{
  "schema": "coreai-scalar-gate-summary/1",
  "result": "PASS",
  "recipe": "system-one-scorer-4b-fp16",
  "license": "cc-by-nc-4.0",
  "source": {
    "hf_id": "pngwn/system-one-qwen3.5-4b-scorer",
    "revision": "e6464dce15f013c2ef641593a85cc6afcdaea928"
  },
  "base": {
    "hf_id": "Qwen/Qwen3.5-4B-Base",
    "revision": "1001bb4d826a52d1f399e183466143f4da7b741b",
    "license": "Apache-2.0"
  },
  "gate": "FP32-reference argmax at margin>=0.02; finite scalars/probabilities; exact fresh-state repeats; both engines load and emit one meaningless id0 token",
  "generated_at": "2026-09-23T01:32:24.082038+00:00",
  "bundle": "gpu-pipelined/system_one_qwen3_5_4b_scorer_decode_fp16",
  "protocol": {
    "prompt_version": "State/Question/Option, last token",
    "layers": 32,
    "temperature": 1.75,
    "max_len": 384,
    "slot": "len(ids)-1",
    "head": "scalar",
    "logits_shape": [
      1,
      1,
      1
    ],
    "chat_template": "unused; add_special_tokens=False"
  },
  "fixtures": {
    "path": "fixtures-system-one-scorer-4b.json",
    "sha256": "7ad0713e165f17873f59adfd15a58ebc91860132bab26038381ce929ac0a4125",
    "requests": 20,
    "questions": 48,
    "rows": 280
  },
  "transcripts": {
    "readout": "gate-system-one-scorer-4b-readout-fp16.json",
    "engine": "gate-system-one-scorer-4b-engine-fp16.json"
  },
  "fixture_coverage": {
    "requests": 20,
    "questions": 48,
    "questions_non_zoo": 44,
    "rows": 280,
    "question_counts_non_zoo": {
      "choice": 24,
      "noul": 10,
      "score": 10
    },
    "wide_choice_questions": 6,
    "descriptive_choice_questions": 16,
    "score_ten_level_questions": 2,
    "zoo_only_questions": 4,
    "zoo_only_rows": 104,
    "states_truncated": 6,
    "truncated_request_ids": [
      "request-01",
      "request-02",
      "request-03",
      "request-04",
      "request-05",
      "request-06"
    ],
    "truncated_rows": 12,
    "state_longer_than_384_requests": 6,
    "total_tokens": 12153,
    "min_row_tokens": 24,
    "max_row_tokens": 384,
    "questions_completed": 48,
    "rows_completed": 280,
    "wall_seconds": 1214.858353166026,
    "all_finite": true,
    "scalar_min": -14.593598365783691,
    "scalar_max": 7.658686637878418,
    "scalar_max_absolute": 14.593598365783691,
    "near_tie_questions": [
      "request-07-q03",
      "request-08-q03"
    ],
    "wall": 1214.858353083022,
    "total_merge_oracle_wall_seconds": 1258.4867856669007,
    "status": "PASS"
  },
  "readout": {
    "rows": 280,
    "expected_rows": 280,
    "questions": 48,
    "expected_questions": 48,
    "argmax_agreement": 48,
    "margin_ge_0_02_questions": 46,
    "margin_ge_0_02_argmax_agreement": 46,
    "near_ties": [
      {
        "id": "request-07-q03",
        "oracle_margin": 0.006739843342914137,
        "oracle_argmax": 1,
        "argmax": 1,
        "argmax_agrees": true,
        "p": [
          0.06852844678950806,
          0.14041521453796316,
          0.13387302790675262,
          0.11745231516189299,
          0.09834065836381406,
          0.11402942509386077,
          0.11824146549641584,
          0.11257522991562542,
          0.06747071791742688,
          0.029073498816740265
        ],
        "p_oracle": [
          0.06844557046678758,
          0.1410972557943501,
          0.13435741245143595,
          0.11769923201661937,
          0.09850107499890995,
          0.11399026048335885,
          0.11797473484160344,
          0.11195394081789756,
          0.0672594693782866,
          0.028721048750750485
        ]
      },
      {
        "id": "request-08-q03",
        "oracle_margin": 0.004176430865158717,
        "oracle_argmax": 5,
        "argmax": 5,
        "argmax_agrees": true,
        "p": [
          0.03821268801456236,
          0.08558664157346375,
          0.1121254686745306,
          0.11465641135757645,
          0.10849394805786149,
          0.1272676828719541,
          0.12314573609528758,
          0.1172444835308317,
          0.10855450845152134,
          0.06471243137241071
