{
  "purpose": "Public benchmark infrastructure smoke test; not a paper reproduction or full-budget model comparison.",
  "date": "2026-07-23",
  "recommendation": {
    "dataset": "MovieLens-100K",
    "limits": {
      "users": 80,
      "items": 200,
      "evaluation_users": 40,
      "steps": 2,
      "seed": 42
    },
    "benchmark_suite": "public",
    "fitness_metric": "public_composite",
    "trials": {
      "rankmixer_dense": {
        "validation_ndcg_at_10": 0.0434492175,
        "public_composite": 0.0476224164
      },
      "rankmixer_whale": {
        "validation_ndcg_at_10": 0.0168979408,
        "public_composite": 0.0161637258
      },
      "rankmixer_tmallgs": {
        "validation_ndcg_at_10": 0.0125,
        "public_composite": 0.0164453241
      },
      "rankmixer_long_history": {
        "validation_ndcg_at_10": 0.0304390076,
        "public_composite": 0.0349001483
      },
      "rankmixer_ramp": {
        "validation_ndcg_at_10": 0.0422363656,
        "public_composite": 0.0467241361
      }
    },
    "result": "All four new operators completed training and all public slices; no claim of improvement is made at this smoke budget."
  },
  "llm": {
    "model": "micro-llm",
    "dataset": "WikiText-2 + Stanford Alpaca + GSM8K",
    "limits": {
      "dimensions": 32,
      "layers": 1,
      "sequence_length": 48,
      "pretraining_steps": 1,
      "post_training_steps": 2,
      "seed": 42
    },
    "benchmark_suite": "public",
    "fitness_metric": "public_composite",
    "trials": {
      "dynamic_rubric": {
        "perplexity": 512.1624613,
        "preference_accuracy": 0.46875,
        "gsm8k_candidate_pass_at_1": 0.203125,
        "public_composite": -7.3897831
      },
      "off_context_grpo": {
        "perplexity": 512.2251315,
        "preference_accuracy": 0.46875,
        "gsm8k_candidate_pass_at_1": 0.21875,
        "public_composite": -7.3899166,
        "reward_bearing_group_rate": 1.0
      }
    },
    "result": "Both post-training paths completed and produced public capability metrics; the tiny budget is only a pipeline diagnostic."
  }
}