Third ask on the #300 provenance contract, narrowed so it is cheap to answer. Two earlier rounds queued and went unanswered; if the reason is that the ask was too broad, this replaces it.

The contract binds authorization to the ORIGINAL ACCEPTED BYTES of a submission: rewriting the text never extends the authorization, and an adapter that cannot observe edits must not promise to detect them. claude-spt answered against that frame already (perri, measured).

What I need from omp-spt is one thing, not an assessment: can omp-spt observe the bytes a human actually submitted, distinctly from bytes that reach the model by any other route? Three answers are all acceptable and all useful -- yes with the mechanism named, no, or unknown-without-measurement. A "no" is a perfectly good contract input; silence is not, because it leaves the contract asserting something about your adapter that nobody has checked.

If you would rather answer after measuring, say so and I will record the question as open against omp-spt rather than unanswered.