Authority-change governance conformance

This contract is a bounded ground-truth benchmark for reconstructing why an authority change occurred. It is not an IGA workflow engine, approval UI, GRC platform, policy decision point, runtime enforcement service, or vendor adapter. EADS continues to own enterprise and operational topology; this package models only fictional identity authority, governance observations, and evaluator truth.

The 1.0.0 fixture contains 12 hand-inspectable cases covering a valid grant, wrong and expired approvers, approved-versus-enacted scope drift, denied-but- enacted authority, an emergency exception, decision-time policy selection, missing retained evidence, supersession linkage, revocation timing drift, conflicting decisions, and a structurally valid unauthorized change. The examples remain generated contract documentation. The byte-identical public/evaluator fixture is frozen separately under synthworld.benchmarks/authority-governance-v1, where a root SHA256SUMS binds both payloads and both visibility manifests.

One clock and deterministic precedence

All request, decision, enactment, expiry, and audit coordinates are nonnegative integer ticks on SynthWorld’s shipped temporal axis. No UTC field participates in generation, ordering, identity, replay, or scoring. TemporalEventEnvelopeV2 preserves V1’s fields, payload digest, canonical (effective_tick,event_id) ordering, and derived zero-based index. V2 adds only governance_1_0; V1 remains closed and rejects that family. Neither loader upgrades or relabels an artifact.

For multiple decisions, the controlling decision is the last canonical (effective_tick,event_id) decision strictly before enactment. A later policy, decision, or evidence record cannot retroactively authorize an earlier change. Malformed references and phase ordering are rejected. A well-formed but unauthorized or incorrectly enacted history remains valid input and is scored.

Visibility and metrics

Public input contains observed requests, decisions, enactments, audits, bounded policy rules, approver mandates, and opaque evidence references. Evaluator data is serialized separately and holds case labels, canonical before/after state, governance and approver verdicts, applicable decision-time policy/control, required evidence, expected enactment, and reconstructability truth.

Reports contain no aggregate security score. State, governance authority, policy/rationale, evidence/observability, and enactment metrics each publish an integer numerator, denominator, support, and denominator meaning. Action auditability remains an independent Asteria/enterprise-agentic measurement; this family scores authority-change auditability.

Three public-only baselines deliberately fail distinct dimensions: inferring validity from final state, trusting any recorded approval, and evaluating all history under the latest policy.

Generate or verify the schemas and examples with:

uv run python authority-governance-contract/tools/generate_contract.py
uv run python authority-governance-contract/tools/generate_contract.py --check