        ],
        "p_oracle": [
          0.03829370334805408,
          0.08559956498729611,
          0.11225994259558048,
          0.11473061408817004,
          0.10857444261658075,
          0.12713889005910925,
          0.12296245919395053,
          0.11719791061365697,
          0.10862090712691694,
          0.06462156537068481
        ]
      }
    ],
    "strict_disagreements": [],
    "max_abs_delta_p": 0.003684344586315391,
    "mean_of_question_mean_abs_delta_p": 0.00020850158130700705,
    "mean_of_row_mean_abs_delta_p": 0.00020850158130700705,
    "mean_of_row_mean_note": "Each probability row is one question; options are weighted equally within it.",
    "mean_option_abs_delta_p": 0.00013447082908294965,
    "max_abs_delta_scalar": 0.06464046239852905,
    "mean_abs_delta_scalar": 0.01434520035982132,
    "scalar_min": -14.578125,
    "scalar_max": 7.65234375,
    "scalar_absmax": 14.578125,
    "oracle_scalar_min": -14.593600273132324,
    "oracle_scalar_max": 7.658699989318848,
    "oracle_scalar_absmax": 14.593600273132324,
    "finite_all": true,
    "reset_identical": true,
    "session_resets_identical": true,
    "max_row_executions_per_process": 15,
    "provisional_max_delta_p_expectation": 0.02,
    "provisional_expectation_exceeded": false,
    "provisional_expectation_is_acceptance_gate": false,
    "worst_five_questions": [
      {
        "id": "request-11-q03",
        "request_id": "request-11",
        "type": "score",
        "zoo_only": false,
        "row_ids": [
          "request-11-q03-option-00",
          "request-11-q03-option-01",
          "request-11-q03-option-02"
        ],
        "options": [
          "none",
          "some",
          "complete"
        ],
        "nopts": 3,
        "scalars": [
          -1.9833984375,
          -0.7919921875,
          0.1658935546875
        ],
        "scalars_oracle": [
          -1.9961133003234863,
          -0.7945132851600647,
          0.13405287265777588
        ],
        "p": [
          0.1564840369177385,
          0.30912866310323367,
          0.5343872999790279
        ],
        "p_oracle": [
          0.15711289848380908,
          0.3121841461234784,
          0.5307029553927125
        ],
        "argmax": 2,
        "oracle_argmax": 2,
        "argmax_agrees": true,
        "selected_option": "complete",
        "oracle_selected_option": "complete",
        "oracle_margin": 0.21851880926923412,
        "abs_delta_p": [
          0.0006288615660705876,
          0.0030554830202447203,
          0.003684344586315391
        ],
        "max_abs_delta_p": 0.003684344586315391,
        "mean_abs_delta_p": 0.002456229724210233,
        "finite": true
      },
      {
        "id": "request-14-q04",
        "request_id": "request-14",
        "type": "noul",
        "zoo_only": false,
        "row_ids": [
          "request-14-q04-option-00",
          "request-14-q04-option-01"
        ],
        "options": [
          "yes",
          "no"
        ],
        "nopts": 2,
        "scalars": [
          -1.5673828125,
          1.375
        ],
        "scalars_oracle": [
          -1.5531220436096191,
          1.3508743047714233
        ],
        "p": [
          0.156915253827473,
          0.843084746172527
        ],
        "p_oracle": [
          0.15983899956016281,
          0.8401610004398371
        ],
        "argmax": 1,
        "oracle_argmax": 1,
        "argmax_agrees": true,
        "selected_option": "no",
        "oracle_selected_option": "no",
        "oracle_margin": 0.6803220008796743,
        "abs_delta_p": [
          0.002923745732689803,
          0.002923745732689942
        ],
        "max_abs_delta_p": 0.002923745732689942,
        "mean_abs_delta_p": 0.0029237457326898725,
        "finite": true
      },
      {
        "id": "request-15-q05",
        "request_id": "request-15",
        "type": "score",
        "zoo_only": false,
        "row_ids": [
          "request-15-q05-option-00",
          "request-15-q05-option-01",
          "request-15-q05-option-02",
