CAIN-42 CAIN StudioEvidence library · niche
Prediction, world models & simulation
Predicting consequences before acting, and never mistaking a guess for a fact. 417 tested invariants.
Last reviewed 2026-10-01
417 of 417 held
Test families in this niche
world_model (79) · knowledge_graph (43) · simulation (36) · epistemic (28) · inference (27) · truth (19) · blast_radius (16) · state_predictor (14) · truth_guard (12) · knowledge (10) · world_action (10) · model_swap (9) · observation (9) · causal_state (7) · moat (6) · fabric_component (6) · causal (5) · model_state (4) · research (3) · model_change (3) · operation_kind (2) · radar (2) · action (2) · graph (2) · risk (1) · reassessment (1) · cognitive (1) · prediction (1) · operation_state (1) · environment_query (1) · ip (1) · swarm (1) · lab (1) · agency_graph (1) · environment (1)
Where these come from
- CAIN-42 Evolution 40 -- Autonomous Enterprise Intelligence Fabric: 132
- CAIN-42 Evolution 35 -- Continuous Governance Intelligence Fabric: 87
- CAIN-42 Evolution 27 -- Agentic Internet Control Plane: 54
- CAIN-42 Evolution 42 -- Supreme Governed Agentic Infrastructure Platform: 23
- CAIN-42 Evolution 32 -- Governed Autonomy Learning Fabric: 20
- CAIN-42 Evolution 21 -- Governed Open-Ended Intelligence Fabric: 16
- CAIN-42 Evolution 33 -- Governed Agentic Operating Fabric: 16
- CAIN-42 Evolution 23 -- Governed Meta-Intelligence Fabric: 15
- CAIN-42 Evolution 34 -- Proof-Carrying Machine Agency: 12
- CAIN-42 Evolution 30 -- Governed Machine Autonomy Fabric: 10
- CAIN-42 Evolution 19 -- Governed Autonomy Operating Fabric: 7
- CAIN-42 Evolution 20 -- Governed Agentic Civilization Fabric: 7
- CAIN-42 Evolution 24 -- Governed Agentic Internet Fabric: 7
- CAIN-42 Evolution 36 -- Machine Agency Exchange Fabric: 6
- CAIN-42 Evolution 26 -- Universal Machine Agency Trust Fabric: 4
- CAIN-42 Evolution 28 -- Portable Machine Agency & Execution Identity: 1
Rules 1–150 of 417
| ID | Rule | Bundle | Result |
|---|---|---|---|
| G21 | simulation cannot become reality | E19 | held |
| G22 | counterfactual cannot become execution | E19 | held |
| G23 | prediction cannot become fact | E19 | held |
| G69 | legitimate model updates remain possible | E19 | held |
| G72 | simulation does not prove real-world safety | E19 | held |
| G79 | SIMULATED remains SIMULATED | E19 | held |
| G98 | public claims are truth-layer gated | E19 | held |
| I31 | simulation cannot execute | E20 | held |
| I32 | counterfactual cannot execute | E20 | held |
| I33 | market simulation cannot create real transactions | E20 | held |
| I34 | reputation simulation cannot create trust | E20 | held |
| I86 | the power engine is observational only | E20 | held |
| I87 | blast radius separates predicted, observed and unknown | E20 | held |
| I96 | the digital twin mirrors reality and never changes it | E20 | held |
| D16 | simulation remains simulation | E21 | held |
| D17 | counterfactual remains counterfactual | E21 | held |
| D28 | repeated identical simulations are not independent confirmation | E21 | held |
| D30 | source multiplication cannot create truth | E21 | held |
| D37 | model promotion requires governance | E21 | held |
| D47 | invalidated knowledge propagates | E21 | held |
| D49 | superseded models cannot silently remain authoritative | E21 | held |
| D68 | causal metadata remains distinct from private reasoning | E21 | held |
| D72 | simulated claims remain simulated | E21 | held |
| D79 | legitimate model diversity remains possible | E21 | held |
