Status: rule adopted and enforced (2026-10-20). Related: Real Code Corpora §4,
crates/sophia-ingest/src/metamath.md, SPEC-001 §4.
Problem. A record’s kind is stored with it but is not part of its id. When two records with the same id arrive with different kinds, the first one wins. Metamath showed this is a real error, not a corner case. A statement is an axiom in one library and a theorem in another, and with identity by statement only one label survives.
Rule. A record’s kind must be a function of its canonical content (for Metamath: statement or syntax, from the typecode). Anything that can differ between users of the same content is a role, recorded where it belongs:
- in a nominal collection (a library’s axioms);
- in an annotation (a proof);
- in a name (a label).
To do.
Done (2026-10-20):Store::addshould reject a record whose id exists with a different kind, rather than keep the first silently.put_records,addand every other write that stores records refuse it withStoreError::KindConflict, and the batch rolls back. Conformance testkinds_follow_from_contenton all four backends. set.mm, iset.mm and LinearAlgebra ingested into one store raise no conflict.- The sophialang normaliser already derives kinds from content.