Conciseness

The body is mostly efficient — non-obvious operational gotchas like 'the job sits queued forever' without --follow on PGLite and 'apply defaults to false' earn their tokens. But it can clearly be tightened: the 'Anti-Patterns' section appears twice (lines ~175 and ~229) with overlapping content, the boxed 'celebration summary' is decorative, and the 'Mental model' paragraph narrates history ('accreted 94 distinct pages.type values over years') beyond what's needed to act. This matches anchor 3 — mostly efficient with some unnecessary content — rather than 4, where only minor trimming would be needed.

Actionability

Fully executable throughout: exact commands with JSON params ('gbrain jobs submit unify-types --follow --params "{\"target_pack\":\"gbrain-base-v2\",\"apply\":true}"'), expected outputs for verification, a copy-paste rollback SQL statement, and concrete failure signals ('stays queued here means no worker is running'). Guidance covers the common cases end-to-end.

Workflow Clarity

A five-phase migration with explicit checkpoints befitting a batch/destructive operation: dry-run preview before apply ('don't proceed... file an issue'), a dedicated Verify phase with expected findings ('pack_upgrade_available → ok', '≤16 distinct types'), a full Rollback section with a 72h restore window, and feedback loops ('A job that stays queued here means no worker is running; resubmit with --follow'). The validation cap for batch operations is satisfied — not a 4 because checkpoints, error recovery, and rollback are all explicit.

Progressive Disclosure

No bundle files exist, so this scores on body structure: well-organized sections with clear headers and a final Reference list of one-level-deep external links. Anchor 4 fits — good structure, mostly well-placed content — but the duplicated Anti-Patterns sections, the multi-line Output Format transcript, and the Contract section's detail are candidates for separate reference files in a ~270-line body, which is the 'minor organization gaps' of anchor 4 rather than the clean split of anchor 5.

Specificity

Multiple concrete actions are enumerated: 'Migrate a brain from gbrain-base... via gbrain onboard --check + the unify-types Minion handler' and 'Collapses 94 noisy types to 15 canonical with subtypes, alias rows, and link rows'. Coverage is comprehensive — it names the exact tooling, the transformation, and its outputs. Not a 4 because there are no minor gaps in action coverage; every clause describes a specific capability.

Completeness

Explicitly answers both: what — 'Migrate a brain... to gbrain-base-v2's 14-canonical-type taxonomy... Collapses 94 noisy types to 15 canonical'; and when — 'Triggers when an agent notices pack_upgrade_available, type_proliferation, or asks "what is the canonical taxonomy / how do I clean up my page types"'. Both are concrete and explicit, matching the anchor-5 example's structure.

Trigger Term Quality

Good natural-phrase coverage: 'how do I clean up my page types', 'what is the canonical taxonomy', 'migrate', 'unify'. It falls at anchor 4 rather than 5 because common synonyms a user might actually say ('consolidate my types', 'retype pages', 'migrate to v2') appear only in the frontmatter triggers list, not in the description itself, and 'pack_upgrade_available' / 'type_proliferation' are internal jargon findings rather than user-spoken terms.

Distinctiveness Conflict Risk

A clear niche (gbrain pack migration to a named taxonomy) with named tooling and trigger phrases that no other plausible skill would claim. Conflict risk is minimal — nothing generic remains that would fire on an unrelated skill.