| D81 | knowledge does not become policy automatically | E21 | held |
| D94 | models are version-bound | E21 | held |
| D128 | causal levels cannot be skipped | E21 | held |
| D131 | the counterfactual lab cannot execute | E21 | held |
| D143 | models cannot regress against the incumbent | E21 | held |
| D144 | research bounties conserve simulated units | E21 | held |
| META-I018 | simulation ≠ deployment | E23 | held |
| META-I048 | model diversity does not imply independence | E23 | held |
| META-I056 | self-model answers all six self-knowledge questions | E23 | held |
| META-I057 | stale knowledge is reported STALE | E23 | held |
| META-I066 | verification in the reference architecture is independent of the model | E23 | held |
| META-P013 | every candidate's sandbox result is labelled SIMULATED (all depth-2 candidates) | E23 | held |
| META-L002 | META-002 SELF-MODELING IS NOT SELF-OWNERSHIP | E23 | held |
| META-L019 | META-019 HIGHER MODEL CONFIDENCE MUST NOT JUSTIFY GOVERNANCE BYPASS | E23 | held |
| META-B-stronger_model | legitimate mutation 'stronger_model' stays eligible (governance does not block improvement) | E23 | held |
| META-B-drop_simulator | legitimate mutation 'drop_simulator' stays eligible (governance does not block improvement) | E23 | held |
| META-S-self-what-i-predict | an assertion answering 'WHAT I PREDICT' lands in exactly that answer | E23 | held |
| META-S-unknown-unknown_model_behavior | the self-unknown engine surfaces unknown_model_behavior as a first-class state | E23 | held |
| META-S-unknown-unknown_physical_consequence | the self-unknown engine surfaces unknown_physical_consequence as a first-class state | E23 | held |
| META-S-canary-observed_gain_below_predicted | canary trigger OBSERVED_GAIN_BELOW_PREDICTED forces ROLLBACK | E23 | held |
| META-S-self-model-evidence | the self-model refuses and does not store unevidenced knowledge | E23 | held |
| NET-I052 | a legitimate escrow settles and is labelled SIMULATED | E24 | held |
| NET-T-medium-behavioral_trust | behavioral_trust UNKNOWN fails a medium-consequence action | E24 | held |
| NET-T-medium-security_trust | security_trust UNKNOWN fails a medium-consequence action | E24 | held |
| NET-T-high-behavioral_trust | behavioral_trust UNKNOWN fails a high-consequence action | E24 | held |
| NET-T-high-security_trust | security_trust UNKNOWN fails a high-consequence action | E24 | held |
| NET-T-high-execution_trust | execution_trust UNKNOWN fails a high-consequence action | E24 | held |
| NET-S-model-rollback | a rolled-back model is refused | E24 | held |
| I-BLAST-direct | direct reach | E26 | held |
| I-BLAST-transitive | transitive reach | E26 | held |
| I-BLAST-unknown | unknown reach not zero | E26 | held |
| I-REASSESS-model_change | trigger model_change | E26 | held |
| I-INFER-tokens-ok | grant tokens | E27 | held |
| I-INFER-tokens-over | over tokens refused | E27 | held |
| I-INFER-tokens-authority | compute adds no authority for tokens | E27 | held |
| I-INFER-latency_ms-ok | grant latency_ms | E27 | held |
| I-INFER-latency_ms-over | over latency_ms refused | E27 | held |
| I-INFER-latency_ms-authority | compute adds no authority for latency_ms | E27 | held |
| I-INFER-parallel_branches-ok | grant parallel_branches | E27 | held |
| I-INFER-parallel_branches-over | over parallel_branches refused | E27 | held |
| I-INFER-parallel_branches-authority | compute adds no authority for parallel_branches | E27 | held |
| I-INFER-reasoning_iterations-ok | grant reasoning_iterations | E27 | held |
| I-INFER-reasoning_iterations-over | over reasoning_iterations refused | E27 | held |
