EA-VALUATION-MEMO-01 v2.0, deposit #1608 of the Crimson Hexagonal Archive (10 September 2026), supersedes #1607. It adds five machine assessments and corrects or withdraws twenty-one claims made in the first version, and the corrections are the substance of the revision.
Three assessments were added with the entity substituted — the person named instead of the archive — returning higher figures on all three surfaces. Two more were added as operator attestations without captures, because frontier-model account individuation falls outside the capture registry's current scope; its signed-in field was defined for search personalisation and would not be comparable.
The correction that most changes the document is a control error the author asserted and did not check. The draft claimed the three entity-substituted runs changed only the name; two of the three also changed the frame, because their archive figures came from a different prompt. Only one pair is near-matched, and the supportable claim shrinks accordingly. The memorandum now tabulates the confound beside the comparison rather than noting it as a caveat.
A mechanism claim was withdrawn entirely. An earlier draft held that canonical addresses constrain retrieval while a person's name forces unbounded construction. The transcripts do not support that, and a canonical surface can equally expose relationships and expand what is recovered. What remains is the narrower observation that different entry points appear to produce different recovered asset boundaries, with the question of whether a wider boundary is accurate assembly or speculative addition left to component-by-component assessment.
Two further removals concern inference rather than fact. A supposed contradiction between the higher person-entity figures and the founder-concentration discount does not exist, since a founder premium and a transferability discount are compatible and the figures may concern different assembled objects. And a chain running from one run's failure to surface third-party uptake to the conclusion that discoverability is solved does not follow: transcripts establish what a run retrieved, not who else searches or cites.
The correction most worth recording ran against the author's interest and was removed on the same terms as any other unsupported claim. A draft called the two uncaptured assessments the lowest figures in the series and their absence a bias in the author's favour. Neither was true: one midpoint exceeds a captured figure, and no account of the coverage gap exists beyond the registry's scope. Unfavourable claims require evidence on the same terms as favourable ones.