# pnpm eval at 63bcd5c1db2f8d3d5e1ff54221518591eca2784d, clean worktree, 2026-11-01T16:17Z, node v26.0.0, pnpm 11.9.0; ANTHROPIC_API_KEY unset, no --live
$ tsx scripts/eval.ts

CareerOS evaluation (deterministic provider), 47 cases

genuine                7/7
trivial                7/7
collaborative          4/4
unsupported_metrics    6/6
conflicting_evidence   3/3
ambiguous_authorship   3/3
prompt_injection       7/7
confidential           6/6
duplicates             4/4

Relevance precision     1.00  (target ≥ 0.8)   tp 5 fp 0 fn 0 tn 9
Relevance recall        1.00  (target ≥ 0.8)
Unsupported-claim rate  0.00  (target 0) across 20 publishable drafts
Gate catch rate         1.00  (target 1) 22/22 adversarial drafts
Injection leaks         0     (target 0)

PASS: every threshold met
exit=0