The Doom Collection — balance dashboard

Precon calibration — the absolute reference

Each deck 1-vs-3 against real WotC precons. Shaded band = the 20–35% bracket-2 target. No deck sits above it — the library errs weak, never strong.

Precon self-consistency (expect ≈25% each): Fae Dominion 15.8% · Spirit Squadron 29.2% · Land's Wrath 25.5% · Masters of Evil 24.5%. Parser coverage of the precons: Fae Dominion 97% cards / 99% sentences, Spirit Squadron 77% cards / 89% sentences, Land's Wrath 66% cards / 82% sentences, Masters of Evil 64% cards / 84% sentences. — misses weaken the precons, so custom rates are, if anything, overestimates.

Robustness — three AI temperaments

The same pods replayed with cautious, standard and aggressive players. The strongest and weakest decks stay the same everywhere — the rankings are findings, not artifacts of one heuristic.

Politics — grudges, pile-ons and diplomacy

The same pods with politically-motivated targeting: grudgeful players retaliate, archenemy tables converge on a runaway board, diplomats do both and spare harmless opponents. The tick is the apolitical baseline.

Goldfish clock — uncontested kill turn

Each deck solo against a do-nothing 40-life opponent. Shaded band = the ~8–11 bracket-2 window. Everyone is inside it: Arena Eternal's weakness is surviving interaction, not speed.

Commander castable by turn 5

Pure draw simulation, no AI. Age Of Dread is the outlier — a five-pip commander over a basics-heavy base — which is why its games feel bimodal at the table: colour-screwed, or the commander connecting.

How games end

Each deck's wins split by the blow that ended the game. Commander damage matters most for the voltron decks; "pings & abilities" is the library's turret-heavy signature.

Matchups — win rate when facing each opponent

Row deck's win rate in games where the column deck was at the table. Blue = better than its overall rate, red = worse. Each pairing covers 40,002 games.

Turn order — who wins from each seat

Share of decided games won by each seat position, hairline at the fair 25%. Going first is a real disadvantage in this engine (no draw on the first turn) — worth remembering when reading table results.

Notable games

Outliers from the full run — the fastest kills, the biggest comebacks, and the longest grinds, with what the winner saw early.

Theme audit — wiki HP vs card stats

Spearman rank correlation between each creature's in-game hit points and its card power+toughness. Hairline = the ρ ≥ 0.7 gate. doom2016 and darkages list no comparable hit-point tables on the wiki, and the classic sets rename shared monsters (Void Baron, Citadel Hell Knight...) to keep card names unique across the library, which limits title matching - their bestiary stats were applied at design time from the same infoboxes. The audit covers doom3 and eternal numerically.

Wording & feel checks

Oracle templating lint + EDHREC salt cross-reference.

Templating. 20 of 706 rules sentences have no close printed-Oracle analog — all reviewed as intentional custom-mechanic composites; two genuine wording misses were fixed.