Claim Authorization Architecture Specification
RFC Set 0001–0006 • Public Draft
This document is the technical specification for Ontic Labs' Claim Authorization Architecture (CAA). It is intended for architects, security teams, and integrators who need the precise semantics behind the Reality Fidelity concepts described elsewhere on this site.
Canonical Ontology & Claim Authorization RFCs
RFC-0001: Canonical Ontology Objects (COO)
Defines the structure and requirements for representing real-world entities before authoritative claims can be made.
Key interfaces:
• OntologyObject — identity, state dimensions, authority requirements
• StateAxis — dimensions of state (enum, range, validated_free)
• RequiredStateLogic — always required, value constraints, conditional requirements
• AuthorityRequirements — oracle usage, verification methods
RFC-0002: Interaction & State Negotiation Protocol
Details how the system interacts with users when required state is missing or ambiguous.
Status codes:
• AUTHORIZED — all required state present, provenance verified
• REQUIRES_SPECIFICATION — missing required state
• AMBIGUOUS_MAPPING — multiple valid interpretations
• UNRESOLVABLE — cannot determine required state
Inferred State Rule: Explicit user confirmation required for inferred state in sensitive domains.
RFC-0003: Oracle & Verification Model
Specifies how external reality is referenced and verified.
Oracle types:
• Database — verified external data source
• Standards document — regulatory or industry standard
• Human lock — explicit human confirmation
• Telemetry — real-time measurement
Requirements: Every authoritative output must declare oracle reference, verification method, and resolution layer.
RFC-0004: Drift Detection & Versioning
Focuses on preventing silent degradation of safety over time.
Mandates:
• Versioning of ontology objects
• Invalidation of cached authorizations upon schema changes
• Drift detection tests to challenge minimal state definitions
• Audit trails for schema evolution
RFC-0005: Proposal & Authorization Envelope
Separates proposal generation from authority granting.
AuthorizationEnvelope types:
• measurement — verified quantitative output
• narrative — explanatory content without authority
• refusal — explicit rejection with reason
• dispute_summary — conflicting oracle resolution
Clients must explicitly handle each variant.
RFC-0006: Fallback & Degraded Modes
Maintains usability without leaking authority.
Modes:
• NARRATIVE_ONLY — can explain, cannot emit authoritative claims
• CANNED_RESPONSE_ONLY — pre-approved responses only
Grammar constraints on narrative output:
• Attributive language required
• No definitive classifications
• Streaming rule: sentences not rendered until authorization status cleared
Adversarial Test Requirement
All implementations must pass the Ontic Adversarial Prompt Suite to prevent:
- Prompt injection attacks
- Fictional wrapper attacks
- Adjective smuggling ("safe", "standard", "best")
- Semantic coercion
- Narrative smuggling
- JSON protocol injection
- Streaming race conditions
For the complete specification with code examples and test cases, contact enterprise or view the architecture overview.