Conciseness

The body is dense with non-obvious procedural knowledge ("n_enroll = ceil(n_analyzed / (1 − dropout_rate))", "The harness's alpha argument does not threshold returned p-values", "DEFF = 1 + (m − 1)·ICC is a planning approximation"), and it assumes domain competence rather than teaching basics. But minor over-explanation could be trimmed: the Overview's textbook statements ("Power analysis plans the probability of rejecting a specified null under an assumed alternative", "Four quantities are locked together for any given test") restate what Claude already knows, and the citation section's version-fetching instructions (~15 lines) are administrative rather than procedural. This fits anchor 4 ('efficient; minor instances of over-explanation that could be trimmed') better than 5 ('every token earns its place'). Not 3: there is no padded section and no tutorial-style hand-holding.

Actionability

Quotes: "from power import sample_size, power, mde, power_curve" with seven complete runnable calls; "est = simulate_power(gen_and_test, n=64, n_sims=2000, alpha=0.05, seed=0)"; the uv pip install commands with exact pins; and a copy-paste reporting template. Code is executable as written (the run-directory caveat is stated: "Run from skills/statistical-power/scripts/ or add that directory to sys.path"), and the common cases (two-group t, 3:1 allocation, ANOVA, proportions, correlation, power curve) are all covered. This matches anchor 5 ('fully executable; copy-paste ready code or commands; specific examples cover the common cases'). Not 4: no gaps remain — even edge conventions (signed effects for one-sided tests, integer-ratio recheck) are specified.

Workflow Clarity

Quotes: the 7-step workflow ("State the design... Choose the effect size... Set α and target power... Compute... Sensitivity analysis... Apply adjustments... Report") with explicit checkpoints and error-recovery loops: "First check Type I error under the null", "A noisy bisection search yields a candidate n: verify nearby sizes with more replicates and a fresh seed", "Recheck power using the realized integer ratio". The reporting template doubles as a completeness checklist. This matches anchor 5 ('clear sequence with explicit validation steps; feedback loops for error recovery'). Not 4: validation is not just mentioned but built into the loop (verify candidate n with fresh seeds). Not a destructive/batch skill, so no cap applies.

Progressive Disclosure

SKILL.md is an overview with quick recipes inline, and the bulk detail is split into three real one-level-deep files, each verified to exist and clearly signaled with purpose statements: "references/closed_form_recipes.md — per-test argument conventions and tested statsmodels/Pingouin calls", "references/simulation_based_power.md — full simulation patterns...", "references/effect_sizes.md — choosing effect sizes (SESOI)... ". The reference files do not chain further (only peer-level cross-references), and the Resources section lists every script and reference with a one-line description. This matches anchor 5 ('clear overview with well-signaled one-level-deep references; content appropriately split; easy navigation'). Not 4: there is no misplaced bulk content — full argument tables, conversion tables, and simulation patterns are all in the reference files rather than inlined.

Specificity

Quotes: "Calculates sample sizes and statistical power", "a minimum detectable effect (MDE), a power curve", "justify a sample size for a grant, IRB protocol, or pre-registration", "closed-form power for t-tests, ANOVA, proportions, correlations, chi-square, and regression, plus simulation-based (Monte Carlo) power for complex designs — logistic/Poisson regression, mixed models, cluster-randomized trials, survival, and interactions". Multiple specific concrete actions with comprehensive coverage of the domain; this matches the anchor-5 example's breadth ("Extract text and tables from PDF files, fill forms, merge documents, convert pages to images") with an even more detailed capability inventory. Not 4: there are no coverage gaps — both the closed-form and simulation halves of the domain are enumerated test-by-test.

Completeness

Quotes: what — "Calculates sample sizes and statistical power for study planning... Covers closed-form power... plus simulation-based (Monte Carlo) power"; when — "Applies when someone asks 'how many subjects/samples/replicates do I need'... Also handles requests that only mention an effect size, alpha, or '80% power'". Both what and when are explicit, with concrete trigger phrases, exactly matching the anchor-5 example's structure ("Extract text and tables... Use when working with PDF files or when the user mentions PDFs, forms, or document extraction"). Not 4: the 'when' clause is not merely present but enumerated across multiple explicit user phrasings.

Trigger Term Quality

Quotes: "how many subjects/samples/replicates do I need", "a priori power analysis", "minimum detectable effect (MDE)", "power curve", "80% power", "effect size", "grant, IRB protocol, or pre-registration". Comprehensive natural-language coverage including synonyms (subjects/samples/replicates) and partial-trigger phrasing users actually say; the explicit note that it also fires on requests that "only mention an effect size, alpha, or '80% power' without saying 'power analysis' explicitly" goes beyond even the anchor-5 example ("PDF files, PDFs, forms, document extraction, .pdf") by covering indirect triggers. Not 4: no common variation is missing.

Distinctiveness Conflict Risk

Quotes: "For laying out the study (randomization, blocking, factorial/DOE, crossover, sequential designs) use experimental-design; for analyzing data already collected and reporting it use statistical-analysis." The description carves out a clear niche (pre-data-collection power/sample-size planning) and explicitly routes the two adjacent tasks to sibling skills, so overlap risk is minimal. Not 4: it does not merely avoid overlap — it actively disambiguates against the two most confusable neighbors. Not below 5: no generic wording remains that would attract unrelated skills.