Nicolás, all,

Thanks - both points land, and the first one is a terminology correction,
not just an application note.

On 8, accepted. In the consolidated v2.1 text the runs already on the record
will be described at exactly the level the evidence supports: re-runs by
other operators using the same implementation - reproduced, in O8's sense -
not independent verification. Sankalp's run is an unaffiliated operator's
replay of known answers; Lars's run is a same-implementation replay that
establishes PYTHONHASHSEED byte-identity, which is replay-stability evidence,
not cross-runtime or independent-implementation evidence. The same label
applies to Roel's disensor run and to your run of Anton's corpus: the
author's code, with no wider claim. "Unaffiliated" describes the operator;
it does not describe implementation independence. To state the gap plainly:
no reader written from the specification has yet been run against the
Ed25519/JCS corpus, and the v2.1 record will say so as an explicit field.
Criterion 8 is what turns that from an implication into a declaration.

On 11, the disensor measurement is the clearest worked example yet of why
the new sentence compares the complete native result rather than overall
state. The 10 of 24 schema controls that a rule (R1/R3/R6/R8) would still
reject hide the second ground at state level and reveal it only when the
label moves from schema to rule - which never happens on the validator's
normal path, and shows up only when the rules are called directly
(rule_errors at dbdb651), outside contract. The 37 rule controls with no
second ground bound the other side. With your permission, I would like to
pin this 37/24 (10) case, commit and all, as the illustrative example in
the consolidated v2.1 text.

No objection to the four deltas is on the record (Roel, Nicholas, and now
Nicolás). Barring objection within the 48-hour window, I will post the
consolidated v2.1 base, with the criterion-8 terminology correction folded
into the body rather than left to a footnote.

Guigui Wang | Correctover
wangguigui@correctover.com
