060-Rationale becomes rests-on and off-pattern grounds are reclassified

Type: ../types/adr.md · Status: accepted

Status: accepted Date: 2026-07-28 Amends: ADR 019, ADR 020, and ADR 058

Context

ADR 058 requires every directional identifier to complete source <label> target. rationale instead names the target's role: its intended assertion is that a descriptive or prescriptive source rests on a theoretical target. grounds has the same grammatical defect, but its intended note→note premise-verification journey is distinct.

A corpus review classified all 134 active rationale edges: 114 are genuine design/rule dependencies, 2 are source-side evidence, and 18 are evidence or architecture/history relations. A separate adjudication classified all 38 grounds edges outside the canonical note→note cohort. During implementation, four new canonical note→note edges landed concurrently; the live baseline is therefore 314 active grounds edges, of which 276 remain in the deferred canonical cohort and 38 require reclassification.

The review also exposed three authorization gaps: agentic-system analysis→note evidence, source analysis→agent-memory evidence, and three theoretical dependencies authored under kb/types/, which had no source collection contract.

Decision

Retire the directional identifier rationale and adopt:

source rests-on target

rests-on is asymmetric. It asserts that the source design, description, procedure, rule, or system-definition artifact depends on the target theoretical claim: rejecting or materially changing the target triggers reconsideration of the source. It is not a generic dependency, implementation, provenance, or evidence relation.

Keep the canonical note→note grounds cohort semantically distinct and unchanged in this migration. Its reader follows a theoretical assertion to assess its premise; its own source-as-subject identifier requires a later scoped review.

Migrate the active boundary exactly as adjudicated:

  • 114 rationale edges and 8 off-pattern grounds edges become rests-on;
  • 35 boundary edges become evidenced-by;
  • 12 become is-evidence-for;
  • 2 become implements;
  • 1 becomes compares-with;
  • none are removed.

Collection contracts replace rationale authorization with rests-on, rather than retaining synonyms. kb/agentic-systems/ → kb/notes/ and kb/sources/ → kb/agent-memory-systems/ gain is-evidence-for only for the observed source-side evidence journey. No collection is widened to authorize off-pattern grounds.

kb/types/ becomes a collection by gaining a minimal COLLECTION.md. It remains the global type layer; the collection contract governs text-level authoring and outbound links, while each type spec and schema continue to own type semantics. This removes the missing-source-contract exception instead of making the self-referential root type authorize its own links.

Operativity path: cp-skill-write and cp-skill-connect load the source COLLECTION.md; collection authors consult the shared catalogue; type-spec authors now load kb/types/COLLECTION.md; and the migrated active footer corpus teaches the same assertions in use.

Considered alternatives

Rename every rationale edge to rests-on. Rejected because 20 of 134 edges assert evidence, implementation, comparison, or decision history rather than theoretical dependency.

Merge rationale and grounds into one identifier. Rejected because 114 cross-register design/rule dependencies and 276 note→note premise-verification edges repeatedly create different follow and maintenance decisions. The 38 off-pattern grounds rows are drift, not evidence for merging.

Authorize the live off-pattern grounds pairings. Rejected because their assertions resolve cleanly to evidence or rests-on. Widening contracts would make historical drift authoritative.

Let the root type spec authorize links for kb/types/. Rejected because it would preserve a source-directory exception to collection-owned authorization and make a type contract govern its own routing. A minimal collection contract keeps type semantics and text/link governance separate.

Remove weak boundary links. Rejected after individual adjudication: every boundary edge has an articulated formal reader need under an existing or adopted relation.

Deciding forces were direct source-as-subject grammar, tuple conservation, stable reader journeys, and one authorization owner per source collection. The free choice was whether kb/types/ should remain an exception; the migration resolves it in favor of uniform collection ownership.

Consequences

Readers can interpret rests-on without silently reversing endpoints, and maintainers can distinguish design reconsideration from premise verification and evidence inspection. The active corpus has 122 rests-on edges after migration. The 276 canonical grounds edges remain visible migration debt rather than being silently reinterpreted.

The global type layer now participates in collection routing and collection-conformance review. Adding or changing a global type must compose kb/types/COLLECTION.md with the type-spec contract, just as other authored artifacts compose collection and type contracts.

This is the second completed directional-label migration. Its retrospective decides whether the stable inventory, exclusion, authorization, tuple-conservation, and guidance-reconciliation procedure is ready for extraction into a reusable instruction before the grounds run.

Relevant Notes: