Milestones
List view
A learned value function that plays fair Done when: Beats the hand-tuned Evaluation under the Sequential test
No due date•0/1 issues closedThe instrument becomes a teacher Done when: Feature-complete per bootstrap §5
No due date•0/1 issues closedWhole-Run planning and a cited knowledge base Done when: A measured win rate, and every heuristic linked to a Rule or a Lore entry with a Tier
No due date•0/1 issues closedBetter fighting, if and only if the numbers say so Done when: A candidate design is accepted against the one-ply model under the Sequential test, or the measurements are published and the question is closed
No due date•0/1 issues closedA person can watch the bot think, step it, and take the controls mid-fight Done when: A full sewers Run watched end to end with a mid-fight takeover and no desync
No due date•0/1 issues closedA hand-built bot that plays the sewers competently Done when: The Warrior kills Goo on at least 75% of the `goo` Seed set with a lower bound of at least 70%
No due date•0/1 issues closedAny claim about a Brain can be measured, published and reproduced by a stranger Done when: A random-agent Baseline is published and a deliberately worse Brain is rejected by the Sequential test
No due date•0/1 issues closedEvery fact the Brain needs about the game, generated from the pinned code and never drifting Done when: One Gradle task regenerates `codex/<tag>/` and CI fails on drift
No due date•0/11 issues closedThe game runs headlessly, reproducibly, and behind a fair Observation Done when: A seeded Warrior Run completes headlessly, the same tuple twice is identical across two JVMs, and the whole fairness suite is green
No due date•19/31 issues closedRepository, upstream pinned, builds without Android SDK, module skeleton with ArchUnit rule, docs system, CI, CLAUDE.md, project skills, BMAD installed, all BMAD planning artifacts through readiness check, issues mirrored. Done when the §11 checklist in docs/BOOTSTRAP-PROMPT.md is complete.
No due date•13/14 issues closed