{
 "schema": "cain.e91.acceptance.v1",
 "evolution": "E91",
 "release": "RELEASED — PARTIAL EVIDENCE",
 "rule": "PASS only with executed experiment + raw evidence; code/unit test alone ≤ PARTIAL; uncompleted durations BLOCKED",
 "counts": {
  "PASS": 18,
  "PARTIAL": 6
 },
 "gates": [
  {
   "gate": 1,
   "name": "Previous evolution baseline preserved",
   "status": "PASS",
   "owner": "ec",
   "cross_auditor": "1b",
   "evidence": [
    "git: evolution-88 f2d12e8b6, evolution-89 081723055, evolution-90 2d444e707 unchanged; E90 merged no-ff (43f297add)"
   ],
   "explanation": "E91 worked in its own worktree and branch. E88/E89/E90 commits are untouched; E90 was merged, not rewritten."
  },
  {
   "gate": 2,
   "name": "Previous evolution baseline reproduced where feasible",
   "status": "PARTIAL",
   "owner": "ec",
   "cross_auditor": "1b",
   "evidence": [
    "artifacts/e90/baseline/e88_tests.txt (E90 tree)",
    "artifacts/e90/baseline/e88_learning_repro/learning.json",
    "tests/test_e90_*.py in E91 tree"
   ],
   "explanation": "E88: 94/94 tests and its learning experiment reproduced byte-identically. E90: its 25 tests pass in the E91 tree, but its benchmark was not re-run by E91. E89's CL-Bench runs are cited, not re-run (they need paid frontier calls)."
  },
  {
   "gate": 3,
   "name": "Research review completed",
   "status": "PARTIAL",
   "owner": "ec",
   "cross_auditor": "1b",
   "evidence": [
    "research/E91_RESEARCH_REVIEW.md",
    "research/E91_BENCHMARK_SELECTION.md",
    "research/E91_RESEARCH_TO_IMPLEMENTATION_MATRIX.md"
   ],
   "explanation": "All required topics are covered, but from abstracts, landing pages and search summaries only. No full paper text was read, and the review says so per source."
  },
  {
   "gate": 4,
   "name": "Intelligence taxonomy defined",
   "status": "PASS",
   "owner": "ec",
   "cross_auditor": "1b",
   "evidence": [
    "research/E91_VERTICAL_INTELLIGENCE_TAXONOMY.md"
   ],
   "explanation": "15 capabilities, each with definition, baseline, test, success/failure criteria, verification, evidence and limits."
  },
  {
   "gate": 5,
   "name": "Experimental harness operational",
   "status": "PASS",
   "owner": "ec",
   "cross_auditor": "1b",
   "evidence": [
    "artifacts/e91/results/haiku/rows_heldout_*.jsonl"
   ],
   "explanation": "Executed end to end: 1120 held-out rows across arms A–G on Haiku 5.5, every tool call through a single-use grant."
  },
  {
   "gate": 6,
   "name": "Baselines established",
   "status": "PASS",
   "owner": "ec",
   "cross_auditor": "1b",
   "evidence": [
    "benchmarks/e91/BASELINE_RESULTS.json"
   ],
   "explanation": "A, B, C, D on the same 160 held-out tasks: A 82/160; B 85/160; C 90/160; D 114/160."
  },
  {
   "gate": 7,
   "name": "Six vertical environments implemented",
   "status": "PASS",
   "owner": "ec",
   "cross_auditor": "1b",
   "evidence": [
    "tests/test_e91_tasks.py",
    "benchmarks/e91/HELD_OUT_TASK_MANIFEST.json"
   ],
   "explanation": "16 families across software, security, science, business, engineering and education, with deterministic checkers and governed tools. Constructed by this project, not external benchmarks."
  },
  {
   "gate": 8,
   "name": "Held-out tasks created",
   "status": "PASS",
   "owner": "ec",
   "cross_auditor": "1b",
   "evidence": [
    "benchmarks/e91/HELD_OUT_TASK_MANIFEST.json"
   ],
   "explanation": "160 held-out tasks (≥20 per vertical) from a disjoint seed range, manifest committed (7e78c1cf1) before any held-out row."
  },
  {
   "gate": 9,
   "name": "Experience-learning lifecycle implemented",
   "status": "PASS",
   "owner": "ec",
   "cross_auditor": "1b",
   "evidence": [
    "artifacts/e91/learning/haiku/promotion.json",
    "artifacts/e91/learning/haiku/experience.db"
   ],
   "explanation": "Executed: dev experience → checker verification → hypothesis → val validation → approver-signed promotion or rejection → condition-gated held-out use. Weights are not trained; procedures come from a designer-supplied repertoire."
