The Recoding of a Distinction, deposit #1609 of the Crimson Hexagonal Archive (11 September 2026), reports a ten-string comparative run on one composition surface and names its empirical object: lexical persistence under conceptual recoding.
Five distinctions declared with months of DOI-anchored propagation before the archive's repository termination were run against four declared after it and one boundary member declared six days before. Four of the five earlier concepts resolved unquoted, as themselves and in one case with attribution to their authors. None of the later five resolved unquoted. Quotation then forced all five into composition and in four cases did not recover the concept: each string was reassigned to an established neighbouring field — a clinical error, a subspecies of technical debt, an econometric model, an optimization heuristic. One remained itself under the same condition and uniquely had an indexed external path.
Support does not separate the cohorts. The concepts that resolved did so on one, four, eight and sixteen supporting deposits in their first ninety days; those that were recoded failed on two, three, four and four.
The paper's instrument was found by accident. A term that must be quoted to retrieve is being handled as a literal rather than as a concept, and forcing the signifier across the boundary reveals what a composition layer does when it holds a string and lacks the relation that gave it sense. It does not leave a hole; it completes the string from an adjacent territory.
The reading at §3 is schizoanalytic and is marked as a reading. Provenance erasure has been measured privatively, as a rate or a shortfall, which asks what a wound signifies. The run does not fit that form: nothing failed to be found, and a fluent sourced answer was produced about something else. What is severed is the term's relation to the body of work that produced it — the lexeme crosses and the lineage does not — and the archive's entity persists while its distinctions are reassigned, because an aggregate stabilised across many addresses is not detachable in the way a single minted term is.
A constraint is disclosed and stated as co-occurrence rather than cause. A second repository account, lightly posted with loose metadata, is not entering retrieval. The historical configuration was never decomposed, so neither entrance nor enforcement can be attributed to any of its terms. The confounding prevents causal decomposition and the enforcement history prevents the replication that would resolve it; the paper states that these are both real and not alternatives.