{
  "schema_version": 1,
  "adapter_key": "sguid",
  "validated_at": "2026-10-12",
  "accelerator": {
    "vendor": "NVIDIA",
    "model": "NVIDIA A100-SXM4-80GB"
  },
  "command": [
    "python",
    "scripts/run_sguid_checkpoint.py",
    "--checkpoint-path",
    "checkpoints/public-sguid-model",
    "--checkpoint-id",
    "Qwen/Qwen3.5-9B",
    "--checkpoint-revision",
    "c202236235762e1c871ad0ccb60c8ee5ba337b9a",
    "--train-path",
    "data/gsm8k-holdout/train.jsonl",
    "--validation-path",
    "data/gsm8k-holdout/validation.jsonl",
    "--split-manifest",
    "data/gsm8k-holdout/split.json",
    "--dataset-revision",
    "3101c7d5072418e28b9008a6636bde82a006892c",
    "--training-examples",
    "200",
    "--probe-steps",
    "200",
    "--internalization-steps",
    "32",
    "--max-tokens",
    "1024",
    "--skill-tokens",
    "256",
    "--candidate-count",
    "32",
    "--retrieve-count",
    "6",
    "--late-minimum",
    "2",
    "--rounds",
    "2",
    "--evaluation-max-tokens",
    "3072",
    "--validation-examples",
    "4",
    "--seed",
    "42",
    "--device",
    "cuda",
    "--output",
    "runs/sguid-seed42.json"
  ],
  "dataset": {
    "id": "openai/grade-school-math",
    "revision": "3101c7d5072418e28b9008a6636bde82a006892c"
  },
  "checkpoint": {
    "id": "Qwen/Qwen3.5-9B",
    "revision": "c202236235762e1c871ad0ccb60c8ee5ba337b9a"
  },
  "result": "passed",
  "evaluation_tier": "l1_mechanism",
  "diagnostic_only": true,
  "capability_passed": false,
  "claim_policy": "execution gate only; no capability or validation gain claim",
  "metrics": {
    "completed_rounds": 2,
    "adapter_parameter_delta_l2": 0.17389918884542147,
    "evaluation_max_tokens": 3072,
    "baseline_validation_exact_match": 0.75,
    "final_validation_exact_match": 0.25,
    "validation_exact_match_delta": -0.5,
    "validation_examples": 4,
    "baseline_validation_eos_completed": 2,
    "baseline_validation_truncated": 2,
    "final_validation_eos_completed": 2,
    "validation_truncated": 2,
    "rounds": [
      {
        "round": 0,
        "candidate_skills": 32,
        "selected_skills": 6,
        "probe_updates": 200,
        "internalization_updates": 32,
        "probe_nonzero_gates": 1200,
        "internalization_nonzero_gates": 192,
        "internalization_parameter_delta_l2": 0.08388192717773137
      },
      {
        "round": 1,
        "candidate_skills": 32,
        "selected_skills": 3,
        "probe_updates": 200,
        "internalization_updates": 32,
        "probe_nonzero_gates": 1200,
        "internalization_nonzero_gates": 96,
        "internalization_parameter_delta_l2": 0.09196466716817275
      }
    ]
  },
  "limitations": [
    "single-seed bounded L1 mechanism diagnostic; not AIME/HMMT improvement",
    "GSM8K training-only holdout instead of DAPO-Math-17K",
    "current LM hidden cosine retrieval, not original embedding checkpoint",
    "q/v rank-8 LoRA and non-original early/late window/h; not original all-linear rank-64 run",
    "signed generalized full-vocabulary KL, not original SGSD bounded discrepancy",
    "EOS/truncations and any negative validation outcome retained",
    "Qwen3.5 hybrid text decoder, not original Qwen3 backbone; vision unused; q/v LoRA applies only to full-attention layers"
  ],
  "executed_source_sha256": {
    "src/auto_research/post_training/sguid_checkpoint.py": "3568a056a7f28314f03d5b26f52831ce3ff465c481fca564516f4f19f76cb294",
    "scripts/run_sguid_checkpoint.py": "d1780751d8ba416e48c4466ff2b94c80f3aa359ffd5d6060aa238588539ae803",
    "src/auto_research/post_training/oct11_objectives.py": "5db5f4eed68843fb2b8ce1f68bf2ff31ee36c0525268db1e295522a8f2cac6bc",
    "src/auto_research/post_training/oct10_checkpoint.py": "247f1f97a3ca701277f241ec81d99296f0ea82e1db8f68006fcd7bfdc661cd74",
    "src/auto_research/public_holdout.py": "033d9c1fd112accf5ac79573b99d9f31cffc646bcab6911cf3e49459fa003b46"
  },
  "provenance": {
    "commit": "319008f69ac6224ad0a2dfe3146d551c1999f5e3",
    "artifact_path": "docs/gpu-validations/sguid-a100-20261012.json",
    "metrics_artifact": "docs/post-training/2610.12367-sguid/metrics/checkpoint-a100-seed42.json",
    "checkpoint_committed": false
  }
}