rig

What problem does it solve? Validating a game-playing bot requires reproducible, statistically sound measurements rather than anecdotal runs. This Skill runs the Shatterfish rig over seeded games for one brain or an A/B pair and produces the depth, turns, cause-of-death, and SPRT verdict numbers the project requires before any brain change merges. ## Core Features & Use Cases - Seeded benchmark runs: Executes the rig via ./gradlew :rig:run with named seed sets or counted seeds, recording per-run results to runs.jsonl, summary.json, and sprt.json. - A/B brain comparison: Runs SPRT comparisons between a candidate brain and a baseline, reporting accept/reject/undecided verdicts with LLR and bounds. - Results publishing: Writes docs/results/<date>-<sha>.md in the project's expected format so numbers land in the same PR as the brain change. - Use Case: After modifying the bot's tactical search, ask to compare the new brain against the last published baseline on the standard seed set, and receive a summary table plus a publishable results file. ## Quick Start Ask the assistant to run the rig comparing the current brain against the last published baseline on the standard seed set and publish the results.