Rendered from docs/obligations/0137-a-british-spelling-reaches-a-serialized-grain-string-so-the-rename-is-a-corpus-change-and-a-derived-bytes-change-at-once.md in the Headwater corpus. Every document on this half of the site is typed by the taxonomy the descriptor names: corpus.json.

A British spelling reaches a serialized grain string, so the rename is a corpus change and a derived-bytes change at once

Context

CLAUDE.md rules American spelling everywhere, and language.controlled.not_met enforces it from the SPELLINGS table in engine/crates/check/src/language.rs. Neither neighbourhood nor centre is in that table. No rule reads a .rs file, so the engine source passes. Governed prose passes too, because the table does not hold the word. Issue #191 measured the gap and split this obligation off, rather than put a fixture re-bless into a change about readability.

Obligation

The corpus owes one spelling of this concept.

The string is in a published, versioned format. Grain::name in engine/crates/check/src/scope.rs writes Grain::Neighbourhood { .. } as the literal string neighbourhood. That string reaches cache keys. It also reaches rules[].scope of the JSON report, which the headwater check contract defines, and the same member of the SARIF report. engine/crates/adapter/tests/fixture.json and engine/crates/adapter/tests/fixture.sarif pin it. Every release from v0.1.0 shipped this value.

A rename is the first change to that shape that adds no member. The contract says that each version of the JSON shape added a member, and that each addition moved the minor. A renamed value removes a value that a reader may branch on. So a rename needs a decision on how the shape version moves: a minor raise, or a major raise to 2.0.

Waiting makes the rename cost more. Before an adopter reads the report, a rename costs a re-bless of the fixtures that pin the bytes, and a version raise. After an adopter reads it, a rename also costs that adopter a migration. So the next act is a ruling, and an adopter is not the trigger.

The rest of the rename is mechanical. A rename of the Rust identifier is a compile-time change, and the compiler finds every site. The specification uses neighbourhood as its own term, in docs/spec/12-check-layer.md and in docs/spec/glossary.md. A rename of the code alone leaves the engine and its own specification with two words for one concept. centre occurs in engine source too, and it reaches no serialized value.

Discharge

The questions for a person: is the serialized grain string renamed to neighborhood before a first adopter? Does the JSON shape version then move to the next minor or to 2.0? And do neighbourhood and centre join the SPELLINGS table? If they join it, headwater check --fix rewrites the specification sentences, and that is the same corpus change through a different path.

When the ruling is to rename, one change moves the identifier, the term in the specification, and the serialized string together. The change description states the serialized string that moved and names every re-blessed fixture, so HEADWATER_BLESS=1 does not carry the diff silently. The contract records the version raise under its own heading. The full suite and every gate pass, and the change reports the cache identity hits + misses + unkeyed == instances before and after.

When the ruling is to keep the string, the contract records that neighbourhood is a fixed value of the shape. That statement discharges the record.