Contributing

AWA's runtime behavior is backed by code tests, PostgreSQL integration tests, compatibility matrices, benchmarks, and TLA+ models. Changes should update the evidence at the same boundary they change.

• Development is the contributor workflow, including migrations and local checks.
• Benchmarking explains which performance results are comparable and which are historical evidence only.
• Architecture decisions records durable design choices and their status.
• Correctness models hold the TLA+ specifications and model-to-code mapping.

Start from the repository's agent instructions when working with a coding agent; it contains the always-on validation and Agent Skills rules.