Keystone

A specification is a claim about what can be built from it. Keystone is how that claim gets tested: a generator that emits a complete core-protocol peer, in any target language, from the published specification — and a gate that says, per peer, whether the result actually conforms.

The result is a cohort of core peers spanning functional, imperative, array, stack, logic and declarative-query substrates — OCaml, Haskell, Lean, Smalltalk, Forth, APL, SQL, Datalog, COBOL, Rexx and thirty-odd more — each built from the same spec and measured against the same oracle.

What that does and does not prove

It shows the specification is complete and unambiguous enough that a peer can be derived from it without recourse to a reference implementation's source. When a language whose number model, dispatch discipline or memory model has nothing in common with the last one still reaches the same wire behaviour, the property being demonstrated belongs to the spec rather than to any implementation.

It is not evidence that many separate teams independently agreed. The peers share a generation lineage — one generator, one spec — which makes the cohort consistent, not independent. The honest name for what it shows is spec-forced convergence across languages. Independent agreement is a different claim, and the three separately written reference implementations (Go, Rust, Python) are what carries it.

The gate

Conformance is a tracked, moving state, not a badge. The gate is validate-peer --profile core — the extension-free categories: connectivity, encoding, type system, origination, resource bounds, concurrency, plus the register and conformance-handler gates. Most of the cohort passes it; the peers that do not are published with a named defect rather than dropped from the count, and one is a deliberately exploratory probe that was never in scope for the gate.

Read the row, not the summary. A peer sitting a spec version behind, or lacking Ed448 agility, or carrying a known gap, is a documented state with a name and a tier — that is the point of publishing the matrix rather than a percentage. Green rows are pinned to a specific oracle commit and are superseded by re-measurement, not preserved by it.

The generator, the cohort and the matrix are maintained in entity-core-keystone.

Convergence is one of two evidence classes this domain publishes. The other is proof: see Formalization.