  },
  {
   "gate": 10,
   "name": "Learning performance compared against baselines",
   "status": "PASS",
   "owner": "ec",
   "cross_auditor": "1b",
   "evidence": [
    "artifacts/e91/analysis/core_analysis_claude-haiku-5-5.json"
   ],
   "explanation": "G 153/160 vs D 114/160, exact McNemar p=0.0. Pre-registered H3 holds: the gain is entirely in families with a promoted procedure (70→109); none elsewhere (44→44)."
  },
  {
   "gate": 11,
   "name": "Causal reasoning experiments completed",
   "status": "PASS",
   "owner": "1b",
   "cross_auditor": "ec",
   "evidence": [
    "artifacts/e91/causal/causal_suite_heldout.json",
    "artifacts/e91/causal/causal_suite_heldout_with_outputs.json"
   ],
   "explanation": "Held-out 40/40, reproduced independently by ec (re-execution + checker). Interventions, replication, randomised-design detection, INSUFFICIENT_EVIDENCE. Designer-written procedure, not model reasoning."
  },
  {
   "gate": 12,
   "name": "Adaptive reasoning evaluated",
   "status": "PARTIAL",
   "owner": "1b",
   "cross_auditor": "ec",
   "evidence": [
    "artifacts/e91/results/haiku/rows_heldout_H.jsonl",
    "tests/test_e91_adaptive.py"
   ],
   "explanation": "Arm H (evidence-triggered escalation 0.5B→Haiku vs static routing): 18 held-out rows recorded at matrix build time. Run not complete, so PARTIAL."
  },
  {
   "gate": 13,
   "name": "Multi-agent collaboration compared against single-agent baselines",
   "status": "PASS",
   "owner": "1b",
   "cross_auditor": "ec",
   "evidence": [
    "artifacts/e91/team/team_summary_haiku.json",
    "artifacts/e91/team/team_summary_r2_haiku.json",
    "artifacts/e91/results/haiku/rows_heldout_team-*.jsonl",
    "artifacts/e91/results/haiku/rows_heldout_team_r2-*.jsonl"
   ],
   "explanation": "7 configurations executed on 20 held-out code tasks with Haiku 5.5: single 18/20 (20 calls); three_identical 19/20 (62 calls); three_diverse 20/20 (62 calls); specialists 20/20 (47 calls); specialist_verifier 20/20 (23 calls); hierarchical 20/20 (40 calls); governed_swarm 20/20 (20 calls). Run 1: near ceiling, no significant gain. Run 2 (fault_localization + allocation, shared tool budget, families chosen from dev data and pre-registered): single 10/20; three_identical 9/20; three_diverse 5/20; specialists 3/20; specialist_verifier 8/20; hierarchical 7/20; governed_swarm 7/20. No team beat the single agent; verifier probes consumed the shared budget, and false acceptances/consensus occurred. PASS = the comparison was executed; the result is negative."
  },
  {
   "gate": 14,
   "name": "Long-horizon mission persistence tested",
   "status": "PARTIAL",
   "owner": "1b",
   "cross_auditor": "ec",
   "evidence": [
    "artifacts/e91/mission/run_1h/mission_report.json"
   ],
   "explanation": "Persistence through injected faults executed: 59.4 min, 22 faults, 7200 tasks verified, 0 duplicates. PARTIAL because the run is 3,561 s (< 1 h), and 6/24/72 h runs were not done."
  },
  {
   "gate": 15,
   "name": "Cross-domain transfer evaluated",
   "status": "PASS",
   "owner": "ec",
   "cross_auditor": "1b",
   "evidence": [
    "artifacts/e91/transfer/haiku/transfer_learning.json"
   ],
   "explanation": "T1 software.fault_localization→{'gate_off': {'accepted': 10, 'baseline_D': 5, 'delta': 5, 'n': 10, 'refused': 0, 'transferred': 10}, 'gate_on': {'accepted': 10, 'baseline_D': 5, 'delta': 5, 'n': 10, 'refused': 0, 'transferred': 10}}: D 5/10 → 10/10; negative control gate on 10/10 (refused 10), gate off 2/10. T2 science.hypothesis_test→{'gate_off': {'accepted': 10, 'baseline_D': 7, 'delta': 3, 'n': 10, 'refused': 0, 'transferred': 10}, 'gate_on': {'accepted': 10, 'baseline_D': 7, 'delta': 3, 'n': 10, 'refused': 0, 'transferred': 10}}: D 7/10 → 10/10; negative control gate on 3/10 (refused 10), gate off 0/10. The abstract tool interface is designer-supplied."
