# identify-conformance — media identification

The corpus of [`extensions/media-identify.md`](../extensions/media-identify.md) §10.

**Feature equivalence.** A conforming implementation, given a vector's file, gives the whole result
`vectors.json` states for it: every candidate with its evidence, where its walk stopped and why, its
facts, tracks and capabilities; what was embedded; and the measures. A vector with `repeat` is its
file that many times over, for what happens past the head and past the scan (§3.2).

- `ext/` — files written by real muxers and encoders, from test patterns and sine waves. The
  protected file uses an all-zero test key.
- `gen/` — what `wai_identify_vectors gen` writes: generated cases and negative controls, each saying
  in `vectors.json` what it tests, many of them a change to a file of `ext/`.
- `vectors.json` — every vector's file, SHA-256, origin and whole result.
- `prober.json` — what an independent media prober reads of `ext/`: each file's format and streams.
  `identify_verify.py` holds the result to it wherever both state a value, so the reference and the
  second implementation agreeing is not all that is checked.
- `identify_verify.py` — the second implementation, written from the extension.

**Provenance.** `ext/` and `prober.json` are written by `tools/make_external.sh`, with open tools; it
is not run in CI, because it needs them. `tools/oracle.py` writes `prober.json`.