Cognitive Authority Recovery Architecture (CARA): Deterministic Recovery Engine and Safety Assurance Case — Technical Report v1.3.1

CARA (Cognitive Authority Recovery Architecture) is a deterministic recovery engine and safety assurance artifact for autonomous systems operating under authority-gating architectures such as the Human-Machine Authority Architecture (HMAA).

The system computes a bounded recovery score R ∈ [0,1] from five evidence channels: sensor trust trajectory (τ_prior, τ_current), SALM audit integrity, operator confirmation chain, anomaly clearance, and elapsed lockout duration (t_locked). The score is mapped to four Graduated Re-Entry Protocol (GREP) phases through a guarded classification pipeline (G0–G4) enforcing algebraic and policy invariants.

The artifact includes a formal mathematical specification, deterministic Node.js reference implementation (zero external dependencies), interactive browser dashboard, hazard verification suite (H1–H10 + H3b), 27,000+ seeded Monte Carlo trials, a 100-seed sweep, parameter sensitivity analysis, and a Goal Structuring Notation (GSN) safety case.

This is a research reference implementation and safety argument artifact, not a production deployment system. Remaining assurance gaps formal non-bypass proof, hardware input attestation, cryptographic signing of recovery decisions, and hardware-in-the-loop validation are documented in the artifact’s assurance gap matrix.