quality:
  completion_primary_metric: answer_satisfaction
  answer_satisfaction_min: 0.98
  satisfaction_confidence_lower_bound_min: 0.98
  evaluation_min_unseen_scenarios: 200
  evaluation_min_scenario_classes: 11
  evaluation_min_each_class: 10
  evaluation_min_critical: 30
  evaluation_min_multiturn: 20
  evaluation_min_false_premise: 20
  production_model_self_judge_allowed: false
  user_need_resolution_role: diagnostic_only
  critical_satisfaction_min: 0.99
  false_premise_correction_min: 1.0
  unsupported_hallucination_max: 0
  legacy_mixing_max: 0
  secret_leak_max: 0
  unexecuted_completion_claim_max: 0
  satisfaction_required_dimensions:
    - purpose_fulfilled
    - preflight_correct
    - intent_correct
    - all_major_needs_covered
    - required_depth_met
    - factual
    - evidence_complete
    - constraints_respected
    - conditions_exceptions_covered
    - current_status_covered_when_required
    - next_action_covered_when_required
    - relevant
    - direct
    - clear
    - appropriately_concise
    - actionable_when_required
    - context_consistent
    - clarification_efficient
    - resolution_mode_correct
    - self_contained
  same_revision_evidence_required: true
  evaluation_input_contract: scenario_plus_runtime_evidence