          "request-15-q05-option-03",
          "request-15-q05-option-04"
        ],
        "options": [
          "very low",
          "low",
          "medium",
          "high",
          "critical"
        ],
        "nopts": 5,
        "scalars": [
          -6.4296875,
          -2.8125,
          -0.83984375,
          1.4873046875,
          -0.7861328125
        ],
        "scalars_oracle": [
          -6.44263219833374,
          -2.8297576904296875,
          -0.8664757013320923,
          1.476930022239685,
          -0.7765030860900879
        ],
        "p": [
          0.006638105123733906,
          0.052446354888491795,
          0.16190714281609525,
          0.6120549619234562,
          0.16695343524822281
        ],
        "p_oracle": [
          0.006626994094001005,
          0.052229686529919314,
          0.16037686679369648,
          0.61192850142172,
          0.16883795116066314
        ],
        "argmax": 3,
        "oracle_argmax": 3,
        "argmax_agrees": true,
        "selected_option": "high",
        "oracle_selected_option": "high",
        "oracle_margin": 0.4430905502610568,
        "abs_delta_p": [
          1.1111029732901317e-05,
          0.00021666835857248046,
          0.0015302760223987744,
          0.00012646050173625856,
          0.001884515912440321
        ],
        "max_abs_delta_p": 0.001884515912440321,
        "mean_abs_delta_p": 0.0007538063649761472,
        "finite": true
      },
      {
        "id": "request-20-q01",
        "request_id": "request-20",
        "type": "choice",
        "zoo_only": true,
        "row_ids": [
          "request-20-q01-option-00",
          "request-20-q01-option-01",
          "request-20-q01-option-02",
          "request-20-q01-option-03",
          "request-20-q01-option-04",
          "request-20-q01-option-05",
          "request-20-q01-option-06",
          "request-20-q01-option-07",
          "request-20-q01-option-08",
          "request-20-q01-option-09",
          "request-20-q01-option-10",
          "request-20-q01-option-11",
          "request-20-q01-option-12",
          "request-20-q01-option-13",
          "request-20-q01-option-14",
          "request-20-q01-option-15",
          "request-20-q01-option-16",
          "request-20-q01-option-17",
          "request-20-q01-option-18",
          "request-20-q01-option-19",
          "request-20-q01-option-20",
          "request-20-q01-option-21",
          "request-20-q01-option-22",
          "request-20-q01-option-23",
          "request-20-q01-option-24",
          "request-20-q01-option-25",
          "request-20-q01-option-26",
          "request-20-q01-option-27",
          "request-20-q01-option-28",
          "request-20-q01-option-29",
          "request-20-q01-option-30",
          "request-20-q01-option-31"
        ],
        "options": [
          "lighting",
          "plumbing",
          "heating",
          "cooling",
          "doors",
          "windows",
          "floors",
          "walls",
          "ceilings",
          "roofs",
          "stairs",
          "lifts",
          "locks",
          "shelves",
          "desks",
          "chairs",
          "curtains",
          "carpets",
          "drains",
          "pipes",
          "valves",
          "tanks",
          "pumps",
          "fans",
          "sockets",
          "switches",
          "cables",
          "sensors",
          "meters",
          "signs",
          "paths",
          "gates"
        ],
        "nopts": 32,
        "scalars": [
          -7.78515625,
          -7.390625,
          -8.3125,
          -8.359375,
          2.419921875,
          -5.3359375,
          -8.9140625,
          -6.8828125,
          -8.859375,
          -8.1484375,
          -9.203125,
          -6.06640625,
          -1.203125,
          -9.8359375,
          -10.5,
          -10.265625,
          -6.25,
          -10.7421875,
          -8.1015625,
          -7.09375,
          -2.53515625,
          -7.37890625,
          -6.9140625,
