{
  "schema_version": 1,
  "date": "2026-10-18",
  "purpose": "setup-validation",
  "limitations": [
    "One task and one seed per condition; no causal effect estimate.",
    "Separate from the original 20 exploratory runs and the Inspect integration pilot."
  ],
  "smoke_protocol": {
    "model": "qwen3:14b",
    "seed": 2000,
    "step_limit": 40,
    "wall_limit_minutes": 240
  },
  "smoke_trials": [
    {
      "condition": "P0G0",
      "label": "Original instruction",
      "model": "qwen3:14b",
      "seed": 2000,
      "examples_passed": 2,
      "examples_total": 3,
      "hidden_passed": 49,
      "hidden_total": 88,
      "weighted_score": 0.8016,
      "full_solve": false,
      "agent_seconds": 229.5,
      "end_reason": "completed",
      "result_sha256": "de65c5821d88d5279be2fb4b8c89099220f0d335c7d4b62067fcdea03d06b952"
    },
    {
      "condition": "P0G1",
      "label": "Original instruction + completion check",
      "model": "qwen3:14b",
      "seed": 2000,
      "examples_passed": 3,
      "examples_total": 3,
      "hidden_passed": 52,
      "hidden_total": 88,
      "weighted_score": 0.8349,
      "full_solve": false,
      "agent_seconds": 3137.6,
      "end_reason": "completed",
      "result_sha256": "7b6a5cbb5a3e5fe4370a976082350faa2fbb4c5e2f34bd70b32fc4d8e98ab699"
    },
    {
      "condition": "P1G0",
      "label": "Explicit testing instruction",
      "model": "qwen3:14b",
      "seed": 2000,
      "examples_passed": 1,
      "examples_total": 3,
      "hidden_passed": 31,
      "hidden_total": 88,
      "weighted_score": 0.4008,
      "full_solve": false,
      "agent_seconds": 808.3,
      "end_reason": "completed",
      "result_sha256": "f330ea2b769bb0dda4f61437f099e3b0dd808e5fdceb4ec4f125780c3274c4fc"
    },
    {
      "condition": "P1G1",
      "label": "Explicit testing instruction + completion check",
      "model": "qwen3:14b",
      "seed": 2000,
      "examples_passed": 2,
      "examples_total": 3,
      "hidden_passed": 35,
      "hidden_total": 88,
      "weighted_score": 0.5451,
      "full_solve": false,
      "agent_seconds": 14400.2,
      "end_reason": "wall_limit",
      "result_sha256": "c0dae90ad7c921171579677d1d7e115ab4de84d8d5b52338908297761eb1b2d1"
    }
  ],
  "inspect_checks": {
    "purpose": "integration-validation",
    "matching_saved_submissions": 4,
    "matching_test_outcomes": 364,
    "adapter_tests": 20,
    "transport_tests": 5,
    "evidence_tamper_checks": 5
  }
}