Contributing to EMET

EMET is a re-derivability-based integrity verifier. Its value is that every verdict it emits can be reproduced, and that it refuses to become an authority. Contributions are welcome - within the boundaries below, which are not negotiable.

The boundaries a change must respect

A pull request must not make EMET:

• emit a verdict outside the closed lattice MATCH | DRIFT | UNVERIFIABLE (in particular, never TRUSTED);
• adjudicate a model's safety or content decision;
• run inside, or depend on being hosted by, the system it audits;
• enforce, block, sign, or actuate of its own accord;
• depend on a secret or a held key, or on any claim it cannot re-derive.

A change that bends one of these is out of scope no matter how useful. See SPEC.md section 6.

Running the checks

python test_membrane.py                  # behavior proof (23 tests)
python conformance/run.py membrane.py    # reference conformance (40/40)
python membrane.py selftest              # self-hash

Both implementations must pass the conformance vectors. If a change alters behavior, update SPEC.md and conformance/vectors.json together - the spec is normative, and the vectors are how an independent implementation reproduces it.

The most valuable contribution: another implementation

An independent implementation written against SPEC.md alone (not by reading existing code), in any language, is the highest-leverage contribution. When it passes conformance/vectors.json, re-derivability gains another witness - and a different-author witness is exactly what the project still needs. Where your implementation and the spec disagree, fix the spec: those divergences are the point.

Markers and the corpus

The marker denylist is a known-signature set, not a proof of completeness (SPEC sections 8, 11, 16). A proposed marker is data: include the signature, a rationale, and a test input with its expected count. Absence of a marker is never absence of injection.

Security issues

See SECURITY.md. Do not open a public issue for an unfixed vulnerability.

Provenance of authorship

Parts of this project were developed with AI assistance and are committed with Co-Authored-By trailers. Contributions are accepted under the project license (MPL-2.0).