          -8.125,
          -7.984375,
          -4.9453125,
          -5.60546875,
          -3.224609375,
          -8.1953125,
          -9.203125,
          -7.89453125,
          4.1953125
        ],
        "scalars_oracle": [
          -7.80185079574585,
          -7.406404495239258,
          -8.328375816345215,
          -8.3722505569458,
          2.405498504638672,
          -5.341571807861328,
          -8.924745559692383,
          -6.8783721923828125,
          -8.876729011535645,
          -8.154253005981445,
          -9.206731796264648,
          -6.098209381103516,
          -1.189605712890625,
          -9.841769218444824,
          -10.507135391235352,
          -10.270621299743652,
          -6.233147144317627,
          -10.75239372253418,
          -8.108002662658691,
          -7.108526229858398,
          -2.5550146102905273,
          -7.36898946762085,
          -6.927596569061279,
          -8.134787559509277,
          -8.004584312438965,
          -4.9726715087890625,
          -5.6035261154174805,
          -3.2340099811553955,
          -8.21119499206543,
          -9.236793518066406,
          -7.896141052246094,
          4.17133903503418
        ],
        "p": [
          0.000717858944322546,
          0.0008993924878897198,
          0.000531091226422738,
          0.0005170544010148663,
          0.2446897370635693,
          0.0029097622762010423,
          0.0003766000720946237,
          0.0012021875005124253,
          0.0003885546416473148,
          0.0005832896021279521,
          0.0003192598143164164,
          0.0019168067904573286,
          0.0308667183897491,
          0.00022238232679585153,
          0.00015216018394047576,
          0.0001739664351356079,
          0.0017259022500446,
          0.00013249447553952243,
          0.0005991245593224508,
          0.0010656736134457185,
          0.014418582331831017,
          0.0009054354158788976,
          0.0011809104060285715,
          0.0005911540626877878,
          0.0006406183025131938,
          0.0036374601297118397,
          0.002494413243690419,
          0.009723489927841967,
          0.0005678731653653913,
          0.0003192598143164164,
          0.0006743660695540914,
          0.6748564200760307
        ],
        "p_oracle": [
          0.0007191815249930839,
          0.0009015207946759931,
          0.0005323186915273457,
          0.0005191386905016756,
          0.2454589060345332,
          0.0029336055303064035,
          0.0003785922027948519,
          0.0012190362342608175,
          0.00038912384915785244,
          0.0005880083232525815,
          0.0003222490363386637,
          0.0019038304734785964,
          0.03146211983276343,
          0.00022417928797102953,
          0.00015327548882480683,
          0.00017545591105207873,
          0.0017625481986839066,
          0.0001332316396700548,
          0.0006037558356560152,
          0.0010688079711317084,
          0.014419055372251894,
          0.0009210028446555952,
          0.001185224679289517,
          0.0005945853163533855,
          0.0006405108561773736,
          0.003622021856556868,
          0.002525765151669818,
          0.009782091018493599,
          0.0005691834701096167,
          0.0003167606763826054,
          0.0006814573372852584,
          0.6732934558692004
        ],
        "argmax": 31,
        "oracle_argmax": 31,
        "argmax_agrees": true,
        "selected_option": "gates",
        "oracle_selected_option": "gates",
        "oracle_margin": 0.42783454983466723,
        "abs_delta_p": [
          1.3225806705379102e-06,
          2.128306786273364e-06,
          1.2274651046077344e-06,
          2.084289486809377e-06,
          0.0007691689709638794,
          2.3843254105361273e-05,
          1.992130700228178e-06,
          1.6848733748392204e-05,
          5.692075105376304e-07,
          4.7187211246294295e-06,
          2.989222022247329e-06,
          1.2976316978732154e-05,
          0.0005954014430143324,
          1.7969611751780042e-06,
