Specification RFCs
The Ontic governance specification. Each RFC addresses one layer of the authority pipeline — from problem targeting through enforcement.
Problem Targeting
Ensure the system is addressing the correct causal layer before state collection or authorization begins.
Canonical Ontology
Define what must be known about a real-world entity before authoritative claims are permitted.
Ground Truth Infrastructure
Define the authoritative data layer and verification protocol that enables CAA to reference external reality.
Model Selection & Training
Define how models are selected, trained, and validated to ensure the simulator is fit for purpose before governance constrains its outputs.
Prompt Derivation
Define how the Canonical Ontology and Oracle configurations are transformed into LLM-executable instructions. This RFC bridges the structural definitions (RFC-0001) with the runtime behavior, ensuring prompts are mechanically derived rather than manually authored.
State Extraction
Address the "Sensor Bottleneck" vulnerability where LLM-based extraction could hallucinate state values, causing the Governor to authorize on false premises.
State Negotiation
Define how the system interacts with users when required state is missing or ambiguous.
Evidence Binding
Canonical claim is invalid if RFC-0006 tests fail.
Opaque Boundary
Ensure control-plane isolation between proposal generation and authority evaluation.
Explicit Absence
Canonical claim is invalid if RFC-0008 tests fail.
Authorization Envelope
Separate proposal generation from authority granting.
Fallback Modes
Maintain usability without leaking authority.
Drift Detection
Prevent silent degradation of safety over time.
Agentic Governance
Extend CAA to govern LLMs operating as active orchestrators with tool access, not merely passive proposal generators.
Enclosed Execution
Define the client-side requirements for full chain of custody, enabling court-defensible attestation of user input before it enters the CAA pipeline.
Hardware Attestation
Specify how the `enclosed` enforcement locus is mechanically realized: what hardware root of trust is required, how measurements flow from boot to inference, how artifacts are hash-bound into the custody chain, and what constitutes a valid attestation for the Ontic Assurance Standard.
Human-in-the-Loop Protocol
Specify when and how human reviewers are integrated into CAA decision flows, ensuring human oversight is deterministic, auditable, and fail-closed.
Ready to implement?
The RFCs describe what Ontic does. Check your risk profile, or see the full architecture.