# AgentProof Century Trust Layer — Draft 0.1

## Purpose
The Century Trust Layer extends authorization into consequence control and long-term continuity.

Traditional authorization asks whether an actor may invoke an operation. AgentProof already resolves inherited authority immediately before an autonomous action. The Century Trust Layer adds two additional questions:

1. **How much change is this agent allowed to cause?**
2. **Can a future verifier prove what was expected, what actually happened, and which cryptographic lineage preserved that record?**

## Consequence Boundary™

A Consequence Boundary is a machine-readable maximum blast radius for one autonomous action. It can constrain:

- allowed action types;
- maximum transaction amount;
- currency;
- allowed counterparties;
- maximum authority-chain depth;
- whether the action must be reversible;
- latest allowable commit time.

A request can provide `consequence_boundary`. If omitted, AgentProof derives a narrow boundary from the mandate and current transaction.

A valid authority chain does **not** override a violated Consequence Boundary. A violation changes the resolution to `DENY`.

## Continuity Capsule™

Every enterprise resolution creates a Continuity Capsule binding:

- request digest;
- resolution digest;
- authority-chain digest;
- policy snapshot digest;
- optional pre-state digest;
- Consequence Boundary digest;
- expected-effects digest;
- execution-permit identifier when one exists;
- a dual SHA-256 + SHA-512 digest suite.

The capsule contains hashes rather than requiring raw private state to be public.

## Closure

After execution, the caller can close the capsule with:

- outcome: success, partial, failed, disputed, reversed, or unknown;
- optional post-state digest;
- observed effects;
- settlement metadata;
- evidence references.

AgentProof computes the difference between expected and observed effects and binds the closure with a new digest suite.

## Append-only correction

Historical records are never silently rewritten. A capsule may be superseded by a successor capsule. The old capsule remains publicly addressable, records its successor, and receives a continuity attestation documenting the correction lineage.

## Portable continuity archive

An authenticated tenant can export a self-describing `agentproof.continuity-archive/1` bundle containing the capsule bindings, consequence boundary, expected effects, closure, succession attestations, correction links, canonicalization profile, and digest suites. The archive can be hash-verified offline even if the AgentProof service later disappears. Current issuer authentication remains online because resolution proofs presently use HS256; a future asymmetric signing profile can remove that dependency.

## Cryptographic succession

A closed or open capsule can be re-attested. Re-attestation does not rewrite history. It creates an append-only continuity attestation whose predecessor points to the prior attestation.

Today's profile emits SHA-256 and SHA-512. Future implementations can add stronger algorithms while preserving the original bytes and original digests.

## Invariant

> Authorization can permit an action. It cannot authorize unlimited consequence.

## APIs

- Resolve: `POST /api/v1/resolve`
- Close: `POST /api/v1/continuity/close`
- Re-attest: `POST /api/v1/continuity/reattest`
- Supersede without deletion: `POST /api/v1/continuity/supersede`
- Export portable archive: `POST /api/v1/continuity/export`
- Public hash-only status: `GET /api/continuity/{capsule_id}`

## Non-claims

The Century Trust Layer does not predict every real-world side effect. It enforces declared machine-readable bounds on facts available at decision time and preserves a before/after evidence structure for later verification.