CAIN LEGION — verifiable autonomy & authority science (2026-10-08)

AUTHORITY IS A STATE, NOT A BOOLEAN. NO VERIFIED AUTHORITY LINEAGE → NO HIGH-ASSURANCE AUTHORIZATION.

What was built

An Authority State Model (VALID/EXPIRED/REVOKED/SUPERSEDED/CONTEXT_INVALID/PROVENANCE_INVALID/POLICY_INVALID/ATTESTATION_INVALID/BUDGET_EXCEEDED/AUTHORITY_ESCALATED/INCONCLUSIVE, where INCONCLUSIVE is never ALLOW), an Authority Continuity Token that refuses commit-time TOCTOU drift, a Context Authority Evaluator, Purpose-Bound Capability, a Mission Authorization Envelope, a derived Authority Decay Model, Emergency Authority with contained escalation, a multi-kind Revocation Fabric, an Authority Fork Detector, a Logical Authority Clock, a canonical Verifiable Authorization Object + Authority Receipt 2.0, Change-Impact Recertification, a Causal Authorization Engine, a Capability Composition Analyzer, an Evidence Graph with trust classes, a hash-chained WAL Authority Store, an Independent Verifier 2.0, and a 20-property formal suite.

What was measured

• Formal properties: 20/20 pass.
• Adversarial population: 24000 attempts, 24000 blocked, 0 breaches across 12 attack classes.
• Governance mutation suite: 21 of 21 killed; survivors [].
• Authority-state evaluate latency (local): p50 0.0036 ms.
• IP landscape: 35 scored invention candidates (33 tested).

Reproduce: python3 scripts/legion/run_evidence.py --trials 2000 --seed 20261009.

Files: RESULTS.json · PROPERTIES.json · SWARM.json · MUTATION.json · LATENCY.json · manifest.json.

Limits. Libraries; the runtime stage is still opt-in and off by default. No staging soak, no third-party verification, no customers, no revenue, no patents. Physical AI is simulated only. The IP landscape is engineering invention-discovery, not legal advice and not a patentability claim. PRE-PRODUCTION.