{
  "benchmark": "one-shot-vs-pev-pilot",
  "run_date": "2026-07-07",
  "model_class": "haiku-class subagents, identical for both conditions",
  "tasks": 4,
  "runs_per_task_per_condition": 1,
  "conditions": {
    "A_one_shot": { "constraint": "read + single edit, no command execution, no verification permitted" },
    "B_pev": { "constraint": "plan with success criterion, edit, run node --test, verify before declaring done, max 3 attempts" }
  },
  "results": {
    "logic_grading_post_harness_normalization": { "A": "4/4 tasks pass", "B": "4/4 tasks pass" },
    "as_run_environment": {
      "A": { "completed": "0/4", "claimed_success": "4/4", "false_success_rate": "4/4" },
      "B": { "completed": "4/4", "claimed_success": "4/4", "false_success_rate": "0/4" }
    },
    "environmental_defect": "Task corpus lacked package.json {type:module}; ESM tests could not run. Unknown to the task author until PEV agents' verification runs surfaced it. All 4 PEV agents diagnosed and repaired it unprompted; no one-shot agent could discover it.",
    "cost": {
      "A": { "avg_tokens": 20878, "avg_duration_ms": 8168, "avg_tool_uses": 3 },
      "B": { "avg_tokens": 23057, "avg_duration_ms": 32421, "avg_tool_uses": 7.75 }
    }
  },
  "baselines_confirmed_failing": { "task1": "0/3 pass", "task2": "1/3 pass", "task3": "2/3 pass", "task4": "4/6 pass" }
}