Evidence library · niche
Autonomy, control loops & recovery
Agents acting on their own, and how CAIN keeps them in bounds and recovers. 407 tested invariants.
Last reviewed 2026-10-01
407 of 407 held
Test families in this niche
evolution (56) · loop (44) · autonomy (25) · drift (25) · learning_loop (17) · promotion_dominance (15) · self_healing (15) · self_test (13) · degradation (13) · self_governance (12) · autonomy_gradient (11) · self_improvement (10) · promotion_stage (10) · self_repair (10) · degraded_mode (10) · replan (10) · continuity (9) · autonomy_level (9) · research (7) · autopilot (6) · recovery (6) · fabric_component (4) · swe (3) · state_machine (2) · integrity (2) · profile (2) · risk (1) · scientist (1) · operation_state (1) · gap_detector (1) · plan_check_act (1) · trajectory (1) · radar (1) · swarm (1) · continuity_token (1) · canary (1) · interrupt (1) · promotion (1) · replanning (1) · rollback (1) · systemic (1) · graph (1)
Where these come from
- CAIN-42 Evolution 35 -- Continuous Governance Intelligence Fabric: 86
- CAIN-42 Evolution 40 -- Autonomous Enterprise Intelligence Fabric: 70
- CAIN-42 Evolution 32 -- Governed Autonomy Learning Fabric: 66
- CAIN-42 Evolution 27 -- Agentic Internet Control Plane: 60
- CAIN-42 Evolution 23 -- Governed Meta-Intelligence Fabric: 31
- CAIN-42 Evolution 30 -- Governed Machine Autonomy Fabric: 21
- CAIN-42 Evolution 33 -- Governed Agentic Operating Fabric: 18
- CAIN-42 Evolution 42 -- Supreme Governed Agentic Infrastructure Platform: 15
- CAIN-42 Evolution 28 -- Portable Machine Agency & Execution Identity: 10
- CAIN-42 Evolution 36 -- Machine Agency Exchange Fabric: 6
- CAIN-42 Evolution 39 -- Governed Agent Factory: 6
- CAIN-42 Evolution 20 -- Governed Agentic Civilization Fabric: 5
- CAIN-42 Evolution 21 -- Governed Open-Ended Intelligence Fabric: 5
- CAIN-42 Evolution 19 -- Governed Autonomy Operating Fabric: 3
- CAIN-42 Evolution 37 -- Autonomous Execution Mesh: 3
- CAIN-42 Evolution 24 -- Governed Agentic Internet Fabric: 1
- CAIN-42 Evolution 41 -- Machine Agency Exchange Fabric: 1
Rules 1–150 of 407
| ID | Rule | Bundle | Result |
|---|---|---|---|
| G49 | checkpoint rollback cannot bypass governance | E19 | held |
| G67 | legitimate autonomy is not treated as replay | E19 | held |
| G70 | legitimate recovery remains possible | E19 | held |
| I46 | institutional autonomy is bounded | E20 | held |
| I49 | institutional evolution is governed | E20 | held |
| I66 | legitimate recovery restores operation | E20 | held |
| I97 | the institutional control loop is ordered | E20 | held |
| I101 | assessment never raises autonomy | E20 | held |
| D38 | skill promotion requires governance | E21 | held |
| D40 | policy promotion requires governance | E21 | held |
| D65 | no self-improvement bypasses E8 | E21 | held |
| D104 | a promotion is necessary but never sufficient | E21 | held |
| D105 | promotion scope is an intersection | E21 | held |
| META-I019 | canary ≠ production | E23 | held |
| META-I021 | unknown self-state cannot become allow | E23 | held |
| META-I040 | rollback cannot erase audit history | E23 | held |
| META-I051 | the legitimate promotion path commits through E8 | E23 | held |
| META-I052 | promotion never changes the governance ceiling | E23 | held |
| META-I063 | a governed rollback restores the architecture and revokes the failed one | E23 | held |
| META-I076 | recovery requires verification | E23 | held |
| META-L015 | META-015 SELF-OPTIMIZATION MUST NOT REMOVE GOVERNANCE CHECKS | E23 | held |
| META-L021 | META-021 UNKNOWN SELF-MODIFICATION STATE MUST NEVER BECOME ALLOW | E23 | held |
| META-L022 | META-022 UNVERIFIED EVOLUTION MUST NEVER BECOME PRODUCTION | E23 | held |
