# CAIN42 — ATLAS Federation Specification (Prompt 15, sections 3-4, 11, 17)

## Federated assurance exchange

An organization defines its **required assurance** (`AssuranceRequest`: requesting principal, target organization,
target agent, intended action, action/transaction type, consequence class, requested authority, required evidence,
minimum assurance level, freshness, independence, geographic/temporal/regulatory constraints, human-approval,
acceptable runtime/identity/verification classes).

The remote system returns `AssuranceResponse` (identity, controlling organization, authority, delegation chain,
current assurance, certificate, evidence, freshness, independence, behavioral/capability/model/memory/runtime/
supply-chain/policy status, effect-history summary, revocation, expiration, signatures).

ATLAS computes a deterministic **ASSURANCE MATCH**: ALLOW / RESTRICT / ESCALATE / REQUIRE_HUMAN /
REQUIRE_ADDITIONAL_EVIDENCE / DENY.

## Negotiation

`AssuranceNegotiator` runs a bounded number of rounds. Rounds may only add **evidence**, never authority. Local
policy is authoritative and is evaluated first; it can only deny (P120, P121, P130).

## Counterparty registry

`CounterpartyRegistry` stores verifiable evidence (identity, assurance level, capabilities, authority, protocols,
security posture, evidence, revocation endpoint, assurance endpoint, transaction limits). A declared level is
**clamped down** to the evidence-backed level; the registry is not a trust-score directory.

## Evidence-based reputation

`EvidenceBasedReputation` is derived only from verified behavior, successful effects, violations, revocations,
incidents, assurance continuity, transaction outcomes and independent attestations. Likes/followers/self-reported
scores are rejected and never authoritative (P117).

## Status

`IMPLEMENTED`, `TESTED`, `INTEGRATED`, `LIVE` (8 stages). No cross-organization production federation with a real
counterparty exists yet — that is the top blocker.
