ADR-044: A corroborates relation between the record's own claims, for evidence one claim gives another without the second depending on it, so that a supporting claim's retirement lowers the other's credence without breaking its argument

Context

A claim filed while porting the LEGO material (CLAIM-310) shows that changing only how a collection is arranged changes the cost of retrieving from it. That is evidence for CLAIM-296: what a system can deploy within a deadline depends on its arrangement, not only on what it holds. It also meets CLAIM-296’s defeat condition head-on, which asks for “a case in which a change only to the arrangement of the system leaves every such capability unchanged where this claim says it should move.”

The record had three ways to say this, and each says something else.

• supports / rests_on. supports on a claim is the converse of rests_on (and of grounds). rests_on is “another of the record’s own claims this one could not stand without — the spine of the argument, walked as the argument chain. Retired, it leaves this claim needing another route” (ADR-032). Filing supports: CLAIM-296 made luria link --fix write rests_on: [the new claim] onto CLAIM-296. That is false. CLAIM-296 was argued before the new claim existed and does not need it. Retiring the new claim would wrongly report CLAIM-296 as needing another route, and the argument chain would draw an edge that is not in the argument. The relation was removed before the merge.
• complements. “Answers a different part of the same question, so the two belong together rather than competing.” It is symmetric. Here the two claims answer the same part (does arrangement change capability?), and one bears evidence for the other, so the relation is directed.
• Prose. It is what was done, and it is invisible to the views. Nothing shows a reader of CLAIM-296 that a later claim bears on it, and nothing tells a reader of the later claim that retiring it costs CLAIM-296 something.

Cases already have the distinction for themselves (ADR-032). supports (via the claim’s grounds) is load-bearing, and illustrates is “a claim this case makes vivid without bearing its weight”. Claims have no counterpart of illustrates for each other.

Decision

• A new reference field on CLAIM, corroborates, scheme CLAIM, many, with converse corroborated_by written by luria link --fix. Its blurb: “a claim of the record’s own that this one gives independent evidence for, without that claim depending on it: if this one is retired, the other loses support, not its footing.”
• Not part of the argument. It is not added to the argument chain’s relations, and nothing that walks rests_on (retraction, the chain views, the report of steps crossing lines) follows it. A view can list “corroborated by” on a claim’s page, as illustrated_by is listed now.
• Directed and claim-to-claim only. Evidence from outside the record’s argument is already grounds (THEORY, CASE, LIT). A case that makes a claim vivid is already illustrates.
• The test for which relation to use. If the target claim’s argument, as written, uses the source claim as a step, it is rests_on. If the target stands on its own argument and the source adds evidence or removes a defeater, it is corroborates. If they answer different parts of one question, it is complements.
• Pilot. Accepted by the owner on 2026-10-15 (“accept ADR-044 and apply it”). It was applied first to the claims that prompted it: CLAIM-310 corroborates CLAIM-296. CLAIM-306, whose reading of “If I can’t find it, it isn’t there” bears on CLAIM-296’s case, keeps complements until the pilot says which fits. Nothing else is retagged until the pilot has run on at least one more batch of filing.

Alternatives considered

• Overload complements. It is free, and wrong in two ways: it is symmetric, and the record already uses it for claims that do not bear on each other’s truth at all. A reader could not tell “these go together” from “this one is evidence for that one”.
• supports with a qualifier (for example supports: {code, weight: weak}). It keeps one relation and makes every consumer of supports check the qualifier. The retraction walk is exactly what must not follow this edge, so it should not be the same edge.
• instance_of / instantiated_by. This is more precise for the pilot: the new claim is a special case of CLAIM-296’s relation (resource, arrangement, deadline) for one resource and one kind of arrangement. But a claim can corroborate another without being an instance of it, for example by removing a defeater or by agreeing from a different line. A special case is one kind of corroboration, and can be a later refinement if the pilot finds it common.
• Status quo, prose only. It costs nothing, and the views and the lint stay blind to a relation the record already relies on. Each time one claim is retired, a reader has to find by search which others lose evidence.

Consequences

• One new pair of fields in luria.yaml under CLAIM’s references. No luria change is needed: a declared reference with a converse is already written back by luria link --fix, as complements and illustrated_by are.
• A new judgement at filing time, between rests_on, corroborates and complements. The test above is meant to make it quick. A misfiled rests_on costs more than a misfiled corroborates, since only the first is walked.
• Credence still has to be revised by hand. The relation tells a reader that retiring a claim weakens another, and by how much is the reader’s call, as with illustrates.
• If accepted, a follow-up: sweep the claims whose prose says one claim “supports” or “bears on” another without depending on it, and decide each. The sweep is not part of this decision.