Conciseness

Efficient overall: every section carries instruction (context to gather, coaching rules, three phases, output format, tone) and it assumes Claude's competence, never explaining AI basics. A few rhetorical asides could be trimmed — "the one thing a three-minute web quiz structurally cannot do", "A number out of forty segments a mailing list" — which matches the 4 anchor ('minor instances of over-explanation that could be trimmed'). Not 5 because of those editorial lines; not 3 because no explanation is genuinely unnecessary — the rationale lines (e.g., why no score is given) guard against predictable agent behavior.

Actionability

Concrete, executable guidance for an instruction-only skill: the exact context items to request, the file to load ("Load references/traction-ladders.md"), phase-by-phase instructions with named patterns ("AI Theater, Pragmatic Builder, or Force Multiplier"), experiment constraints ("a leading indicator I can see inside the window, and a stopping rule"), and a six-field output format that is copy-paste ready. Not 5: no worked example of quoting evidence to place a rung, and question phrasing is left to judgment ("Choose what to ask next from what I just said") beyond one verbatim question. Not 3: guidance is far more complete than pseudocode-level hints.

Workflow Clarity

Clear three-phase sequence (place on ladders from evidence → name the pattern → one experiment), each with explicit completion criteria: "Stop when the binding constraint is clear", the verbatim sitting question in Phase 2, and the experiment's required indicator and stopping rule ("'We'll see how it goes' is not a stopping rule"). It includes an explicit verification loop — "Where what I told you and what the artifacts say disagree, name the gap out loud" — and the Output Format section acts as a final checklist. Not 4: no validation gaps are visible; checkpoints are explicit at every phase, including the global gate ("do not produce the final output until you have enough to produce it honestly").

Progressive Disclosure

The SKILL.md is a ~110-line process overview; the detailed rung definitions live in the one-level-deep, clearly signaled references/traction-ladders.md — referenced twice ("Load references/traction-ladders.md for the behavioural ladders" and in the References section with a description of its contents) — and the file exists with the promised content. Matches the 5 anchor: clear overview, well-signaled single reference, content appropriately split. Not 4: nothing that belongs in a separate file is inlined and navigation is trivial.

Specificity

The description lists multiple concrete actions — "Places the organisation on behavioural ladders across strategic focus, execution model, team integration and pragmatic mindset, names the binding constraint, and ends in one 2-4 week experiment" — which matches the 5 anchor's comprehensive action coverage. However, the rubric penalizes second-person voice by one point, and the text directly addresses the reader ("your AI work", "what you have actually shipped"), so the raw 5 drops to 4. Not 3: the action list is comprehensive, well beyond '1-2 concrete actions'.

Completeness

Explicitly answers both halves. What: "Find out whether your AI work is producing business traction or AI Theater, using evidence from what you have actually shipped", "Places the organisation on behavioural ladders... names the binding constraint, and ends in one 2-4 week experiment". When: an explicit "Use when there is a lot of AI activity and no defensible line to a business outcome, before an AI budget or steering-committee conversation, or when someone asks how the AI programme is going and the honest answer is a list of pilots", plus a routing boundary ("If the question is broad... run ai-activity-to-impact first"). Matches the 5 anchor exactly; 4 would require the 'when' to be less explicit or concrete than it is.

Trigger Term Quality

Good natural keyword coverage: "a lot of AI activity", "AI pilots with no clear return", "before an AI budget or steering-committee conversation", "how the AI programme is going", "honest answer is a list of pilots", "where do we start" — phrases a sponsor would actually say, with synonyms (activity/work/programme, pilots/initiatives). Not 5: common variations like "AI strategy", "AI ROI", or "is the AI investment paying off" are absent. Not 3: coverage clearly exceeds 'some relevant keywords' — several natural phrasings are present.

Distinctiveness Conflict Risk

Clear niche — separating business traction from "AI Theater" — with distinct triggers ("the honest answer is a list of pilots", "AI budget or steering-committee conversation") that no generic AI skill would match, and an explicit routing clause to the adjacent broad-triage skill (ai-activity-to-impact) that minimizes misfires. Not 4: the main overlap risk with closely related skills (sniff-test, outcome-framing-coach) is more than compensated for by the routing boundary and the distinctive 'AI Theater / traction' framing.