{
 "phase": "Phase 0 - proof",
 "generated": "2026-09-27",
 "datasets": [
  {
   "slug": "clinc150",
   "name": "CLINC150 (in-scope)",
   "n_test": 4500,
   "n_classes": 150,
   "blend_accuracy_pct": 92.56,
   "lr_accuracy_pct": 92.56,
   "tuatara_accuracy_pct": 80.4,
   "mcnemar_exact_p": 1.0,
   "predictions_url": "https://cymetica.com/static/research/decide-benchmarks/clinc150/predictions.csv",
   "result_url": "https://cymetica.com/static/research/decide-benchmarks/clinc150/result.json",
   "score_script_url": "https://cymetica.com/static/research/decide-benchmarks/clinc150/score.py"
  },
  {
   "slug": "hwu64",
   "name": "HWU64",
   "n_test": 1076,
   "n_classes": 64,
   "blend_accuracy_pct": 87.73,
   "lr_accuracy_pct": 87.17,
   "tuatara_accuracy_pct": 81.51,
   "mcnemar_exact_p": 0.03125,
   "predictions_url": "https://cymetica.com/static/research/decide-benchmarks/hwu64/predictions.csv",
   "result_url": "https://cymetica.com/static/research/decide-benchmarks/hwu64/result.json",
   "score_script_url": "https://cymetica.com/static/research/decide-benchmarks/hwu64/score.py"
  },
  {
   "slug": "ag-news",
   "name": "AG News",
   "n_test": 7600,
   "n_classes": 4,
   "blend_accuracy_pct": 91.13,
   "lr_accuracy_pct": 90.99,
   "tuatara_accuracy_pct": 87.24,
   "mcnemar_exact_p": 0.16078,
   "predictions_url": "https://cymetica.com/static/research/decide-benchmarks/ag-news/predictions.csv",
   "result_url": "https://cymetica.com/static/research/decide-benchmarks/ag-news/result.json",
   "score_script_url": "https://cymetica.com/static/research/decide-benchmarks/ag-news/score.py"
  },
  {
   "slug": "sst2",
   "name": "SST-2",
   "n_test": 1821,
   "n_classes": 2,
   "blend_accuracy_pct": 84.4,
   "lr_accuracy_pct": 82.59,
   "tuatara_accuracy_pct": 80.23,
   "mcnemar_exact_p": 0.006435,
   "predictions_url": "https://cymetica.com/static/research/decide-benchmarks/sst2/predictions.csv",
   "result_url": "https://cymetica.com/static/research/decide-benchmarks/sst2/result.json",
   "score_script_url": "https://cymetica.com/static/research/decide-benchmarks/sst2/score.py"
  }
 ]
}