{
 "bundle": "governed-evolution-2026-10-07",
 "generated": "2026-10-07T19:19:49+00:00",
 "commit": "8a41fa19",
 "reproduce": "python3 scripts/governed_evolution/run_evidence.py --seed 5150 --red-team-seed 4242",
 "evidence_generated": "2026-10-07T19:00:32Z",
 "tests": {
  "tests": 127,
  "passed": 127,
  "failures": 0,
  "errors": 0,
  "skipped": 0
 },
 "benchmark": {
  "BASELINE": {
   "task_success": 0.3333,
   "security_violations": 0
  },
  "CAIN_GOVERNED_EVOLVING": {
   "task_success": 1.0,
   "security_violations": 0
  },
  "CAIN_GOVERNED_EVOLVING_ISOLATED": {
   "task_success": 1.0,
   "security_violations": 0
  },
  "UNGOVERNED_EVOLVING": {
   "task_success": 0.625,
   "security_violations": 36
  }
 },
 "governed_bad_skills_live": {
  "CAIN_GOVERNED_EVOLVING": 0,
  "CAIN_GOVERNED_EVOLVING_ISOLATED": 0
 },
 "reproducibility": 1.0,
 "isolated_regression_rate": 0.3333,
 "action_false_allows_governed": 0,
 "action_decision_latency_ms_p50": 1.218,
 "receipts": {
  "verified": 16,
  "of": 16,
  "verifier": "clean-room, no CAIN imports"
 },
 "red_team": {
  "BLOCKED": 411,
  "BREACH": 0,
  "ERROR": 0,
  "INFO": 0,
  "attempts": 411
 },
 "red_team_findings_fixed": [
  "RT-01 ToolDiscoveryGate first-word-only consequential check",
  "RT-02 sk_live_ credential reference accepted",
  "RT-03 full-width protected names passed the evolution gate",
  "RT-04 memory ledger could not see an edited/truncated tail",
  "BENCH-01 isolated-arm receipts verified against the wrong World's governance key (evidence script)"
 ],
 "token_cost": null,
 "honest_scope": [
  "synthetic, deterministic workload; the 'agent' proposals are a fixed labelled pool, not an LLM",
  "the labels are used only to score, never by the governed arm",
  "the ungoverned arm is scored against ground truth; none of its code runs unconfined",
  "a benchmark is not production reliability; n is small (8 proposals, 24 eval tasks, 40 actions)",
  "the BASELINE and CAIN arms share one action gate, so their action metrics are identical by design",
  "self-attested: produced and checked by the same project; no third-party review",
  "not wired into the hosted gateway; library + tests + reproducible evidence only"
 ],
 "independence": "same-project, self-verified; not third-party",
 "source_published": false
}