ID

R302

Status

Backlog

Bucket

structural

Priority

4

Theme

classification-model

Created

2026-06-13

Updated

2026-06-13

Rename ChildField to SourceField (carrier-named field hierarchy)

Split out of R290 (datafetcher-field-dimensional-slots). R222’s refined field-side model names the three carriers after their GraphQL parent-type category: Query → QueryField, Mutation → MutationField, and Source → SourceField. The first two already match their carrier; the third is still called ChildField, a name that predates the carrier vocabulary and describes position ("a child of some parent") rather than the carrier it is ("the Source carrier"). The corpus already asserts carrier: Source for every ChildField leaf (R299), so the rename closes the last gap between the model’s vocabulary and the code’s.

This is pure mechanical churn: a type rename with no behavioural, classification, or emit change. ChildField and its nested leaves (ColumnField, TableField, NestingField, …​ the full sealed set) rename to SourceField; the ~940 references across ~100 files (model, generators, validators, catalog, and the test corpus) update with them. No leaf is added, removed, or reclassified, and no @classified verdict changes, so the R281/R299 corpus stays byte-identical.

Why split from R290

R290 materialises carrier x intent x mapping on the field, derives the fetcher mechanism, deletes the throwaway leaf-to-tuple adapter, and dissolves the ConstructorField leaf (the SingleRecordTableField collapse was split to R305). That is load-bearing architectural work whose reviewability depends on a small, readable diff. Folding a ~940-reference rename into the same commit range would bury the substantive change under mechanical noise, exactly the bundling R222’s "vertical slices each ship their own vocabulary" technique exists to avoid. The rename touches the same files R290 does but is independent of it: neither blocks the other, and whichever lands second rebases trivially (a rename has no semantic merge conflict). Landing this first lets R290’s diff speak in the final SourceField vocabulary from its opening commit; landing it last is equally fine. No ordering edge is declared for that reason.

Scope

  • Rename the sealed type ChildField → SourceField and update every reference (imports, case arms, javadoc, switch patterns, fixture references).

  • Keep the nested leaf names as-is (SourceField.ColumnField, SourceField.TableField, …​); only the enclosing type renames. (If any leaf name independently reads as position-named rather than carrier/intent-named, flag it but do not rename it here, that is a separate judgement call.)

  • Update prose references in graphitron-rewrite/docs/ and the R290 appendix that still say ChildField where they mean the Source carrier.

Acceptance

mvn -f graphitron-rewrite/pom.xml install -Plocal-db green with no other diff: the R281/R299 corpus byte-identical (every carrier: Source verdict unchanged), LeafCoverageReportTest leaf count unchanged, and no generated-output change. A rename that needs any test assertion edited beyond the type name itself is out of scope, that signals an unintended behavioural change.