Conciseness

The body is dense and decision-oriented — tables, formulas, exact commands, no tutorial padding — and the high-value pitfalls (log-transform ratios, the scale foot-gun, CI→SE derivation) earn every token. However, the I² interpretation table, the fixed-vs-random effects table, and the tau_squared explanation partly restate standard meta-analysis knowledge Claude already has; trimming these would save tokens without losing skill-specific value.

Actionability

Everything is executable: exact conversion formulas per input type ('(ln(U) − ln(L)) / (2 × 1.96)'), a copy-paste `tu run MetaAnalysis_run` invocation with a complete JSON studies array, the helper script's CSV column schemas, and a worked reporting example with concrete numbers. A user can go from raw paper numbers to a pooled estimate without inventing any step.

Workflow Clarity

A 5-step workflow diagram is stated up front and each step is then elaborated in order, with explicit validation checkpoints: input rules (log-transform, 'one direction', 'same effect measure'), a retraction check before trusting studies, error-recovery guidance ('flip sign / invert the ratio for studies coded the opposite way'), interpretation thresholds for reading results, and an ordered reporting checklist. Not a destructive or unvalidated batch operation — input verification and output interpretation are both built in.

Progressive Disclosure

The bundle is one helper script (verified present at scripts/meta_analysis.py, and its behavior matches the body's claims: ratio log-transformation with back-transformed output, forest plot, publication-bias note). The body is well-sectioned per workflow step with a summary diagram for navigation, the script is well-signaled with its usage inlined, and nothing at this size warrants splitting into reference files — the conversion table that could be a reference is the core error-prone step and belongs inline.

Specificity

Multiple concrete actions with comprehensive coverage: 'pool effect sizes across studies … with fixed- or random-effects models, quantify heterogeneity (Q, I², τ²), and build a forest plot', plus a full enumeration of handled effect measures (odds ratios, risk ratios, hazard ratios, mean differences, correlations, GWAS betas) and the input-conversion capability. This matches the anchor-5 example's breadth with no coverage gaps in the domain.

Completeness

Both what and when are explicit: the what is a complete capability list (pooling, heterogeneity quantification, forest plot, effect-size/SE preparation), and the when uses a concrete trigger clause — 'Use when you have results from MULTIPLE studies and need a single pooled estimate, or to synthesize evidence from a systematic review / multiple GWAS / replicated experiments'. This mirrors the anchor-5 exemplar structure.

Trigger Term Quality

Natural user vocabulary is comprehensively covered with synonyms: 'meta-analysis', 'evidence synthesis', 'pool effect sizes', 'systematic review', 'GWAS', 'replicated experiments', 'forest plot', plus specific measure names (odds ratios, hazard ratios) users would cite. These are exactly the phrases someone needing this skill would say.

Distinctiveness Conflict Risk

Clear niche — statistical pooling of multiple studies — with a distinctive trigger (MULTIPLE studies, pooled estimate, systematic review) that separates it from single-study statistics, literature-search, or GWAS-analysis siblings. The 'replicated experiments' and 'multiple cohorts' framing makes mis-triggering on a single-study request unlikely.