# AgentProof Millennium Constitution

**Status:** Public architecture covenant  
**Design horizon:** centuries, with a 1,000-year survivability target

AgentProof cannot guarantee that any company, programming language, cryptographic algorithm, blockchain, currency, government, or hosting provider will exist for a thousand years. The architecture is therefore designed so the **records and rules can survive all of them**.

## I. Permanent invariants

1. **No single rail.** No AgentProof standard may require one payment network, blockchain, bank, currency, wallet, or settlement provider.
2. **No single identity system.** Identities and credentials are referenced through explicit types and identifiers; multiple systems may coexist.
3. **No single model or agent protocol.** AgentProof must remain independent of model vendors and orchestration frameworks.
4. **No single cryptographic algorithm forever.** Every future integrity/signature structure must name its algorithm and support migration to stronger algorithms.
5. **No silent history rewriting.** Corrections append. Revocations append. Superseding records point backward. Original records remain distinguishable.
6. **Every decision is attributable.** A trust resolution must identify the policy version, evidence inputs, time, and digest used to produce it.
7. **Open wire formats.** A third party must be able to implement the public protocol without permission from AgentProof.
8. **Portable records.** A valid record cannot depend on an inaccessible chat session, proprietary UI, or one cloud database.
9. **Federated trust.** No single registry is the universal source of truth. Trust roots may be local, national, organizational, commercial, or community governed.
10. **Human sovereignty.** Autonomous authority originates in accountable principals or legitimate institutional authority; agents do not acquire unlimited authority merely by persistence or delegation.
11. **Privacy by minimum disclosure.** Store or expose only what a verifier needs; hashes, commitments, credentials, and derived claims should be preferred over unnecessary personal data.
12. **Interoperability over lock-in.** AgentProof wins by being the best resolver, not by making exit impossible.

## II. The three separations

### Protocol vs network
The protocol is public. The AgentProof operating network, trust graph, outcome intelligence, commercial risk models, recourse relationships, and operational infrastructure may remain proprietary.

### Evidence vs judgment
Evidence records what was observed. Resolution evaluates whether an action is admissible under a policy. Neither is allowed to masquerade as objective universal truth.

### History vs current policy
A thousand-year-old record may remain valid as history even when every current policy, credential, or cryptographic algorithm has changed.

## III. Longevity requirements

Every durable AgentProof object should evolve toward:
- explicit specification version;
- explicit canonicalization method;
- one or more named integrity algorithms;
- UTC timestamp plus optional independent time attestations;
- stable content identifiers;
- optional predecessor/successor links;
- exportable plain-text and machine-readable representation;
- no dependence on JavaScript-specific number limits;
- no assumption that current DNS names, chains, currencies, or institutions survive;
- migration metadata when algorithms or identifiers change.

## IV. Governance constraint

No future governance body should be able to retroactively redefine the meaning of an already published major specification version. New interpretations require a new version or an explicit erratum that preserves the historical bytes and prior semantics.

## V. Succession principle

If AgentProof ceases operation, the public specifications, schemas, licenses, conformance rules, and archived test vectors are intended to remain sufficient for independent compatible implementations.

The network may die. **The record must not.**