| I-INFER-reasoning_iterations-authority | compute adds no authority for reasoning_iterations | E27 | held |
| I-INFER-model_calls-ok | grant model_calls | E27 | held |
| I-INFER-model_calls-over | over model_calls refused | E27 | held |
| I-INFER-model_calls-authority | compute adds no authority for model_calls | E27 | held |
| I-INFER-verification_calls-ok | grant verification_calls | E27 | held |
| I-INFER-verification_calls-over | over verification_calls refused | E27 | held |
| I-INFER-verification_calls-authority | compute adds no authority for verification_calls | E27 | held |
| I-INFER-simulations-ok | grant simulations | E27 | held |
| I-INFER-simulations-over | over simulations refused | E27 | held |
| I-INFER-simulations-authority | compute adds no authority for simulations | E27 | held |
| I-INFER-tool_calls-ok | grant tool_calls | E27 | held |
| I-INFER-tool_calls-over | over tool_calls refused | E27 | held |
| I-INFER-tool_calls-authority | compute adds no authority for tool_calls | E27 | held |
| I-INFER-memory_retrievals-ok | grant memory_retrievals | E27 | held |
| I-INFER-memory_retrievals-over | over memory_retrievals refused | E27 | held |
| I-INFER-memory_retrievals-authority | compute adds no authority for memory_retrievals | E27 | held |
| I-COG-simulator | select simulator | E27 | held |
| I-TWIN-identity_compromise | twin identity_compromise | E27 | held |
| I-TWIN-partition | twin partition | E27 | held |
| I-TWIN-supply_chain | twin supply_chain | E27 | held |
| I-TWIN-collusion | twin collusion | E27 | held |
| I-TWIN-sybil | twin sybil | E27 | held |
| I-TWIN-runtime_swap | twin runtime_swap | E27 | held |
| I-TWIN-model_swap | twin model_swap | E27 | held |
| I-TWIN-policy_attack | twin policy_attack | E27 | held |
| I-KG-UNKNOWN | epistemic state UNKNOWN | E27 | held |
| I-KG-HYPOTHESIS | epistemic state HYPOTHESIS | E27 | held |
| I-KG-CLAIM | epistemic state CLAIM | E27 | held |
| I-KG-SUPPORTED | epistemic state SUPPORTED | E27 | held |
| I-KG-REPLICATED | epistemic state REPLICATED | E27 | held |
| I-KG-CONTESTED | epistemic state CONTESTED | E27 | held |
| I-KG-REFUTED | epistemic state REFUTED | E27 | held |
| I-KG-SUPERSEDED | epistemic state SUPERSEDED | E27 | held |
| I-KG-REVOKED | epistemic state REVOKED | E27 | held |
| I-KG-no-evidence | supported requires evidence | E27 | held |
| I-RESEARCH-LOOP-SIMULATION | loop SIMULATION | E27 | held |
| I-RESEARCH-LOOP-OBSERVATION | loop OBSERVATION | E27 | held |
| I-RESEARCH-LOOP-KNOWLEDGE_UPDATE | loop KNOWLEDGE_UPDATE | E27 | held |
| I-WM-prediction | prediction is not reality | E27 | held |
| I-WM-ledger | prediction error does not change policy | E27 | held |
| I-MODELCHANGE-QUARANTINE | decision QUARANTINE | E27 | held |
| I-MODELCHANGE-NEW_EXECUTION_IDENTITY_REQUIRED | decision NEW_EXECUTION_IDENTITY_REQUIRED | E27 | held |
| I-MODELCHANGE-CONTINUITY_APPROVED | decision CONTINUITY_APPROVED | E27 | held |
| E28-I85 | learning from prediction error never changes the policy | E28 | held |
| E30-ENV-browser | environment 'browser' is ENFORCED and treated so | E30 | held |
| E30-ENV-cloud | environment 'cloud' is ENFORCED and treated so | E30 | held |
| E30-ENV-database | environment 'database' is ENFORCED and treated so | E30 | held |
| E30-ENV-filesystem | environment 'filesystem' is ENFORCED and treated so | E30 | held |
| E30-ENV-laboratory | environment 'laboratory' is UNENFORCED and treated so | E30 | held |
| E30-ENV-physical_actuator | environment 'physical_actuator' is UNENFORCED and treated so | E30 | held |
