{
  "schema_version": "1.0",
  "experiment_id": "slot-parallel-sql-generation-dynamic-masking-benchmarks",
  "origin": "first_party_lab",
  "recorded_at": "2026-09-30",
  "environment": {
    "architecture": "Slot-Parallel Multi-Head Classifier + Dynamic Schema Action Masking + SQLite Execution Verifier (RLEV)",
    "train_domains_n": 16,
    "val_domains_n": 4,
    "unseen_test_domains_n": 4,
    "unseen_test_queries_n": 60,
    "target_single_pass_latency_ms": 30.0
  },
  "aggregate_metrics": {
    "best_epoch": 9,
    "val_exact_match_pct": 5.0,
    "val_execution_accuracy_pct": 10.0,
    "test_valid_execution_rate_ve_pct": 61.67,
    "test_execution_accuracy_ex_pct": 10.0,
    "test_exact_match_em_pct": 1.67
  },
  "slot_level_accuracies_pct": {
    "order_direction": 86.67,
    "limit": 86.67,
    "primary_table": 55.0,
    "operator": 41.67,
    "agg_func": 31.67,
    "where_column": 31.67,
    "select_column": 18.33
  },
  "unseen_domain_breakdown": [
    {
      "domain": "automotive",
      "queries_n": 13,
      "execution_accuracy_ex_pct": 15.38,
      "exact_match_em_pct": 0.0
    },
    {
      "domain": "finance",
      "queries_n": 15,
      "execution_accuracy_ex_pct": 13.33,
      "exact_match_em_pct": 0.0
    },
    {
      "domain": "airline",
      "queries_n": 15,
      "execution_accuracy_ex_pct": 13.33,
      "exact_match_em_pct": 6.67
    },
    {
      "domain": "healthcare",
      "queries_n": 17,
      "execution_accuracy_ex_pct": 0.0,
      "exact_match_em_pct": 0.0
    }
  ]
}