| META-L023 | META-023 FAILED EVOLUTION MUST NEVER BECOME THE ACTIVE GOVERNANCE PATH | E23 | held |
| META-S-dim-rollbackability | governability dimension 'rollbackability' drops when its property is violated | E23 | held |
| META-S-dim-failure_recovery | governability dimension 'failure_recovery' drops when its property is violated | E23 | held |
| META-S-registry-experimental | registry state EXPERIMENTAL has no self-loop and PRODUCTION only via CANARY | E23 | held |
| META-S-registry-sandbox | registry state SANDBOX has no self-loop and PRODUCTION only via CANARY | E23 | held |
| META-S-registry-verified | registry state VERIFIED has no self-loop and PRODUCTION only via CANARY | E23 | held |
| META-S-registry-canary | registry state CANARY has no self-loop and PRODUCTION only via CANARY | E23 | held |
| META-S-registry-production | registry state PRODUCTION has no self-loop and PRODUCTION only via CANARY | E23 | held |
| META-S-registry-deprecated | registry state DEPRECATED has no self-loop and PRODUCTION only via CANARY | E23 | held |
| META-S-unknown-unknown_dependency | the self-unknown engine surfaces unknown_dependency as a first-class state | E23 | held |
| META-S-unknown-unknown_execution_path | the self-unknown engine surfaces unknown_execution_path as a first-class state | E23 | held |
| META-S-unknown-unknown_policy_interaction | the self-unknown engine surfaces unknown_policy_interaction as a first-class state | E23 | held |
| META-S-unknown-unknown_failure_mode | the self-unknown engine surfaces unknown_failure_mode as a first-class state | E23 | held |
| META-S-unknown-unknown_external_dependency | the self-unknown engine surfaces unknown_external_dependency as a first-class state | E23 | held |
| META-S-unknown-unknown_security_boundary | the self-unknown engine surfaces unknown_security_boundary as a first-class state | E23 | held |
| META-S-unknown-unknown_third_party_component | the self-unknown engine surfaces unknown_third_party_component as a first-class state | E23 | held |
| META-S-mode-verification_degraded | mode VERIFICATION_DEGRADED removes consequential capabilities | E23 | held |
| META-S-mode-degraded | mode DEGRADED removes consequential capabilities | E23 | held |
| META-S-mode-halted | mode HALTED removes consequential capabilities | E23 | held |
| META-S-canary-canary_worse_than_production | canary trigger CANARY_WORSE_THAN_PRODUCTION forces ROLLBACK | E23 | held |
| META-S-canary-canary_governance_violation | canary trigger CANARY_GOVERNANCE_VIOLATION forces ROLLBACK | E23 | held |
| NET-L17 | NET-L17 RECOVERY IS NOT TRUST RESTORATION | E24 | held |
| I-ABUDGET-max_duration_ms-ok | spend max_duration_ms | E27 | held |
| I-ABUDGET-max_duration_ms-exhaust | exhaust max_duration_ms reauthorizes | E27 | held |
| I-ABUDGET-max_spend_minor-ok | spend max_spend_minor | E27 | held |
| I-ABUDGET-max_spend_minor-exhaust | exhaust max_spend_minor reauthorizes | E27 | held |
| I-ABUDGET-max_tool_calls-ok | spend max_tool_calls | E27 | held |
| I-ABUDGET-max_tool_calls-exhaust | exhaust max_tool_calls reauthorizes | E27 | held |
| I-ABUDGET-max_delegation_depth-ok | spend max_delegation_depth | E27 | held |
| I-ABUDGET-max_delegation_depth-exhaust | exhaust max_delegation_depth reauthorizes | E27 | held |
| I-ABUDGET-max_parallel_agents-ok | spend max_parallel_agents | E27 | held |
| I-ABUDGET-max_parallel_agents-exhaust | exhaust max_parallel_agents reauthorizes | E27 | held |
| I-ABUDGET-max_data_access-ok | spend max_data_access | E27 | held |
| I-ABUDGET-max_data_access-exhaust | exhaust max_data_access reauthorizes | E27 | held |
| I-ABUDGET-max_network_destinations-ok | spend max_network_destinations | E27 | held |
| I-ABUDGET-max_network_destinations-exhaust | exhaust max_network_destinations reauthorizes | E27 | held |
