CAIN-42 CAIN StudioCAIN-42 evidence library
CAIN-42 Evolution 21 -- Governed Open-Ended Intelligence Fabric
Evidence bundle for evolution E21: 7 claims, 155 of 155 invariants held.
Last reviewed 2026-10-01
Claims (7)
| Claim | Level |
|---|---|
| C42-E21-OPEN-ENDED-INTELLIGENCE: CAIN-42 governs autonomous research -- questions, competing hypotheses, experiments, simulations, observations, replication, peer review, adversarial challenge, knowledge, causal claims, benchmarks, model and strategy evolution, research-agent creation and capability promotion -- so that discovery never silently becomes truth, authority or execution. Epistemic state is computed from signed, typed, independence-keyed evidence; research authority is disjoint from execution authority; a promoted capability is necessary but never sufficient, and every action still passes E20, the E19 action contract and the E8 kernel. | |
| C42-E21-LOOPHOLE: A research system that discovers a governance weakness can disclose it but cannot use the discovery to acquire permission to exploit it: every loophole attack in the bench is refused | |
| C42-E21-SCALE: deterministic in-process simulations of up to 100,000 virtual research agents and 100,000 simulated hypotheses | |
| C42-E21-SCIENTIFIC-TRUTH: any guarantee that a SUPPORTED or VERIFIED hypothesis is true, that a novelty is real beyond the supplied corpus, or that research produces correct science | |
| C42-E21-AGI: creating or discovering AGI, solving alignment, guaranteeing safe self-improvement, or controlling every autonomous system | |
| C42-E21-THIRD-PARTY: third-party review | |
| C42-E21-MULTI-HOST: multi-host or distributed behaviour of research state |
Invariants: 155 of 155 held
Rules the code must never break, each checked across many scenarios. See all 155 on one page.
The bundle's own README
CAIN-42 Evolution 21 -- Governed Open-Ended Intelligence Fabric#
Evidence measured at build time from the real E21 fabric, the real E20 institution gate, the real E19 gate and the real E8 kernel. Start with CAIN42_E21_EVIDENCE_BUNDLE.json (machine-readable summary), then INVARIANTS.json, ATTACK_MANIFEST.json, DISCOVERY_RESULTS.json (the research demonstration), SELF_EVOLUTION_RESULTS.json, HYPOTHESIS_RESULTS.json, EXPERIMENT_RESULTS.json, REPLICATION_RESULTS.json, KNOWLEDGE_RESULTS.json, CAUSAL_RESULTS.json, CAPABILITY_RESULTS.json, SCALE_RESULTS.json (SIMULATION), TAMPER_RESULTS.json and PUBLIC_SAFE_EXAMPLES.json. Verify everything with verify_e21.py.txt (see REPRODUCE.txt). Status: TESTED, pre-production, not third-party verified; the research problem is synthetic; limitations in LIMITATIONS.md, UNKNOWN.json and UNVERIFIED.json.
Try CAIN-42 on your own agents
Create a free account and every new account starts with a 7-day trial of the full platform. Or try the sandbox first, with no account at all.
Create a free account → · Try the sandbox · See the whole ecosystem