Skip to content

T-7LA2-capability-attachment-rate

Status: planning/draft · Impact: medium · Complexity: small

A capability earns storage by carrying information the code cannot express — otherwise it is a hand-maintained table of contents. Ship a deterministic metric that measures, per capability, what is attached to it, so the corpus can be pruned on evidence instead of argument (2026-07-17 model discussion; the falsifiable test behind keeping Capability stored while deriving code-fact edges).

Prose (greenfield): no attachment measurement exists. The only live capability consumers are index sectioning and site pages; attachment surfaces are wikilink references (related: + body prose across the corpus) and parent_key. locations[] arrives with T-2KK8-capability-kind-grains-and-locations; Requirement and Relation endpoints arrive with their entities.

sdlc capability attachments (or an entities audit metric): for each capability instance, count and list (a) inbound wikilink references from tasks/decisions/milestones, (b) locations[] entries (post-T-2KK8), (c) future surfaces as they activate (Requirement refinements, Relation endpoints). Report an attachment rate for the corpus and flag zero-attachment capabilities as prune candidates. Read-only; no frontmatter changes.

  1. Define v1 attachment surfaces (wikilinks, locations) and the report shape; wire as a registry op with --output json.
  2. Fold a summary line into entities audit (report-only, non-failing).
  3. Feed the regrade triage: the pass that assigns structural kind to the 20 instances also reviews zero-attachment flags for pruning.
LocationKindChange
apps/sdlc/lib/model/entities/capability/modifyattachment-count op colocated with the entity package
apps/sdlc/lib/model/ops/audit.tsmodifyreport-only attachment summary
  • AC-1: the op reports per-capability attachment counts and sources with --output json.
  • AC-2: zero-attachment capabilities are flagged (report-only).
  • AC-3: entities audit carries the corpus attachment-rate summary without failing on low rates.
  • Acting on the metric (pruning) — that is the regrade-triage judgment pass, not this op.
  • Derived-edge surfaces (calls/reads/writes analysis) — separate discussion thread, undecided.
  • 2026-07-17 Socratic model review: “if relations are derivable, why keep capabilities?” — resolved as derive facts, store commitments, with attachment rate as the falsifiable keep/prune test for the capability corpus.

← Back to Tasks