Parity system

The project tracks LiveKit Agents behavior through a manifest-driven parity gate.

Use this page when a docs or source claim depends on LiveKit-reference behavior. A passing Go test can be good regression evidence, but parity-sensitive claims should point to the manifest, a cross-runtime case, or a Go test that intentionally encodes reference behavior.

The central manifest is:

scripts/parity-fixtures/test-cases.tsv

Supported case types are:

• go-test: runs Go tests that encode reference behavior.
• cross-runtime: runs a Python reference runner and Go runner with the same TSV-embedded input_json.
• symbol-report: validates symbol inventory fixtures.

Run the gate:

scripts/parity-gate.sh

Run one case:

scripts/parity-gate.sh --case <case-name>

Do not add one-off JSON fixture files or one runner per behavior. New parity-sensitive cases should use the TSV manifest and existing runner types whenever possible.

Evidence levels

Case type	What it proves
go-test	Go behavior passes a named test linked to reference behavior.
cross-runtime	Python reference and Go target produced matching normalized JSON for the same TSV input.
symbol-report	Symbol inventory matches a fixture; it is not behavioral proof.

Use scripts/parity-gate.sh --local only for fast local feedback. It is not final validation.