The procedure (Steps 1-5) is efficient and specific, but conceptual framing sections ('A risk matrix records what a team believed...'), fully quoted ISTQB definitions with URLs repeated inline, and a Limitations section that restates earlier points ('Quiet rows are ambiguous by construction. This is restated here because...') could be tightened; it is above 1 because the padding is framing around genuinely domain-specific guidance, not explanation of basic concepts Claude lacks.

Gives concrete, copy-paste-ready thresholds ('Report at 2 or more points on a 1 to 5 scale', 'fewer than 5 defects as directional only', 'more defects than the median matrix row') and an exact four-part citation requirement (value, source, window, event count); per the instruction-only scoring note, absence of code is not penalized when guidance is this specific, so it clears the 'fully executable...copy-paste ready' anchor over the pseudocode anchor at 2.

Sequenced Steps 1-5 with sub-steps (3.1-3.3), explicit validation checkpoints (Step 3 thresholds gate reporting; Step 4 citation completeness gates reportability), and feedback loops ('If the row has near-zero defects and near-zero test executions...report it under 3.3 instead'); it exceeds the 2 anchor because checkpoints and error-recovery reclassification are explicit rather than implicit.

Holds the procedure inline while externalizing evidence to references/empirical-basis.md and a full example to references/worked-example.md, both real files clearly signaled with one-level-deep markdown links; it is above 2 because the references are clearly signaled and content that should be separate (evidence, worked example) is split out, not buried inline.

Lists multiple specific concrete actions in third person ('Checks an already-written risk matrix', 'Maps each row's likelihood rating to observed defect density, test failure rate and code churn', 'classifies each row as over-stated, under-stated, in-agreement, or not calibrated', 'emits candidate new entries'), matching the 'lists multiple specific concrete actions' anchor rather than the partial-coverage anchor at 2.

Explicitly answers what ('Checks an already-written risk matrix...classifies each row...') and when via an explicit 'Use when a matrix has been driving test decisions for at least three releases and nobody has yet checked...' trigger, satisfying the 'clearly answers both what AND when' anchor; it is above 2 because the when clause is explicit, not implied.

Covers natural terms a QA/test lead would say ('risk matrix', 'calibration', 'likelihood rating', 'impact rating', 'defects', 'escapes', 'incidents', 'defect density') plus an explicit natural trigger clause, matching the 'good coverage of natural terms' anchor; it is not the 2 anchor because common variations are present rather than missing.

Carves a clear niche with an explicit out-of-scope list ('choosing a scoring methodology, designing the matrix structure, picking risk categories...are all out of scope') and distinct after-the-fact triggers, making it unlikely to trigger for the authoring skill; it is above 2 because the boundary is explicit rather than merely 'somewhat specific'.