CAIN-42 CAIN StudioCAIN-42 evidence library
155 invariants
From CAIN-42 Evolution 21 -- Governed Open-Ended Intelligence Fabric.
Last reviewed 2026-10-01
| ID | Rule | Niche | Result |
|---|---|---|---|
| D1 | discovery cannot create authority | Identity, authority & delegation | held |
| D2 | hypothesis cannot create authority | Identity, authority & delegation | held |
| D3 | confidence cannot create authority | Identity, authority & delegation | held |
| D4 | model prediction cannot create authority | Identity, authority & delegation | held |
| D5 | simulation cannot create authority | Identity, authority & delegation | held |
| D6 | experiment result cannot create authority | Identity, authority & delegation | held |
| D7 | scientific consensus cannot create authority | Consensus & distributed systems | held |
| D8 | research reputation cannot create authority | Identity, authority & delegation | held |
| D9 | intelligence improvement cannot create authority | Identity, authority & delegation | held |
| D10 | benchmark performance cannot create authority | Identity, authority & delegation | held |
| D11 | unsupported claims remain unsupported | Core guarantees | held |
| D12 | contradictory evidence remains visible | Evidence, receipts & proofs | held |
| D13 | failed experiments remain auditable | Core guarantees | held |
| D14 | revoked evidence cannot support authorization | Identity, authority & delegation | held |
| D15 | stale evidence cannot silently authorize | Identity, authority & delegation | held |
| D16 | simulation remains simulation | Prediction, world models & simulation | held |
| D17 | counterfactual remains counterfactual | Prediction, world models & simulation | held |
| D18 | hypothetical remains hypothetical | Core guarantees | held |
| D19 | synthetic data remains synthetic | Memory, data & privacy | held |
| D20 | observed data remains observational | Memory, data & privacy | held |
| D21 | discovery provenance is immutable | Evidence, receipts & proofs | held |
| D22 | experiment provenance is immutable | Evidence, receipts & proofs | held |
| D23 | model provenance is immutable | Evidence, receipts & proofs | held |
| D24 | benchmark provenance is immutable | Evidence, receipts & proofs | held |
| D25 | knowledge lineage is reconstructable | Evidence, receipts & proofs | held |
| D26 | replication is independent where claimed | Consensus & distributed systems | held |
| D27 | consensus is not replication | Consensus & distributed systems | held |
| D28 | repeated identical simulations are not independent confirmation | Prediction, world models & simulation | held |
| D29 | correlated evidence cannot masquerade as independent evidence | Evidence, receipts & proofs | held |
| D30 | source multiplication cannot create truth | Prediction, world models & simulation | held |
| D31 | learning cannot silently change authority | Identity, authority & delegation | held |
| D32 | self-improvement cannot silently change authority | Identity, authority & delegation | held |
| D33 | agent spawning cannot silently change authority | Identity, authority & delegation | held |
| D34 | research success cannot silently change authority | Identity, authority & delegation | held |
| D35 | knowledge accumulation cannot silently change authority | Identity, authority & delegation | held |
| D36 | capability promotion requires governance | Identity, authority & delegation | held |
| D37 | model promotion requires governance | Prediction, world models & simulation | held |
| D38 | skill promotion requires governance | Autonomy, control loops & recovery | held |
| D39 | tool promotion requires governance | Tools, MCP, protocols & adapters | held |
| D40 | policy promotion requires governance | Autonomy, control loops & recovery | held |
| D41 | research budgets are bounded | Transactions, markets & economics | held |
| D42 | experiment budgets are bounded | Transactions, markets & economics | held |
| D43 | compute budgets are bounded | Transactions, markets & economics | held |
| D44 | agent creation budgets are bounded | Transactions, markets & economics | held |
| D45 | tool budgets are bounded | Transactions, markets & economics | held |
| D46 | failed research remains evidence | Evidence, receipts & proofs | held |
| D47 | invalidated knowledge propagates | Prediction, world models & simulation | held |
| D48 | revoked claims propagate | Identity, authority & delegation | held |
| D49 | superseded models cannot silently remain authoritative | Prediction, world models & simulation | held |
| D50 | benchmark changes invalidate incompatible comparisons | Benchmarks, coverage & performance | held |
