# Illustrative implementation checklist. This is not a normative ContextOS runtime schema.
schema_version: contextos.ai/examples/multi-agent-replay-readiness/v1
kind: MultiAgentReplayReadiness

metadata:
  system: renewal_assistant
  owner: agent_platform_team
  reviewed_at: "2026-07-19"

scoring:
  absent: 0
  partial: 1
  implemented_and_measured: 2
  maximum: 24

assessment:
  single_agent_baseline:
    score: 0
    evidence_refs: []
    measures:
      task_success: null
      policy_pass_rate: null
      p95_latency_ms: null
      cost_per_trusted_outcome: null

  single_agent_replay:
    score: 0
    evidence_refs: []
    contract:
      pinned_artifacts:
        - run_context
        - context_pack
        - policy_bundle
        - model_profile
        - tool_manifest
      recorded_events:
        - tool_call
        - tool_result
        - evaluator_verdict
        - approval
        - decision_record
      expected_decision_record: true
      reexecute_external_effects: false

  task_decomposability:
    score: 0
    evidence_refs: []
    parallelizable_work: []
    sequential_dependencies: []
    shared_context_requirements: []

  cost_normalized_gain:
    score: 0
    evidence_refs: []
    includes:
      - model_tokens
      - tool_calls
      - evaluator_calls
      - human_review
      - retries
      - recovery

  lane_contracts:
    score: 0
    evidence_refs: []
    lanes:
      - id: contract_review
        mission: extract_material_renewal_terms
        input_artifacts:
          - signed_contract_snapshot
        output_artifact: contract_terms
        allowed_tools:
          - contracts.read
        approval_mode: read_only
        budget:
          max_tool_calls: 8
          max_duration_ms: 45000
        evaluator: contract_term_coverage_v3
        requires_evidence_refs: true

  authority_isolation:
    score: 0
    evidence_refs: []
    minimum_scopes_per_lane: true
    shared_ambient_credentials: false
    unrestricted_shared_tool_pool: false

  causal_trace:
    score: 0
    evidence_refs: []
    parent_child_trace_links: true
    artifact_hashes: true
    retry_reasons: true
    acceptance_decisions: true

  artifact_provenance:
    score: 0
    evidence_refs: []
    material_claims_require_sources: true
    uncertainty_survives_handoff: true

  lane_evaluation:
    score: 0
    evidence_refs: []
    lane_specific_evaluators: true
    final_answer_only_evaluation: false

  parent_acceptance:
    score: 0
    evidence_refs: []
    parent_can_reject: true
    parent_can_retry: true
    parent_can_escalate: true
    reject_when:
      - evidence_refs_missing
      - evaluator_failed
      - output_schema_invalid
      - lane_budget_exceeded

  side_effect_ownership:
    score: 0
    evidence_refs: []
    parent_owns_final_decision_record: true
    workers_can_mutate_final_state: false
    one_governed_effect_path: true

  recovery:
    score: 0
    evidence_refs: []
    cancellation_supported: true
    idempotency_keys_required: true
    rollback_rehearsed: false
    kill_switch_rehearsed: false

release_gates:
  thresholds:
    keep_single_agent: "0-11"
    shadow_one_read_only_lane: "12-18"
    centralized_pilot: "19-22"
    expansion_eligible: "23-24"
  hard_stops:
    - single_agent_replay_missing
    - worker_can_mutate_final_state
    - final_decision_record_has_no_single_owner
    - unrestricted_shared_tool_pool
    - cost_normalized_baseline_missing