Benchmark harnesses, golden bug fixtures, reference reports, and local session artifacts used to evaluate review quality and regression behavior.
The directory mixes durable benchmark inputs with generated .ca outputs. Root-level codemaps describe shared benchmark behavior so leaf fixture codemaps can stay minimal or disappear when redundant.
Fixture diffs and expectations are consumed by scripts and test runners, which then emit session metadata, reports, and scored results under .ca/.
Connects to shared golden-bug scoring, CLI/script automation, and reviewer evaluation workflows; it also stores generated evidence for inspection.