[Levaj2000]

Why

#584 moved the plugin onto the keyed content digest praxis-proxy/policy#84 records, and #585 wrote the digest form into AID-EMIT-1, but no committed vector exercised it: the existing demos never enable provenance, so the form was covered by one unit test only. This adds the vector.

What

• examples/provenance_demo.rs: four finalized decision records carrying unmapped."cpex.content". Unchanged content (equal digests, one key id); a redaction (different digests, same key id, and the redacted value is in neither the record nor a digest, since the OCSF event carries no tool arguments at all); the same content under a rotated key (a different key id, so a reader knows not to compare); and the explicit unkeyed form.
• On PPE the digests are the engine's own ContentKey, the function the executor calls at entry and at emission, under keys the demo resolves through the engine's secret store (env backend, the path engine_settings.content_provenance_key takes at startup). The example feeds it the bytes and places the result on the log the way the executor does; the emitter copies both values and never holds a key. On cpex, which has no provenance key, records 1 to 3 carry the unkeyed form over the same bytes.
• SAMPLE-OUTPUT-PROVENANCE.md: the verbatim PPE output, what each record demonstrates, the cpex delta in full, and a recompute recipe. The demo keys, the key-id label and the canonical audit bytes are printed as // verify lines, so every digest recomputes with Python hmac and hashlib alone.
• docs/specs/aid-emit-1.md: 1.1.2-draft, naming the vector in section 12 and the patch in section 13. No covered byte and no existing vector changes.
• PRAXIS-PORT-RESULTS.md and the crate README name the vector.

Verification

rustup 1.96.1, cpex sibling at the CI rev:

Check	cpex host	ppe host
cargo check --locked --all-targets, RUSTFLAGS=-D warnings	clean	clean
cargo test --locked	35 passed, 0 failed	35 passed, 0 failed
provenance_demo, two runs	byte-identical	byte-identical
provenance_demo vs the committed vector	differs only by the documented digest lines of records 1 to 3	byte-identical
emit_sample, decision_sink_demo vs their vectors	unchanged	byte-identical

All eight digests in the vector were recomputed independently from the printed keys and bytes with Python hmac and hashlib, and the key ids differ across the rotation. The four records parse as JSON and none contains the demo SSN. cargo fmt --check is clean on the example.

One thing to know when reading the vector: the // verify lines deliberately print the pre-redaction bytes of record 2 so the digests can be checked. The records themselves do not carry them.

Generated with Claude Code

https://claude.ai/code/session_019BKU46zuw1Yu8E8yBQj4DP