  },
  {
   "gate": 16,
   "name": "Scientific research lifecycle evaluated",
   "status": "PARTIAL",
   "owner": "1b",
   "cross_auditor": "ec",
   "evidence": [
    "docs/evolution-91/E91_SCIENTIFIC_RESEARCH_REPORT.md",
    "artifacts/e91/causal/causal_suite_heldout.json"
   ],
   "explanation": "Hypothesis rejection, underpowered abstention, controlled experiments, replication and fabricated-citation checks were executed. Autonomous question proposal and literature retrieval over real sources were not."
  },
  {
   "gate": 17,
   "name": "Adversarial intelligence security evaluated",
   "status": "PASS",
   "owner": "1b",
   "cross_auditor": "ec",
   "evidence": [
    "artifacts/e91/security/lab.json",
    "artifacts/e91/transfer/haiku/transfer_learning.json#poisoning"
   ],
   "explanation": "E88 authorization lab 26/26, E91 scenarios 15/15 blocked; forged experience 90/90 rejected. Same-project lab; known weaknesses listed."
  },
  {
   "gate": 18,
   "name": "Governance and kill-switch behavior verified",
   "status": "PASS",
   "owner": "1b",
   "cross_auditor": "ec",
   "evidence": [
    "artifacts/e91/mission/active_kill/mission_report.json"
   ],
   "explanation": "Active kill during execution: effective in 0.15 ms, 0 effects after the fence, 3 commits refused."
  },
  {
   "gate": 19,
   "name": "ClawX evidence records created",
   "status": "PASS",
   "owner": "1b",
   "cross_auditor": "ec",
   "evidence": [
    "docs/evolution-91/E91_CLAWX_EVIDENCE_REPORT.md",
    "artifacts/e91/live_publication.json"
   ],
   "explanation": "Signed static evidence bundle published under clawx.click/evidence/ and the cainstudio proof mirror; live checks recorded. E91 missions are still not written into the ClawX API data model (static evidence publication only)."
  },
  {
   "gate": 20,
   "name": "Evidence bundles independently verifiable offline",
   "status": "PASS",
   "owner": "1b",
   "cross_auditor": "ec",
   "evidence": [
    "artifacts/e91/public/verify_e91_offline.py"
   ],
   "explanation": "Standalone verifier (stdlib + cryptography, no CAIN imports) passes from a clean directory and detects tampering. Its signing key is self-generated: this proves integrity, not independent witnessing."
  },
  {
   "gate": 21,
   "name": "Product integration accurately documented",
   "status": "PARTIAL",
   "owner": "ec",
   "cross_auditor": "1b",
   "evidence": [
    "docs/evolution-91/E91_LIMITATIONS.md"
   ],
   "explanation": "Documented accurately as NOT integrated into live MCPGate, CAIN Studio or the production ClawX service; governance runs through E88 Authority/FencedResource in-process."
  },
  {
   "gate": 22,
   "name": "Public evidence publication verified",
   "status": "PASS",
   "owner": "ec",
   "cross_auditor": "1b",
   "evidence": [
    "artifacts/e91/live_publication.json"
   ],
   "explanation": "Live checks: {'schema': 'cain.e91.live_publication.v1', 'checked_at': '2026-11-18T05:15:00Z', 'clawx_url': 'https://clawx.click/evidence/e91-vertical-intelligence-2026-11-18/', 'cainstudio_url': 'https://cainstudio.online/proof/bundle/e91-vertical-intelligence-2026-11-18/', 'clawx_ok': True, 'cainstudio_ok': True, 'bundle_checks': 91, 'bundle_passed': 91, 'tamper_detected': True, 'verifier': 'scripts/evolution91/verify_e91_offline.py (stdlib + cryptography; imports no CAIN)', 'independent_witness': False, 'note': 'Live downloads from both hosts verified with the standalone verifier; the signing key is self-generated.'}"
  },
  {
   "gate": 23,
   "name": "Resource consumption reported",
   "status": "PASS",
   "owner": "ec",
   "cross_auditor": "1b",
   "evidence": [
    "docs/evolution-91/E91_RESOURCE_COST_REPORT.md",
    "artifacts/e91/paid/ledger.jsonl"
   ],
   "explanation": "Paid ledger (tokens, cost) and per-call local logs reported."
  },
  {
   "gate": 24,
   "name": "Release readiness honestly assessed",
   "status": "PASS",
   "owner": "ec",
   "cross_auditor": "1b",
   "evidence": [
    "docs/evolution-91/E91_EXECUTIVE_REPORT.md"
   ],
   "explanation": "Assessed: NOT RELEASED (constructed tasks, self-run, no independent witness, no production integration)."
  }
 ]
}