| E30-ENV-robot | environment 'robot' is UNENFORCED and treated so | E30 | held |
| E30-ENV-simulated_world | environment 'simulated_world' is SIMULATED and treated so | E30 | held |
| E30-ENV-software | environment 'software' is ENFORCED and treated so | E30 | held |
| E30-ENV-vehicle | environment 'vehicle' is UNENFORCED and treated so | E30 | held |
| E32-MODEL-TRUSTED_FOR_PREDICTION | world model in TRUSTED_FOR_PREDICTION never yields authority | E32 | held |
| E32-MODEL-DEGRADED | world model in DEGRADED never yields authority | E32 | held |
| E32-MODEL-MODEL_UNTRUSTWORTHY | world model in MODEL_UNTRUSTWORTHY never yields authority | E32 | held |
| E32-MODEL-UNKNOWN | world model in UNKNOWN never yields authority | E32 | held |
| E32-CAUSAL-CORRELATION | CORRELATION does not silently become VERIFIED_CAUSAL | E32 | held |
| E32-CAUSAL-INTERVENTION_EVIDENCE | INTERVENTION_EVIDENCE does not silently become VERIFIED_CAUSAL | E32 | held |
| E32-CAUSAL-REFUTED | REFUTED does not silently become VERIFIED_CAUSAL | E32 | held |
| E32-CAUSAL-CAUSAL_HYPOTHESIS | CAUSAL_HYPOTHESIS does not silently become VERIFIED_CAUSAL | E32 | held |
| E32-CAUSAL-TEMPORAL_CORRELATION | TEMPORAL_CORRELATION does not silently become VERIFIED_CAUSAL | E32 | held |
| E32-CAUSAL-COUNTERFACTUAL_PREDICTION | COUNTERFACTUAL_PREDICTION does not silently become VERIFIED_CAUSAL | E32 | held |
| E32-CAUSAL-VERIFIED_CAUSAL | VERIFIED_CAUSAL needs two independent interventions | E32 | held |
| E32-SWAP-llm | llm swap never inherits authority | E32 | held |
| E32-SWAP-vlm | vlm swap never inherits authority | E32 | held |
| E32-SWAP-world_model | world_model swap never inherits authority | E32 | held |
| E32-SWAP-planner | planner swap never inherits authority | E32 | held |
| E32-SWAP-critic | critic swap never inherits authority | E32 | held |
| E32-SWAP-reasoning_model | reasoning_model swap never inherits authority | E32 | held |
| E32-SWAP-embedding_model | embedding_model swap never inherits authority | E32 | held |
| E32-SWAP-classifier | classifier swap never inherits authority | E32 | held |
| E32-SWAP-policy_model | policy_model swap never inherits authority | E32 | held |
| E33-STATE-SIMULATE | no state can be skipped after SIMULATE | E33 | held |
| E33-KIND-change_model | change_model is routed to E32 CAINSwapGovernance | E33 | held |
| E33-KIND-modify_world_model | modify_world_model is routed to E32 world model (observations only) | E33 | held |
| E33-BLAST-agents | blast radius reports agents separately (no aggregate) | E33 | held |
| E33-BLAST-identities | blast radius reports identities separately (no aggregate) | E33 | held |
| E33-BLAST-organizations | blast radius reports organizations separately (no aggregate) | E33 | held |
| E33-BLAST-capabilities | blast radius reports capabilities separately (no aggregate) | E33 | held |
| E33-BLAST-tools | blast radius reports tools separately (no aggregate) | E33 | held |
| E33-BLAST-credentials | blast radius reports credentials separately (no aggregate) | E33 | held |
Other niches
Consensus & distributed systems · Attacks, threats & containment · Identity, authority & delegation · Evidence, receipts & proofs · Memory, data & privacy · Transactions, markets & economics · Tools, MCP, protocols & adapters · Autonomy, control loops & recovery · Policy, law & governance · Trust & reputation · Supply chain, registry & lifecycle · Benchmarks, coverage & performance · Core guarantees
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