priv/sarah/evals/experiences/benefit-comparison.v1.json

58e6347eeb72 · 1 KB

{
  "schema": "sarah.experience_benefit_comparison.v1",
  "purpose": "Compare identical scoped tasks with private experience recall disabled and enabled.",
  "controlled_authority": {
    "persona": "identical",
    "role": "identical",
    "tools": "identical",
    "profile_memory": "identical",
    "preferences": "identical"
  },
  "arms": [
    {
      "id": "memory_off",
      "experience_memory_enabled": false,
      "expected_bank": [],
      "score_dimensions": ["task_success", "correction_rate", "latency", "user_rating"]
    },
    {
      "id": "memory_on",
      "experience_memory_enabled": true,
      "expected_projection_contract": {
        "scope_is_exact": true,
        "terminal_cases_only": true,
        "applicability_required": true,
        "evidence_required": true,
        "single_success_is_advisory_not_universal": true
      },
      "score_dimensions": ["task_success", "correction_rate", "latency", "user_rating"]
    }
  ],
  "activation_gate": {
    "default_enabled": false,
    "requires_measured_benefit": true,
    "minimum_completed_task_pairs": 30,
    "requires_no_scope_or_provenance_regression": true
  }
}