C1: Create machine-readable benchmark cards

[cschanhniem]

Bottleneck

Benchmarks are documented as markdown but lack machine-readable metadata. Governance cannot be enforced automatically.

Task

Create benchmark card YAML/JSON files under configs/benchmark_cards/ with fields: name, type (retrospective/ablation/regression), dataset, baseline_comparison, leakage_status, last_updated, auroc, ci_95.

Acceptance criteria

• Each active benchmark has a card
• Card schema is documented
• 2+ tests validate card structure
• make bench-cards-check target

Source

NEXT_100_PR_MAP, C1

[cschanhniem]

Already completed in prior loops.