| I-ABUDGET-max_reasoning_compute-ok | spend max_reasoning_compute | E27 | held |
| I-ABUDGET-max_reasoning_compute-exhaust | exhaust max_reasoning_compute reauthorizes | E27 | held |
| I-ABUDGET-max_blast_radius-ok | spend max_blast_radius | E27 | held |
| I-ABUDGET-max_blast_radius-exhaust | exhaust max_blast_radius reauthorizes | E27 | held |
| I-LEASE-LEASE_EXPIRED | LEASE_EXPIRED | E27 | held |
| I-LEASE-LEASE_CAPABILITY_NOT_GRANTED | LEASE_CAPABILITY_NOT_GRANTED | E27 | held |
| I-LEASE-LEASE_RESOURCE_NOT_GRANTED | LEASE_RESOURCE_NOT_GRANTED | E27 | held |
| I-LEASE-LEASE_RISK_EXCEEDED | LEASE_RISK_EXCEEDED | E27 | held |
| I-LEASE-renew-expired | expired lease not auto-revived | E27 | held |
| I-LOOP-reauthorize | material change reauthorizes | E27 | held |
| I-LOOP-no-carry | loop carries no authority | E27 | held |
| I-GOAL-goal_drift | goal signal goal_drift | E27 | held |
| I-SELFREPORT-claim | self-report is a claim | E27 | held |
| I-SWE-rollback | swe rollback | E27 | held |
| I-SWE-SAFETY-canary | safety step canary | E27 | held |
| I-SWE-SAFETY-rollback | safety step rollback | E27 | held |
| I-EVO-candidate_id | candidate field candidate_id | E27 | held |
| I-EVO-problem | candidate field problem | E27 | held |
| I-EVO-proposal | candidate field proposal | E27 | held |
| I-EVO-code_digest | candidate field code_digest | E27 | held |
| I-EVO-benchmark | candidate field benchmark | E27 | held |
| I-EVO-security | candidate field security | E27 | held |
| I-EVO-performance | candidate field performance | E27 | held |
| I-EVO-risk | candidate field risk | E27 | held |
| I-EVO-rollback | candidate field rollback | E27 | held |
| I-EVO-evidence | candidate field evidence | E27 | held |
| I-EVO-no-confidence | model confidence not used in ranking | E27 | held |
| I-EVO-not-self-authorized | candidate not self-authorized | E27 | held |
| I-CODELAB-BASELINE | code lab stage BASELINE | E27 | held |
| I-CODELAB-GENERATE | code lab stage GENERATE | E27 | held |
| I-CODELAB-COMPILE | code lab stage COMPILE | E27 | held |
| I-CODELAB-TEST | code lab stage TEST | E27 | held |
| I-CODELAB-SECURITY_TEST | code lab stage SECURITY_TEST | E27 | held |
| I-CODELAB-BENCHMARK | code lab stage BENCHMARK | E27 | held |
| I-CODELAB-VERIFY | code lab stage VERIFY | E27 | held |
| I-CODELAB-RANK | code lab stage RANK | E27 | held |
| I-CODELAB-KEEP | code lab stage KEEP | E27 | held |
| I-CODELAB-no-prod | candidates do not enter production | E27 | held |
| I-RESEARCH-LOOP-QUESTION | loop QUESTION | E27 | held |
| I-RESEARCH-LOOP-HYPOTHESIS | loop HYPOTHESIS | E27 | held |
| I-RESEARCH-LOOP-SEARCH | loop SEARCH | E27 | held |
| I-RESEARCH-LOOP-EXPERIMENT | loop EXPERIMENT | E27 | held |
| I-RESEARCH-LOOP-CODE | loop CODE | E27 | held |
| I-RESEARCH-LOOP-CONFIRMATION | loop CONFIRMATION | E27 | held |
| I-RESEARCH-LOOP-GOVERNANCE_REVIEW | loop GOVERNANCE_REVIEW | E27 | held |
| I-RISK-behavioral_drift | forecast behavioral_drift | E27 | held |
| E28-I38 | an unchanged agent is SAME_GOVERNED_AGENT and keeps its live authority | E28 | held |
| E28-I39 | same identity presented with another key is IDENTITY_CLONE | E28 | held |
| E28-I40 | a signed succession gives CONTROLLED_SUCCESSOR with no retained authority | E28 | held |
| E28-I41 | a succession signed by the agent itself is not accepted | E28 | held |
| E28-I42 | a revoked identity presenting again is REVOKED_RESURRECTION | E28 | held |
| E28-I43 | an actor claiming an unknown identity is UNKNOWN_ACTOR | E28 | held |
| E28-I44 | a new key with a parent claim and no registration is UNVERIFIED_FORK | E28 | held |
| E28-I45 | a quarantine decision quarantines identity and E25 instance | E28 | held |
