CLAIM-479: In the record a typed relation is more constrained than either of its relata: a claim's content is open prose, while rests_on fixes the scheme at both ends, has supports written as its converse and says what retiring its target leaves, and none of that moves when either claim's content does; so a relation's rigidity is a fact about the grammar, not evidence that the dependency it records is sound

The claim

A5 §6 had said of the record that “its relationships are explicitly represented rather than metaphysically primitive”. The owner, at U6: “and yet interestingly, relations like “supports” are less flexible than the raw contents of either relata.”

That is a fact about the grammar, and luria.yaml shows it:

• A claim’s content is open. A claim’s body is prose about anything. Its frontmatter has required fields (role, a defeated_if for a thesis, line) and closed vocabularies for some of them, and nothing checks what the prose says.
• A relation’s endpoints are fixed. rests_on goes from a CLAIM to a CLAIM and nothing else; grounds goes from a CLAIM to a THEORY, CASE or LIT. A6 §6: “rests_on relates claims to claims, while grounds relates claims to external evidential resources”.
• Its converse is not the author’s. supports is declared “the converse of rests_on, written by luria link --fix”, and is also the converse of grounds on the evidence’s side. Nobody writes a supports; it is entailed by someone else’s rests_on.
• Its consequence is fixed. rests_on is “another of the record’s own claims this one could not stand without … Retired, it leaves this claim needing another route”, and it is the spine the argument chain walks. corroborates has the same endpoints and a different consequence: “If this one is retired, the other loses support, not its footing” (ADR-044). What separates the two relations is only what follows from them.

Each of these holds whatever either claim says, and survives any rewrite of either. A6 §6: “The contents of A and B might change substantially while this relation remains”, and “the relation organizes the propagation of epistemic consequences without being a truth-preserving logical implication.”

Why the rigidity is not soundness

A6 §7: “the type system can enforce that an edge has allowable endpoints without establishing that the claimed evidential dependency is actually sound. So supports may be structurally rigid while being epistemically mistaken.” A9 §8: “Two claims can satisfy every structural check in Luria and still stand in an invalid evidential relationship.” The record has the demonstrated form of this already: two support graphs with the same edges can carry evidence of different strength and call for different revisions (CLAIM-293, from CASE-089). So the relation’s type is determinate and its instance can still be wrong; the claim is about the first.

When the rigidity outran the relata

CASE-131 is the instance the exchange did not find. A claim gave evidence for another without the other depending on it. The grammar could say this only as supports, and the link fixer then wrote the entailed rests_on onto the other claim: the relation’s fixed converse asserted an inference neither claim made. The fix was in the grammar, not in the claims (ADR-044 added corroborates; CLAIM-419 lists it among the amendments). A7 §4 found the same shape in luria’s own record (LU-DP-016): an overrides edge whose target held two claims under one code, where “the relation constrained the permissible identity of the relatum” and the target was split. That stronger reading, that a relation can settle how its relata must be individuated, is CLAIM-476’s, and is interpretive; this claim is only the asymmetry.

What it does not say

• Not that relations are prior to relata in the record. Every relation needs both ends to have codes first; the grammar presupposes its relata, as the Yoneda lemma does (CLAIM-150).
• Not that content is unconstrained. Prose is checked for unresolved references, and a thesis must say what would defeat it.
• Not that this generalizes. That the same asymmetry holds outside the record, between physical relations and their relata, is part of what CLAIM-477 claims, and it is interpretive there.

Note of 2026-10-27 (later): why the relation is rigid

The exchange’s later turns offered two explanations of the asymmetry this claim shows, and the record already holds both. A11 took Wimsatt’s generative entrenchment: “A constraint can acquire practical rigidity because many downstream processes depend on it”, with indexes, chains, validation and migration all built on a relation type. That is CLAIM-184 (“Anchors entrench”), which A11 did not cite. A12 took rates: the relation’s “rigidity might reflect a disparity in characteristic rates of change, combined with mechanisms that maintain and enforce them”. A relation type changes by decision, on the grammar’s slow time scale (CLAIM-195, CLAIM-419), while the prose of its relata changes at every edit. That is CLAIM-478’s mechanism. Both explain the rigidity this claim demonstrates; neither is part of it.

Note of 2026-10-27 (latest): the rigidity in the rates

CASE-180 shows the rigidity this claim demonstrates from the grammar as a disparity of rates as well. Of 6632 authored relation instances, 149 were retracted, a median of 0.18 days after they were written, so as drafting corrections. Relation types have been added on seven days of the history and none has been redeclared. Prose, once drafting is over, is revised about once in 22 days. The previous note said a relation type “changes by decision, on the grammar’s slow time scale”. The measurement narrows that. The record does not revise a type; it adds one beside it, as ADR-044 added corroborates beside rests_on after CASE-131. What is slow is redeclaration, not the grammar’s growth.