{
  "schema": 1,
  "study": "cast-checksum-v11",
  "date": "2026-10-27",
  "title": "Reuse verification. Finish sooner.",
  "result": "accepted under the frozen combined gate",
  "comparison": "Previous accepted C++ Cast version versus V11 C++ Cast version; no third-party engine was measured.",
  "primary_case": "count_product_all_1m",
  "calls": 2400,
  "blocks": 20,
  "replicates_per_block_per_version": 5,
  "cases": 12,
  "exact_output_calls": 2400,
  "controls_passed": 11,
  "contended_calls": 1694,
  "worker_admission_shortfall_calls": 23,
  "workers_admitted_counts": {
    "1": 5,
    "2": 18,
    "3": 2377
  },
  "retries": 0,
  "exclusions": 0,
  "recorded_operating_anomalies": 0,
  "hardware": "Apple M5 Max, 18 cores, macOS",
  "topology": "Three requested loopback workers, four threads each, on one shared Mac.",
  "method": {
    "primary_metric": "Cast native wall_ms: runLead, including completed-result verification and leader cleanup; excludes worker spawning, worker reap, CSV export and final CLI reporting.",
    "inference": "Twenty paired blocks. Five raw calls per version per block are reduced to one geometric mean. A paired log-latency t interval uses 19 degrees of freedom and alpha=.05/12; the primary also uses all 2^20 two-sided sign swaps.",
    "primary_gate": "Adjusted upper latency bound <= -10% and raw two-sided p < .05/12.",
    "control_gate": "Every one of 11 other cases has adjusted upper latency bound <= +5%.",
    "correctness": "All 2,400 calls exit0 with exact complete CSV SHA and exact printed/native-JSON/journal-result checksums against retained previous-version outputs.",
    "cache": "Input identity reads before every call warm the OS file cache symmetrically. No extra native warmup or cache flush.",
    "retention": "No retries, exclusions, or early success stopping. Every condition/admission observation is included.",
    "scope": "Shared-host internal comparison; no quiet-host, multi-host, Photon, market-wide or new cross-language speed claim."
  },
  "primary": {
    "previous_wall_ms_median": 417.4145205,
    "candidate_wall_ms_median": 310.55112499999996,
    "paired_latency_change_percent": -32.236065847169336,
    "adjusted_latency_interval_percent": [
      -38.45241894950827,
      -25.39185629241747
    ],
    "raw_sign_randomization_p": 0.0000019073486328125
  },
  "case_results": [
    {
      "case": "count_product_all_250k",
      "role": "nonregression",
      "previous_wall_ms_median": 105.42543749999999,
      "candidate_wall_ms_median": 105.43618749999999,
      "paired_latency_change_percent": -0.8161216534475371,
      "adjusted_latency_interval_percent": [
        -3.002171067728571,
        1.419194967061538
      ],
      "passed": true,
      "label": "Count by product + join · 250k"
    },
    {
      "case": "count_product_all_1m",
      "role": "primary",
      "previous_wall_ms_median": 417.4145205,
      "candidate_wall_ms_median": 310.55112499999996,
      "paired_latency_change_percent": -32.236065847169336,
      "adjusted_latency_interval_percent": [
        -38.45241894950827,
        -25.39185629241747
      ],
      "passed": true,
      "label": "Count by product + join · 1M"
    },
    {
      "case": "pipeline_synth_terminal_250k",
      "role": "nonregression",
      "previous_wall_ms_median": 210.0773335,
      "candidate_wall_ms_median": 210.33266700000001,
      "paired_latency_change_percent": 0.23964033122162945,
      "adjusted_latency_interval_percent": [
        -0.40070115091694825,
        0.8840986818370499
      ],
      "passed": true,
      "label": "Filter + aggregate + join · 250k"
    },
    {
      "case": "pipeline_synth_terminal_1m",
      "role": "nonregression",
      "previous_wall_ms_median": 315.15475000000004,
      "candidate_wall_ms_median": 312.817562,
      "paired_latency_change_percent": -4.649748178498887,
      "adjusted_latency_interval_percent": [
        -8.939970978989892,
        -0.15739485075309753
      ],
      "passed": true,
      "label": "Filter + aggregate + join · 1M"
    },
    {
      "case": "pipeline_okay_terminal",
      "role": "nonregression",
      "previous_wall_ms_median": 105.421854,
      "candidate_wall_ms_median": 105.4045625,
      "paired_latency_change_percent": -0.057841221283510796,
      "adjusted_latency_interval_percent": [
        -0.5964145421604377,
        0.48365011528293483
      ],
      "passed": true,
      "label": "Small pipeline"
    },
    {
      "case": "duplicate_right_64",
      "role": "nonregression",