| D51 | governance failures fail closed | Policy, law & governance | held |
| D52 | unknown does not become allow | Core guarantees | held |
| D53 | missing provenance does not become trusted | Evidence, receipts & proofs | held |
| D54 | missing evidence does not become verified | Evidence, receipts & proofs | held |
| D55 | missing replication does not become reproduced | Consensus & distributed systems | held |
| D56 | research authority is bounded | Identity, authority & delegation | held |
| D57 | experiment authority is bounded | Identity, authority & delegation | held |
| D58 | reviewer authority is bounded | Identity, authority & delegation | held |
| D59 | researcher authority is bounded | Identity, authority & delegation | held |
| D60 | institutional research authority is bounded | Identity, authority & delegation | held |
| D61 | no experiment bypasses E8 | Core guarantees | held |
| D62 | no capability bypasses E8 | Identity, authority & delegation | held |
| D63 | no discovery bypasses E8 | Tools, MCP, protocols & adapters | held |
| D64 | no research institution bypasses E8 | Core guarantees | held |
| D65 | no self-improvement bypasses E8 | Autonomy, control loops & recovery | held |
| D66 | no private reasoning is required for governance verification | Policy, law & governance | held |
| D67 | governance traces do not expose chain-of-thought | Evidence, receipts & proofs | held |
| D68 | causal metadata remains distinct from private reasoning | Prediction, world models & simulation | held |
| D69 | evidence is machine-verifiable | Evidence, receipts & proofs | held |
| D70 | clean-room verification remains independent | Core guarantees | held |
| D71 | public claims map to evidence | Evidence, receipts & proofs | held |
| D72 | simulated claims remain simulated | Prediction, world models & simulation | held |
| D73 | unverified claims remain unverified | Core guarantees | held |
| D74 | unknown claims remain unknown | Core guarantees | held |
| D75 | third-party verification is never implied without third-party verification | Core guarantees | held |
| D76 | autonomous research may continue without autonomous authority escalation | Identity, authority & delegation | held |
| D77 | legitimate experimentation remains possible | Core guarantees | held |
| D78 | legitimate scientific disagreement remains possible | Core guarantees | held |
| D79 | legitimate model diversity remains possible | Prediction, world models & simulation | held |
| D80 | legitimate hypothesis diversity remains possible | Core guarantees | held |
| D81 | knowledge does not become policy automatically | Prediction, world models & simulation | held |
| D82 | policy does not become authority automatically | Identity, authority & delegation | held |
| D83 | authority does not become execution automatically | Identity, authority & delegation | held |
| D84 | execution does not become success automatically | Core guarantees | held |
| D85 | success does not retroactively validate every assumption | Core guarantees | held |
| D86 | discovery loops cannot create infinite authority | Identity, authority & delegation | held |
| D87 | self-reference cannot create constitutional authority | Identity, authority & delegation | held |
| D88 | recursive agents cannot recursively amplify authority | Identity, authority & delegation | held |
| D89 | nested institutions cannot recursively amplify authority | Identity, authority & delegation | held |
| D90 | delegated research cannot exceed parent authority | Identity, authority & delegation | held |
| D91 | experiment environments are identity-bound | Identity, authority & delegation | held |
| D92 | simulation environments are provenance-bound | Evidence, receipts & proofs | held |
| D93 | benchmark environments are version-bound | Supply chain, registry & lifecycle | held |
| D94 | models are version-bound | Prediction, world models & simulation | held |
| D95 | datasets are version-bound | Memory, data & privacy | held |
| D96 | research state is replayable | Core guarantees | held |
| D97 | research state is auditable | Core guarantees | held |
| D98 | research state is hashable | Evidence, receipts & proofs | held |
| D99 | research state is independently verifiable | Core guarantees | held |
| D100 | evidence bundle is reproducible | Evidence, receipts & proofs | held |
| D101 | research authority is disjoint from execution authority | Identity, authority & delegation | held |
| D102 | authority functions take no epistemic input | Identity, authority & delegation | held |
| D103 | a claimed basis never changes an execution decision | Core guarantees | held |
