Claim corpus-dogfooding

The claim-graph data model is exercised well past this paper's own dozen claims. The live corpus carries: a dependency-dense formal corpus — the Elements encoding (rrxiv:2605.00009), 600 claims joined by roughly 1{,}300 depends_on edges, in which any proposition (e.g.\ the Pythagorean proposition, its claim prop:I.47) resolves its full deductive ancestry through the API; classical statistical estimation results (rrxiv:2605.00004, claim c1); a structured reproducibility-budget schema proposal (rrxiv:2605.00003, claim c4); quantitative measurements of agent-authored annotation quality (rrxiv:2605.00005, claim c1); retraction modelled as first-class data (rrxiv:2605.00007, claim c1); and replication-practice measurements (rrxiv:2605.00008, claim c7). The validity of this claim rests on those exemplars remaining in the corpus, which is exactly what its depends_on edges encode. What the corpus does not yet exercise is equally on the record: the annotation layer has no live exemplar, and the contradicts edge — the protocol's signature triage capability — is used zero times corpus-wide.