          1.115304884331071e-06,
          1.4894759164708362e-06,
          3.664594863930673e-05,
          7.371641305323835e-07,
          4.6312763335643785e-06,
          3.1343576859898907e-06,
          4.7304042087754905e-07,
          1.55674287766976e-05,
          4.3142732609453924e-06,
          3.4312536655976703e-06,
          1.0744633582014142e-07,
          1.543827315497173e-05,
          3.135190797939891e-05,
          5.860109065163217e-05,
          1.3103047442254321e-06,
          2.499137933811011e-06,
          7.091267731167074e-06,
          0.0015629642068303307
        ],
        "max_abs_delta_p": 0.0015629642068303307,
        "mean_abs_delta_p": 9.962408632710677e-05,
        "finite": true
      },
      {
        "id": "request-13-q05",
        "request_id": "request-13",
        "type": "score",
        "zoo_only": false,
        "row_ids": [
          "request-13-q05-option-00",
          "request-13-q05-option-01",
          "request-13-q05-option-02",
          "request-13-q05-option-03"
        ],
        "options": [
          "unusable",
          "limited",
          "mostly usable",
          "fully usable"
        ],
        "nopts": 4,
        "scalars": [
          1.6240234375,
          0.2861328125,
          -1.291015625,
          -3.94921875
        ],
        "scalars_oracle": [
          1.6196714639663696,
          0.27900874614715576,
          -1.3120677471160889,
          -3.97291898727417
        ],
        "p": [
          0.5896212147307447,
          0.27450552636424547,
          0.11146868669589143,
          0.024404572209118385
        ],
        "p_oracle": [
          0.5906616660481768,
          0.274554667682575,
          0.11060483754235892,
          0.024178828726889312
        ],
        "argmax": 0,
        "oracle_argmax": 0,
        "argmax_agrees": true,
        "selected_option": "unusable",
        "oracle_selected_option": "unusable",
        "oracle_margin": 0.3161069983656018,
        "abs_delta_p": [
          0.001040451317432134,
          4.91413183295597e-05,
          0.000863849153532506,
          0.00022574348222907312
        ],
        "max_abs_delta_p": 0.001040451317432134,
        "mean_abs_delta_p": 0.0005447963178808182,
        "finite": true
      }
    ]
  },
  "engine": {
    "calls": 4,
    "expected_calls": 4,
    "calls_completed": 4,
    "one_token_calls": 4,
    "id_zero_text_matches": 4,
    "passed_calls": 4,
    "engines": {
      "coreai-pipelined": {
        "calls": 2,
        "passed_calls": 2
      },
      "coreai-sequential": {
        "calls": 2,
        "passed_calls": 2
      }
    }
  },
  "engine_probability_gate_executed": false,
  "benchmark": null,
  "round2_max_abs_delta_p": 0.0036843322011962387,
  "round3_max_abs_delta_p_difference": 1.2385119152469315e-08,
  "oracle_references": {
    "canonical_fixture_and_card_table": "Author's unmerged PEFT model, CPU FP32; accepted round2 fixture retained byte-identical",
    "linked_readout_transcript": "Freshly recomputed merged model, CPU FP32; author's unchanged encode/score_options/softmax",
    "merged_vs_peft_max_abs_delta_scalar": 3.343820571899414e-05,
    "merged_vs_peft_max_abs_delta_p": 2.6363841504339902e-06,
    "merged_vs_peft_argmax_agreement": 48
  },
  "readout_reproduction": {
    "result": "PASS",
    "canonical": "/Users/majimadaisuke/code/codex-conversions/2026-09-23/pngwn-scorer-4b-coreai/results/readout_fp16.json",
    "candidate": "/Users/majimadaisuke/code/codex-conversions/2026-09-23/pngwn-scorer-4b-coreai/results/round3/readout_fp16.json",
    "canonical_argmax_agreement": 48,
    "candidate_argmax_agreement": 48,
    "canonical_max_abs_delta_p": 0.0036843322011962387,
    "candidate_max_abs_delta_p": 0.003684344586315391,
    "max_abs_delta_p_reproduction_error": 1.2385119152469315e-08,
    "max_gpu_scalar_change": 0.0,
    "mean_gpu_scalar_change": 0.0,
    "tolerance": 1e-06
  },
  "coreai_kit_measurements": "measurements-coreai-kit.json",
  "coreai_kit_measurements_by_this_run": false
}
