StandardsCPP — Constitutional Policy Protocol

The Policy Constitution for Governed AI.

CAES defines how decisions are enforced. CPP defines how decisions are determined. Together, they make AI execution reviewable, replayable, and policy-bound.

Why CPP Exists

Policy without a standard is not governance.

Without CPP, policy remains application logic, prompt convention, or mutable configuration. It can be claimed, changed, and argued about — but it cannot be proven.

With CPP, policy becomes a deterministic artifact that can be hashed, replayed, audited, and bound into receipts. The policy hash is the checksum of governance itself.

CPP applies to all policy systems used to evaluate execution actions, behavioral outputs, data operations, and any output crossing a CAES Effect Boundary.

System Relationship

CPP, CAES, and the Receipt form one chain.

01
CPPDetermines the decision

Defines deterministic, versioned, immutable, hashable policy semantics. Produces the PolicyHash bound at evaluation time.

02
CAESEnforces the execution boundary

Requires a valid Decision Receipt before any governed action crosses an Effect Boundary. Fail-closed on absence.

03
ReceiptProves what happened

A cryptographically signed artifact containing the PolicyHash. Independently replayable and offline-verifiable.

Core Requirements

Six non-negotiable properties.

Deterministic
Requires

Identical inputs produce identical outputs. No stochastic evaluation paths.

Prevents

LLM-only or probabilistic authorization paths.

Versioned
Requires

Every policy instance must carry a version identifier. Versions must be immutable.

Prevents

Ambiguous policy identity and undetectable version drift.

Immutable
Requires

Updates must produce a new policy version. No in-place modification.

Prevents

Silent retroactive policy changes and audit gaps.

Hashable
Requires

Every policy instance must produce a canonical PolicyHash: SHA-256 over the canonicalized policy. Same inputs always produce the same hash.

Prevents

Undetectable policy tampering and non-reproducible audits.

Auditable
Requires

Rule evaluation must be traceable. Matched rules must be explicitly recorded in evaluation output.

Prevents

Black-box policy outcomes and unverifiable authorization claims.

Portable
Requires

Policies must be evaluable outside the originating system. Offline verification must be possible.

Prevents

Vendor-locked governance and offline-verification failure.

PolicyHash

The cryptographic fingerprint of governance itself.

Every CPP-compliant policy must produce a deterministic hash bound to the policy state at evaluation time:

PolicyHash = SHA-256(canonical(policy))

Verifier can recompute this offline — no network call required.

Canonicalization removes key ordering variance, whitespace, and encoding differences
Computed at evaluation time — not precomputed and reused
Embedded in Decision Receipts, not referenced by pointer
Independently recomputable by any verifier with the policy definition

Without PolicyHash, audits become interpretive, policy drift becomes invisible, and decisions become non-reproducible.

CAES Requirement

Mandatory for Level 2 and Level 3 conformance.

Level 1Optional
Level 2Required
Level 3Required

CPP enables PolicyHash binding, deterministic audit, offline verification, and cross-system governance portability. These are required properties for CAES Level 2 conformance.

Rejected by CPP

Explicitly prohibited patterns.

LLM-only authorization decisions
Non-deterministic or probabilistic evaluation
Mutable policies without versioning
Hidden logic in application code
PolicyHash implemented as a name or pointer
Silent policy fallback
Execution without policy binding
Behavioral output without evaluation
Read Together

CPP is a companion to CAES, not a standalone thesis.

CAES defines the structural requirements for governed AI execution. CPP defines the policy substrate that CAES Level 2 and Level 3 implementations must use. Read them together.