      "previous_wall_ms_median": 105.404563,
      "candidate_wall_ms_median": 105.420708,
      "paired_latency_change_percent": 0.30722415522610114,
      "adjusted_latency_interval_percent": [
        -0.407016175437272,
        1.0265867267377327
      ],
      "passed": true,
      "label": "Duplicate join keys · 64"
    },
    {
      "case": "empty_right_left_64",
      "role": "nonregression",
      "previous_wall_ms_median": 105.419417,
      "candidate_wall_ms_median": 105.393271,
      "paired_latency_change_percent": -0.03827048259441166,
      "adjusted_latency_interval_percent": [
        -0.632831279517104,
        0.5598478529578833
      ],
      "passed": true,
      "label": "Empty right input · 64"
    },
    {
      "case": "empty_left_inner_64",
      "role": "nonregression",
      "previous_wall_ms_median": 105.432563,
      "candidate_wall_ms_median": 105.437792,
      "paired_latency_change_percent": 0.35434983177658147,
      "adjusted_latency_interval_percent": [
        -0.3612283276589666,
        1.0750670760650123
      ],
      "passed": true,
      "label": "Empty left input · 64"
    },
    {
      "case": "pipeline_synth_sorted_1m",
      "role": "nonregression",
      "previous_wall_ms_median": 314.246729,
      "candidate_wall_ms_median": 312.94822899999997,
      "paired_latency_change_percent": -5.00203705166741,
      "adjusted_latency_interval_percent": [
        -8.062930623381673,
        -1.8392360608795475
      ],
      "passed": true,
      "label": "Pipeline + sort · 1M"
    },
    {
      "case": "inferred_product_keys_1m",
      "role": "nonregression",
      "previous_wall_ms_median": 416.45945900000004,
      "candidate_wall_ms_median": 305.497896,
      "paired_latency_change_percent": -34.22879294118594,
      "adjusted_latency_interval_percent": [
        -39.75241992678982,
        -28.198748020803325
      ],
      "passed": true,
      "label": "Inferred join keys · 1M"
    },
    {
      "case": "count_product_all_1m_default_control",
      "role": "nonregression",
      "previous_wall_ms_median": 417.211208,
      "candidate_wall_ms_median": 310.916646,
      "paired_latency_change_percent": -32.55139364959988,
      "adjusted_latency_interval_percent": [
        -38.64998923020797,
        -25.8465574573092
      ],
      "passed": true,
      "label": "Count + join default control · 1M"
    },
    {
      "case": "sum_then_sort_1m",
      "role": "nonregression",
      "previous_wall_ms_median": 105.51025,
      "candidate_wall_ms_median": 105.50233349999999,
      "paired_latency_change_percent": 0.011609677789049982,
      "adjusted_latency_interval_percent": [
        -0.7302332685064173,
        0.7589964162690693
      ],
      "passed": true,
      "label": "Sum + sort · 1M"
    }
  ],
  "primary_descriptive_medians": {
    "previous_full_command_ms": 1333.5222295000335,
    "candidate_full_command_ms": 1220.2172500000188,
    "previous_result_t_minus_stats_ms": 232.11991650000002,
    "candidate_result_t_minus_stats_ms": 116.8947915,
    "previous_collect_ms": 79.978625,
    "candidate_collect_ms": 79.4098125,
    "previous_whole_command_cpu_seconds": 0.88,
    "candidate_whole_command_cpu_seconds": 0.65,
    "paired_full_command_change_percent": -8.6577410873749
  },
  "interpretation": "V11 removes duplicate completed-result checksum/reporting work. It does not make the join, aggregate or SIMD kernels faster. Full-command and phase medians are descriptive, separate from the paired native-wall acceptance estimate.",
  "source_ids": {
    "previous_source_commit": "4fd0ed9cbc722e7128786111ee6aa8d9d304d7fc",
    "candidate_source_commit": "7d94e68c043c5111f52a624bda8e347b9a9da34b",
    "previous_cast_binary_sha256": "03f3c9cba4e45e77ec07232b588f80e94af5581e71ee4d4c6343cb9d27063701",
    "candidate_cast_binary_sha256": "0f3dd00e29bd3a2ee2ce80aa57256e0bfc5496c78c8670a7c3de32d535ca7198",
    "protocol_sha256": "64b281e778aa02de66c3dc809ae8f5d0ffef41f19dc3c4b905032602dd2641c7",
    "records_sha256": "3bb5978f50e1df48140f61e65d16dfb8ec4964c6bb5b39ddf575e9ae3ee7e65e",
    "formal_analysis_sha256": "444beddb442026806b4ee8c754c4d96a2b9dbb0b60529c399aa0f0da987508d8",
    "independent_statistics_audit_sha256": "5eda1cec860cd1a9010b5408b0a54c46860719d4e824d158afbdf3149c90dd2f"
  },
  "historical_language_tables": {
    "changed_by_this_publication": false,
    "data_date": "2026-10-15",
    "existing_site_evals_json_sha256": "26aacea70db62f9bd5f9528187ba09c25cfbd45feaab72c71c8d49f94d2ff70f",
    "v11_gain_applied_to_language_ratios": false
  }
}