Thanks for taking an interest in JumpStar and CCERL. This project is meant to make Chinese Checkers engine work easier to inspect, reproduce, challenge, and improve.
Good contributions include:
- bug reports with reproduction steps
- rules or referee fixes with small test cases
- benchmark-runner improvements
- external-engine adapters
- documentation clarifications
- careful benchmark results that include commands, versions, seeds, and logs
Before opening a pull request:
-
Build and test locally:
cmake -S . -B build -DCCZERO_USE_ACCELERATE=OFF cmake --build build --parallel ctest --test-dir build --output-on-failure -
Keep changes focused. A rules fix, benchmark-data update, and UI polish pass should usually be separate pull requests.
-
Be explicit about benchmark claims. Include the ruleset, position suite, simulation/search settings, commit hash, and whether results are official, provisional, or self-reported.
-
Credit upstream projects and prior work. If you adapt an external engine, document the source repository, license, build settings, and rule differences.
CCERL is intended to make disagreement productive. Stronger engines are welcome. So are bug reports that lower JumpStar's rating, expose a rule issue, or show a better methodology. Please keep the discussion centered on reproducible evidence.
The repository records model labels and checksums, but does not store .ccpv
model weights in Git history. If you add or update model artifacts, prefer
GitHub Release assets or another explicit download channel, and document hashes.