Conciseness

The guidance is operational and never explains concepts Claude already knows, but the same rules are restated several times: investigator-skip justification appears in the header note, 'When to skip an investigator', Troubleshooting, and the NEVER list; the code-graph-vs-raw-git mapping appears in both the header note and Step 2; 'Evidence before narrative' appears in both Operating Posture and Core Epistemics. This matches anchor 3 ('mostly efficient but... could be tightened') rather than 2 — the excess is repetition of skill-specific guidance, not background explanation or fluff.

Actionability

Fully executable throughout: exact git commands ('git blame -L <start>,<end> <file>', 'git log --follow -p -- <file>'), a copy-paste-ready 'gh pr view <number> --json title,body,author,createdAt,mergedAt,labels,closingIssuesReferences,comments,reviews', tokensave tool signatures, explicit subagent config (subagent_type, model), a numbered per-investigator input checklist, and a worked Sources Consulted example. Anchor 5 — specific commands cover the common cases.

Workflow Clarity

Steps 1–5 are clearly sequenced, and a dedicated Verification section supplies an explicit validation checkpoint ('spot-check at least one citation... Run this the same way you'd run a test before shipping'), with error-recovery feedback in Troubleshooting and Common Failure Modes (each with a WHY). Anchor 5: clear sequence, explicit validation, feedback loops.

Progressive Disclosure

All cited references exist and are one level deep, clearly signaled both inline and in the Reference Files section, with references/source-playbook.md indexing references/sources/*.md — good structure matching anchor 4. Not anchor 5: the ~370-line body itself carries substantial detail (the full investigator roster, a summary of the epistemics framework that duplicates references/epistemics.md, and the long porting/precedence header note) that is closer to reference material than overview.

Specificity

Lists multiple concrete third-person actions — 'Discovers available MCPs and queries each evidence category (source control, issue tracker, long-form docs, real-time chat, infrastructure observability, error tracking, product analytics warehouse) in parallel, then returns a cited read on decisions and tradeoffs' — with comprehensive coverage of the mechanism. Anchor 5 fits better than 4 because no capability area is missing or generic.

Completeness

Explicitly answers both: what it does ('Discovers available MCPs and queries each evidence category... in parallel, then returns a cited read on decisions and tradeoffs') and when to use it ('Use for...' plus a concrete 'Triggers:' list), and even adds explicit negative triggers ('DO NOT use for... use how instead... use blast-radius instead'). Clearly anchor 5, not 4 — the 'when' is as explicit and specific as the 'what'.

Trigger Term Quality

'Triggers: why does X work this way, why did we pick Y over Z, what motivated this, why does this still exist, history of X' covers the natural phrasings users would actually say, plus synonyms in the opening ('design rationale, regressions, postmortems, or data-backed thresholds'). Comprehensive synonym coverage matches anchor 5; nothing common is missing.

Distinctiveness Conflict Risk

It carves a clear niche (motivation/rationale archaeology) and actively disambiguates from adjacent skills: 'DO NOT use for explaining current runtime behavior (use how instead), or for proving a change is safe to ship (use blast-radius instead)'. Minimal conflict risk — anchor 5, not 4, because the sibling-skill boundaries are stated rather than merely implied.