The claim
Luria’s working agreement is “Never edit a generated file”. A view is rebuilt from sources, and an edit made to the view is lost or drifts (LU-DP-003). When an argument exists only in its manuscript, the manuscript is a view that is being edited as if it were the source.
The translation exchange is the case. At U38 the owner asked for “the whole (crystallized) argument”, and three minutes later for a manuscript. The draft:
- changed the thesis from CLAIM-017 to CLAIM-115, with no turn arguing the move;
- added “necessary” to “not sufficient”, supported by an example that shows only insufficiency (CLAIM-053);
- dropped 24 items of the crystallized argument, none of them on request.
The second crystallized argument (A203), the terminal state at U56 (TERM-052), announced its change of question. It also left out Tannaka reconstruction, which A198 had offered as a model for rebuilding a communicative kind’s symmetries from its realizations (QUESTION-036), CLAIM-125’s two open problems, and more. Maturing can legitimately drop things, so some of these may have dissolved rather than been lost, but nothing in the exchange says which. Each draft read as finished. None of the omissions could be seen, let alone classified, without a list of what came before.
The record supplies that list. A claim has a status (does the record
believe it?) and a works membership (is it in the paper?), and ADR-032
keeps the two apart because they diverge. A manuscript is then one
marshalling of claims whose status can be read, and a dropped commitment is
a claim that is still Proposed and that the work no longer includes.
What it does not say
- It does not say a manuscript can be generated from the record. The prose is written, not assembled. “View” means that the record is the reference the draft is checked against.
- It does not say that every change of thesis made in a draft is wrong. It says the change should be visible.
- It does not say the record makes every dropped commitment visible. A move that fits none of the record’s schemes or relations was never filed, so dropping it from a draft leaves no trace against the record either (CLAIM-181). The record shows what was dropped only of what it held.
Note of 2026-10-11: why drafts lose material silently
Under the owner’s reframing (CLAIM-185), this claim gets a mechanism.
A draft persists, but its claims are not addressable one by one, so nothing
rests on them (CLAIM-182). When a draft drops one, no reference breaks
and nothing reports the loss. A claim in the record is an anchor. Dropping
it from a work leaves its inbound references standing and its works
membership visibly gone.
Note of 2026-10-15: split under ADR-043
ADR-043’s classification found this claim compound: that a manuscript should be a view over the argument is interpretive, while that the drafts of the What Survives Translation? exchange dropped commitments and changed the thesis without saying so is empirical, from one exchange. The empirical part, scoped to that exchange, is now CLAIM-324, on which this claim rests, as it already rests on CLAIM-177. This claim keeps the rest and stays interpretive: the general form, that a draft written straight from an exchange behaves this way, generalizes from one case, and the remedy is a proposal. Its title already states the general form, so it was kept.
Note of 2026-10-29: what holds it back
It generalizes from one exchange (CLAIM-324) to any draft written straight from an exchange, and the comparison its defeat condition names has not been made: no draft written against the record has been checked, beside one written straight from an exchange, for dropped commitments and unannounced changes of thesis.