Mission Specification: Test-Quality Doctrine Series

Created: 2026-07-26 Status: ⏸ NOT YET SPECCED — deliberately. Scope sketch only. Programme: Mission C of five — see the programme record doctrine-canonical-structure-remediation-01KYEYSD. Order: Fourth. Blocked by B2. Tracker: the original #2935 deliverable.

Why this is a sketch and not a specification

Operator guidance, 2026-07-26: do not over-specify or plan any mission after phase A. Missions run one at a time, each specced and planned only after the previous phase is finalized.

This mission is last for a substantive reason: C-001 and C-002 require the series to be authored edge-native from birth — zero inline references: blocks, every relationship a DRG edge — and that is only coherent against a tree where B2's authored-edge tier already exists. Authoring it earlier would mean writing the legacy surface and then migrating it. Speccing it now would describe an authoring surface that does not yet exist.

Nothing is lost by waiting. The user stories, the traps and the requirement set live in the programme record's spec.md (US3, US4, US5) and plan.md (IC-14).

Scope sketch

Turn the #2934 over-mocking failure into a citable rule. A reviewer seeing a test that mocks an internal of the system under test to pin a call contract can today only argue from taste; after this they can cite a rule, an anti-pattern node, and a remediation procedure.

  • Author the series: paradigm + DIRECTIVE_047 + procedure + 2 anti-patterns + 4 assets, plus the augments.
  • Split DIRECTIVE_041's intent to the new paradigm, linked by a refines edge.
  • Excise duplicated checklists to assets so one copy of each exists.
  • Update doctrine-daphne with the layout convention, the edges-only rule, and the regeneration command.
  • Give the PowerShell toolguide a real inbound edge — or restate its requirement honestly.
  • Close the CLI/CI validator parity gap. Characterize before fixing.

Estimated 3–5 agent-days, ~30–45 files.

Carried constraints — do not lose these when speccing

  • The 041 → paradigm link is the first refines edge any built-in fragment has ever carried (histogram: 0). #2079 fixed a silent refines→applies downgrade that nothing shipped has exercised, so the round-trip assertion is a real risk item, not a formality. The path is the built-in fragment load (load_graph_or_dir), not the org→DRG bridge.
  • Editing DIRECTIVE_041 is its own reviewed change (C-005). It is live at enforcement: required; audit inbound edges first, and do its migration with the intent split rather than as an anonymous line in a bulk pass.
  • Every new resolved-only node needs a proven inbound edge. A loadable, id-resolvable, unreachable artefact is exactly the PowerShell toolguide's failure — "resolves" was operationalized as id-lookup, not reachability.
  • Anti-pattern nodes hold only urn/kind/label/tags — structurally incapable of carrying a rationale. Design for that rather than discovering it mid-authoring.
  • ⚠️ anti_pattern and asset are among the five kinds extractor._KIND_MAP silently drops (measured 2026-07-26: it maps 11 of 16 NodeKind members; the missing five are anti_pattern, asset, glossary, glossary_pack, glossary_scope). This mission authors 2 anti-patterns and 4 assets. Without mission A's site-3 fix they would vanish at extraction with no error — the series would validate, load, and then partly not exist. This makes A a hard prerequisite for C on its own merits, independent of the B2 ordering.
  • De-duplication must be measured, not assumed: the excised content appears exactly once under src/doctrine/.
  • Part of the reported validator gap is likely the frozen inline surface behaving correctly. Where that is so, document it as intended rather than "fixing" it.
  • Zero inline references: blocks in anything this mission authors.