scaling_1784742712/scaling_summary.json

{
  "seed": 20260722,
  "seq_len": 3920,
  "eval_samples": 30,
  "epochs_cap": 16,
  "rows": [
    {
      "arm": "scale_900",
      "samples": 900,
      "points": 3528000,
      "best_val_loss": 0.39776119589805603,
      "Affiliation-F": 0.20117460220618313,
      "F1_T": 0.20117460220618313,
      "Standard-F1": 0.20117460220618313,
      "VUS-PR": 0.1645291101006374
    },
    {
      "arm": "scale_3000",
      "samples": 3000,
      "points": 11760000,
      "best_val_loss": 0.29562093019485475,
      "Affiliation-F": 0.27190405017097413,
      "F1_T": 0.27190405017097413,
      "Standard-F1": 0.27190405017097413,
      "VUS-PR": 0.2497031037917982
    },
    {
      "arm": "scale_9000",
      "samples": 9000,
      "points": 35280000,
      "best_val_loss": 0.2114131361246109,
      "Affiliation-F": 0.45809959626880686,
      "F1_T": 0.45809959626880686,
      "Standard-F1": 0.45809959626880686,
      "VUS-PR": 0.6130390115046325
    }
  ],
  "elapsed_seconds": 1659.87,
  "note": "Reduced scaling run: nested synthetic-RCD sample counts, checkpoint-free, same architecture/loss."
}