{
  "status": "passed",
  "diagnostic_only": true,
  "paper_result_reproduced": false,
  "dataset": {
    "name": "official GSM8K",
    "revision": "3101c7d5072418e28b9008a6636bde82a006892c",
    "train_sha256": "17f347dc51477c50d4efb83959dbb7c56297aba886e5544ee2aaed3024813465",
    "test_sha256": "3730d312f6e3440559ace48831e51066acaca737f6eabec99bccb9e4b3c39d14",
    "train_examples": 4,
    "validation_examples": 2,
    "test_examples": 2,
    "split": "first 4 official train rows; next 2 official train rows validation; first 2 official test rows, read only after training"
  },
  "checkpoint": {
    "model_id": "Qwen/Qwen3-4B-Instruct-2507",
    "revision": "cdbee75f17c01a7cc42f958dc650907174af0554",
    "weights_committed": false
  },
  "seed": 42,
  "budget": {
    "optimizer_updates": 2,
    "generation_cap": 192,
    "lora_rank": 8,
    "lora_target_modules": [
      "q_proj",
      "v_proj"
    ],
    "accelerator": "NVIDIA A100-SXM4-80GB"
  },
  "controls": {
    "dynamic_dce_srcl": {
      "baseline_validation_accuracy": 0.5,
      "final_validation_accuracy": 0.5,
      "baseline_validation_mean_tokens": 164.5,
      "final_validation_mean_tokens": 174.5,
      "final_test_accuracy": 1.0,
      "srcl_accepted_rounds": 2,
      "adapter_parameter_delta_l2": 0.01001878839044344
    },
    "fixed_teacher_same_assistant_prompt": {
      "baseline_validation_accuracy": 0.5,
      "final_validation_accuracy": 0.5,
      "baseline_validation_mean_tokens": 164.5,
      "final_validation_mean_tokens": 174.0,
      "final_test_accuracy": 1.0,
      "srcl_accepted_rounds": 2,
      "adapter_parameter_delta_l2": 0.010018953337736036
    },
    "dynamic_dce_without_srcl": {
      "baseline_validation_accuracy": 0.5,
      "final_validation_accuracy": 0.5,
      "baseline_validation_mean_tokens": 164.5,
      "final_validation_mean_tokens": 174.5,
      "final_test_accuracy": 1.0,
      "srcl_accepted_rounds": 0,
      "adapter_parameter_delta_l2": 0.00995904023904781
    }
  },
  "comparison_caveat": "Two validation/test problems and two updates have no statistical power; fixed-teacher uses assistant-side placement and is not the paper OPSD baseline.",
  "omitted": [
    "14,717 OpenThoughts problems",
    "AIME/HMMT Average@12 evaluation",
    "200-step rank-128 eight-H100 training",
    "paper-matched SFT, GRPO, OPSD and test-time scaling controls"
  ],
  "schema_version": 2,
  "manifest_ref": "reproduction:recursive-opsd",
  "evaluation_protocol": {
    "tier": "l1_mechanism",
    "seeds": [
      42
    ],
    "formal_comparison": false,
    "claim_policy": "single/few-seed smoke result; do not claim a stable improvement"
  },
  "provenance": {
    "artifact_path": "docs/reproductions/2609.30652-recursive-opsd/metrics/qwen3-4b-gsm8k-seed42.json",
    "dataset_fingerprint": "sha256:17f347dc51477c50d4efb83959dbb7c56297aba886e5544ee2aaed3024813465",
    "code_commit": "4c3051124c219e0ceb0bea7cee63cf123e52bb33"
  }
}