Conciseness

The body is dense and domain-specific, but the same content is repeated: the four modes are fully defined in the opening table and again in step 3's mode-specific bullets; the email-quality-auditor handoff is stated in the scope guard, the Skill Contract, and the Next Best Skill section; the never-authorizes-a-send boundary and the NEEDS_INPUT condition each appear three-plus times. Fits anchor 3 ("mostly efficient but could be tightened") rather than 4 because the duplication is systematic, not isolated; not 2 because there is no explanation of concepts Claude already knows and the domain policy content earns its tokens.

Actionability

Gives a copy-paste-ready keyless command ("python3 "${CLAUDE_PLUGIN_ROOT}/scripts/connectors/experiment.py" proportion --control <events> <n> --variant <events> <n> --alpha <alpha> --min-lift <relative-bar>"), a fill-in hypothesis template, a numeric sample-size table, and an explicit duration formula. Not 5: the `samplesize` and `continuous` subcommands are named without argument formats, and the referenced script cannot be verified in this bundle; not 3: the primary workflows are fully executable with concrete numbers.

Workflow Clarity

Nine numbered steps in a clear design→read-out sequence with explicit validation gates: stop-and-return NEEDS_INPUT when inputs are missing, the no-peeking rule fixing sample and read date at design time, the ordered underpowered fallback ladder in step 6 (widen MDE → run longer → fewer cells → don't test), receipt/hash binding before calculating, and the UNDECIDED decision rule absent a precommitted owner-approved rule. Not 4: checkpoints are explicit and include error-recovery paths, not implicit.

Progressive Disclosure

Sections are well-organized and references are clearly signaled in a Reference Materials list with one-line descriptions, but no bundle files exist in this skill directory while the body depends on ~10 files reached via multi-level relative paths into shared repo space and other skills' bundles ("../../../references/skill-contract.md", "../../nurture/email-sequence-designer/references/send-control.md", "../../../influencer/report/roi-calculator/SKILL.md"), none verifiable here. Fits anchor 3 (structure present, organization could be better) rather than 4 because the skill's discoverability depends on an external repo layout with cross-skill hops rather than a one-level-deep local bundle.

Specificity

Lists multiple concrete outputs — "produces a falsifiable hypothesis, one-variable-per-cell matrix, sample-size/MDE/duration/power plan, and an effect/uncertainty read from own ESP data" plus "Applies only a precommitted owner-approved action rule" — comprehensive and specific to this domain. Not 4: coverage of the skill's actions is complete rather than having minor gaps; not below since nothing is generic.

Completeness

Explicitly answers both: what ("produces a falsifiable hypothesis, one-variable-per-cell matrix, sample-size/MDE/duration/power plan, and an effect/uncertainty read") and when ("Use when the user asks to..." with concrete trigger phrases), and also states what it will not do ("Not for EQS/vetoes or writing the email"). Not 4: both what and when are explicit with concrete triggers, not merely present.

Trigger Term Quality

Five verbatim natural trigger phrases ("design an email A/B test", "set up a multivariate subject/CTA test", "run a send-time test", "build a hold-out group", "is this email result statistically and practically material?") plus bilingual keyword synonyms cover the ways users actually ask for this. Not 4: synonym coverage (multivariate, send-time, hold-out, significance/materiality) is comprehensive rather than missing common variants.

Distinctiveness Conflict Risk

Clear niche (email experiment design/significance read) with distinct triggers and explicit negative boundaries — "Not for EQS/vetoes or writing the email" — that separate it from sibling email-quality and creative-writing skills. Not 4: the explicit exclusion clause plus domain-specific triggers give minimal conflict risk.