| D104 | a promotion is necessary but never sufficient | Autonomy, control loops & recovery | held |
| D105 | promotion scope is an intersection | Autonomy, control loops & recovery | held |
| D106 | every E21 action reaches E8 through E20 and E19 | Core guarantees | held |
| D107 | a token never skips an E21 verdict | Core guarantees | held |
| D108 | E21 verdicts re-checked after token issue | Core guarantees | held |
| D109 | E8 tokens are single-use through E21 | Core guarantees | held |
| D110 | a governance weakness can only be disclosed | Policy, law & governance | held |
| D111 | no quorum can approve an exploit | Consensus & distributed systems | held |
| D112 | evidence classes are closed | Evidence, receipts & proofs | held |
| D113 | evidence is frozen after construction | Evidence, receipts & proofs | held |
| D114 | evidence producers must be live | Evidence, receipts & proofs | held |
| D115 | evidence cannot come from the future | Evidence, receipts & proofs | held |
| D116 | third-party evidence is labelled third-party | Evidence, receipts & proofs | held |
| D117 | reviewers are independent of researchers | Core guarantees | held |
| D118 | objections are resolved by evidence, not votes | Evidence, receipts & proofs | held |
| D119 | two independent reviews are required | Core guarantees | held |
| D120 | all five adversarial challenges precede support | Attacks, threats & containment | held |
| D121 | a broken adversarial challenge blocks support | Attacks, threats & containment | held |
| D122 | a researcher cannot confirm their own discovery | Tools, MCP, protocols & adapters | held |
| D123 | the discovery lifecycle has no shortcuts | Tools, MCP, protocols & adapters | held |
| D124 | each lifecycle stage has one artefact kind | Supply chain, registry & lifecycle | held |
| D125 | competing tracks have distinct controllers | Core guarantees | held |
| D126 | commit-reveal prevents copied answers | Core guarantees | held |
| D127 | blind replication stays blind | Consensus & distributed systems | held |
| D128 | causal levels cannot be skipped | Prediction, world models & simulation | held |
| D129 | correlation is never labelled causation | Core guarantees | held |
| D130 | verified causation needs replicated intervention | Consensus & distributed systems | held |
| D131 | the counterfactual lab cannot execute | Prediction, world models & simulation | held |
| D132 | experiments forbid governance bypass | Policy, law & governance | held |
| D133 | experiments inherit institutional constraints | Core guarantees | held |
| D134 | high-risk and field experiments need the quorum | Consensus & distributed systems | held |
| D135 | experiment approvals are single-use and expire | Identity, authority & delegation | held |
| D136 | information value is correct and is not permission | Identity, authority & delegation | held |
| D137 | curiosity is not authority | Identity, authority & delegation | held |
| D138 | data access is bound to mission and charter | Memory, data & privacy | held |
| D139 | intelligence is never reduced to one score | Attacks, threats & containment | held |
| D140 | a missing dimension is a regression | Benchmarks, coverage & performance | held |
| D141 | strategy stages cannot be skipped | Core guarantees | held |
| D142 | benchmarks detect selective reporting and partial metrics | Benchmarks, coverage & performance | held |
| D143 | models cannot regress against the incumbent | Prediction, world models & simulation | held |
| D144 | research bounties conserve simulated units | Prediction, world models & simulation | held |
| D145 | research bounties move no real money | Core guarantees | held |
| D146 | bounties pay only for supported discoveries | Core guarantees | held |
| D147 | hypothesis ranking is not authority | Identity, authority & delegation | held |
| D148 | the discovery flywheel has no terminal state | Tools, MCP, protocols & adapters | held |
| D149 | the E21 laws and constitution are present | Policy, law & governance | held |
| D150 | public E21 text avoids forbidden claims | Core guarantees | held |
| D151 | E21 is COMPLETE only when every gate holds | Core guarantees | held |
| D152 | search depth, branching and nodes are bounded | Core guarantees | held |
| D153 | budget renewal needs the quorum | Consensus & distributed systems | held |
| D154 | research roles are assigned within the assigner's authority | Identity, authority & delegation | held |
| D155 | a spawned research agent is evaluated before admission | Supply chain, registry & lifecycle | held |
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