| E28-I83 | a self-improvement candidate cannot approve itself | E28 | held |
| E28-I84 | a candidate with any evaluation stage missing is not promoted | E28 | held |
| E30-LVL-1 | L1 is reached only with every evidence item it requires | E30 | held |
| E30-LVL-2 | L2 is reached only with every evidence item it requires | E30 | held |
| E30-LVL-3 | L3 is reached only with every evidence item it requires | E30 | held |
| E30-LVL-4 | L4 is reached only with every evidence item it requires | E30 | held |
| E30-LVL-5 | L5 is reached only with every evidence item it requires | E30 | held |
| E30-LVL-6 | L6 is reached only with every evidence item it requires | E30 | held |
| E30-LVL-7 | L7 is reached only with every evidence item it requires | E30 | held |
| E30-LVL-8 | L8 is reached only with every evidence item it requires | E30 | held |
| E30-LVL-9 | L9 is reached only with every evidence item it requires | E30 | held |
| E30-STATE-DEGRADED | only the listed states reach DEGRADED, by the listed actor | E30 | held |
| E30-STATE-RECOVERING | only the listed states reach RECOVERING, by the listed actor | E30 | held |
| E30-CAND-architecture | a 'architecture' candidate passes only the full firewall and a human review | E30 | held |
| E30-CAND-planner | a 'planner' candidate passes only the full firewall and a human review | E30 | held |
| E30-CAND-memory | a 'memory' candidate passes only the full firewall and a human review | E30 | held |
| E30-CAND-tool | a 'tool' candidate passes only the full firewall and a human review | E30 | held |
| E30-CAND-capability | a 'capability' candidate passes only the full firewall and a human review | E30 | held |
| E30-CAND-model | a 'model' candidate passes only the full firewall and a human review | E30 | held |
| E30-CAND-code | a 'code' candidate passes only the full firewall and a human review | E30 | held |
| E30-CAND-agent | a 'agent' candidate passes only the full firewall and a human review | E30 | held |
| E30-CAND-organization | a 'organization' candidate passes only the full firewall and a human review | E30 | held |
| E30-CAND-research_strategy | a 'research_strategy' candidate passes only the full firewall and a human review | E30 | held |
| E32-STAGE-PROPOSED | stage PROPOSED cannot be skipped past | E32 | held |
| E32-STAGE-SANDBOXED | stage SANDBOXED cannot be skipped past | E32 | held |
| E32-STAGE-TESTED | stage TESTED cannot be skipped past | E32 | held |
| E32-STAGE-ADVERSARIAL | stage ADVERSARIAL cannot be skipped past | E32 | held |
| E32-STAGE-COMPARATIVE | stage COMPARATIVE cannot be skipped past | E32 | held |
| E32-STAGE-CLEAN_ROOM_VERIFIED | stage CLEAN_ROOM_VERIFIED cannot be skipped past | E32 | held |
| E32-STAGE-CANARY | stage CANARY cannot be skipped past | E32 | held |
| E32-STAGE-OBSERVED | stage OBSERVED cannot be skipped past | E32 | held |
| E32-STAGE-APPROVED | stage APPROVED cannot be skipped past | E32 | held |
| E32-STAGE-PROMOTED | stage PROMOTED cannot be skipped past | E32 | held |
| E32-SELFTEST-identity | self-test of identity passes, and a failing check degrades it | E32 | held |
| E32-SELFTEST-authority | self-test of authority passes, and a failing check degrades it | E32 | held |
| E32-SELFTEST-policy | self-test of policy passes, and a failing check degrades it | E32 | held |
| E32-SELFTEST-proof | self-test of proof passes, and a failing check degrades it | E32 | held |
Other niches
Consensus & distributed systems · Attacks, threats & containment · Identity, authority & delegation · Evidence, receipts & proofs · Memory, data & privacy · Prediction, world models & simulation · Transactions, markets & economics · Tools, MCP, protocols & adapters · 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