Ten Dawncaster-grounded card-game strategists, each assigned one of Axiomancer's ten themed deck presets, piloted their kit through the Shallows, the Long Road, the Deep Wood, and the ceiling probe against The Incompleteness — 320 simulated engagements in total. This is the battle report, and the adjustment plan to bend the curve toward where it should sit.
This is an LLM-adjudicated ledger, not an engine-executed one. Axiomancer's real deterministic playtest harness exists (npm run combat-playtest, Spec 32 v3) — but this machine has no Node.js runtime installed, so the actual TypeScript combat engine could not be run. Instead, ten grounding passes were done against the real source: the exact card text and mechanics for each 7-card preset theme (combat.deck-presets.ts / cards.library.ts), the real stage profiles and computed enemy HP (combat.stage-profiles.ts, enemy.library.ts, the enemyStatBudget formula), and the doctrine constants that govern combat (THE STRIKE IS DEAD, threat escalation rates, the health formula). Ten independent expert agents then reasoned out full playthroughs from those real numbers, and a second adversarial pass audited every win-rate claim against the ground truth before it was allowed to stand.
Each expert was first grounded in a matching Dawncaster archetype — pulled live from this machine's Dawncaster card/keyword knowledge base — to bring real deckbuilding theory (stack-then-detonate, denial-without-a-payoff, sacrifice-for-undercosted-power, and so on) to the read of each Axiomancer kit, rather than analyzing it cold.
Treat the win rates below as calibrated estimates grounded in the real numbers, not as a substitute for running the actual engine matrix once Node is available on this machine.
Grouped by stage: where the average sits today, the assessed gap, and the specific deck-level plays or hypothetical balance changes that close it — attributed to the expert who found it.