Before calling a capability complete, capture exact source/toolchain identity, build hashes, runtime exit codes, required marker counts, fatal-marker sweep results, and any media-integrity hashes relevant to the test.

Synthetic tests are useful for development but do not replace external runtime evidence.