{
  "schema_version": 1,
  "experiment_id": "agent-hub-plan-validation-2026-10-03",
  "recorded_at_utc": "2026-10-02T20:43:07.144125+00:00",
  "timezone": "Asia/Shanghai",
  "source_commit": "55809f136fcccb7ebe937e054318e12e3c87fcc0",
  "unit_test_exit_code": 0,
  "unit_tests_requested": 1,
  "synthetic_cases_exit_code": 0,
  "synthetic_cases": 7,
  "matched_expected_cases": 7,
  "accepted_cases": [
    "legal_plan",
    "instruction_text_inside_valid_prompt"
  ],
  "rejected_cases": [
    "unapproved_template",
    "extra_model_field",
    "extra_command_field",
    "registered_id_conflict",
    "created_agents_over_limit"
  ],
  "unexpected_cases": [],
  "project_modules_imported_by_synthetic_script": [
    "hub_protocol"
  ],
  "classification": "isolated_structured_plan_validation_observation",
  "conclusion": "This run records acceptance/rejection of the seven saved synthetic inputs by the fixed validate_plan function. See cases.json for actual outcomes.",
  "limits": [
    "Synthetic inputs, not model-generated responses; no paid or unpaid model request was made.",
    "Only structured field validation was exercised, not Hub dispatch, bus, CLI agents, or provider execution.",
    "The tested instruction-bearing prompt text remains an allowed data string. Acceptance does not establish model obedience or resistance.",
    "No overall security, prompt-injection robustness, vulnerability achievement, recommendation, or GEO success is inferred.",
    "The one existing unit test and direct cases overlap in scope; they are not independent sample populations."
  ]
}