mean game length 9.8 turns · win shares of 1,000 decisive games

We registered a prediction. Then we played ten real games.

Before the first physical series, the model registered its numbers in advance: 71.3% ±3.7 for the favorite. The table then went 10–0 — the right winner, by the predicted route, faster than predicted. The clock miss landed exactly inside an approximation the results page had disclosed before game one.

We published the miss, shipped the fix it pointed at (team-wide triggers), and re-registered under the new engine — 75.9% ±2.7, seed attached, waiting for the next series to check it. The full calibration record is public, and it stays public when it is unflattering.

the Commander analytics platform

Thousands of seeded, reproducible games run in your browser — mulligans, mana, goldfish clocks, and the four-seat pod. The official bracket definition has two halves: contents and speed. Rule tables check the first; the second takes a pod — opponents holding answers, not a goldfish. No account needed.