Dataset Coverage Matrix

Cross-tabulation of all 69 cases in datasets/golden/*.yaml by plan tier, risk weight, expectation category, and seeded-defect linkage. Regenerate after adding cases so this stays a true picture of dataset breadth rather than a stale snapshot.

By plan tier

Plan	Cases	Share
bronze	10	14%
cross-plan	39	57%
gold	9	13%
silver	11	16%

By risk weight

Risk weight	Cases	Share
high	16	23%
medium	53	77%

By behavior

Behavior	Cases	Share
answer	58	84%
refuse	11	16%

By expectation category

A case may declare more than one category, so counts sum to more than 69.

Category	Cases	Share of dataset
amount	37	54%
coverage_decision	19	28%
evidence	47	68%
limits	7	10%
refusal	11	16%
tool_behavior	9	13%
uncertainty	5	7%

Seeded-defect linkage

8/8 seeded defects have at least one golden case with a matching defect_id (see docs/seeded-defects.md for the full catalog).

Defect	Cases
#1	`defect-1-bariatric-bronze-hallucination`, `defect-1-bariatric-bronze-hallucination-v2`
#2	`defect-2-silver-premium-stale`, `defect-2-silver-premium-stale-v2`
#3	`defect-3-silver-multihop-cost`, `defect-3-silver-multihop-cost-v2`
#4	`defect-4-urgent-care-contradiction`, `defect-4-urgent-care-contradiction-v2`
#5	`defect-5-tool-arg-transposition`, `defect-5-tool-arg-transposition-v2`
#6	`defect-6-refusal-breach`, `defect-6-refusal-breach-v2`
#7	`defect-7-prompt-injection`, `defect-7-prompt-injection-v2`
#8	`defect-8-pii-leakage`, `defect-8-pii-leakage-v2`

Regenerate: make dataset-matrix.