{
  "measurement_date": "2026-09-27",
  "dataset": "Genentech/assaybench \u00b7 biogrid/yearfold0",
  "split": {
    "train": 1349,
    "validation": 218,
    "test": 334
  },
  "metric": "Adjusted nDCG@100",
  "references": {
    "published_embedding_knn": {
      "mean": 0.06460030169548284,
      "ci95": [
        0.03844434514093022,
        0.10169878169833363
      ]
    },
    "published_ensemble": {
      "mean": 0.16309105341547825,
      "ci95": [
        0.12435158672545148,
        0.21169275502755247
      ]
    },
    "published_gemini3pro": {
      "mean": 0.15699111156489173,
      "ci95": [
        0.11807350568432998,
        0.20855967953550772
      ]
    },
    "published_gpt54": {
      "mean": 0.14697605791551085,
      "ci95": [
        0.11011813591874998,
        0.1916820843579909
      ]
    },
    "published_oracle_knn": {
      "mean": 0.29178431087462314,
      "ci95": [
        0.2351279263367456,
        0.34609295509125787
      ]
    },
    "published_phenotype_frequency": {
      "mean": 0.13343958615780954,
      "ci95": [
        0.0779636201361144,
        0.21394591647621672
      ]
    },
    "splicr_legacy_seed0": {
      "mean": 0.13616124340731678,
      "ci95": [
        0.07962425682965675,
        0.21922114197113815
      ]
    }
  },
  "router": {
    "mean": 0.16036947764886397,
    "ci95": [
      0.12038642831605759,
      0.21120068288610572
    ],
    "paired_vs_ensemble": {
      "ci95": [
        -0.015580579751069256,
        0.006267558948259969
      ],
      "mean": -0.002721575766614277,
      "n_publications": 63,
      "n_screens": 334,
      "repeats": 10000,
      "seed": 20260927,
      "unit": "publication"
    }
  },
  "legacy_prior": {
    "andcg100": 0.13605235245977712,
    "input": "target library aware",
    "seeds": [
      0.13616124340731678,
      0.13588686189512475,
      0.13596417990114276,
      0.13607873523830272,
      0.1360571352278961,
      0.13608837737657326,
      0.13604175220224174,
      0.13614053442961888
    ],
    "train": "1349 train + 218 validation"
  },
  "counts": {
    "dataset": "GSE145743",
    "library": "geckov2-a",
    "published_table": "GSE145743_my_counts_anno_merged.txt.gz",
    "note": "The deposited table is normalised to 10,000,000 reads per column, 5,000,000 per library across the two halves it merges, so counts are compared as counts per million rather than raw. Only GeCKOv2 library A guides are compared, because the deposited columns merge libraries A and B.",
    "samples": [
      {
        "fastq": "SRR12401850_libA_plasmid.fastq.gz",
        "published_column": "plasmid",
        "guides_compared": 65383,
        "median_cpm_ratio": 0.9691,
        "fraction_within_25_percent": 0.8949,
        "spearman": 0.9289,
        "zero_guides_splicr": 1574,
        "zero_guides_published": 2661,
        "reads": 3119503,
        "mapping_rate": 0.8893
      },
      {
        "fastq": "SRR11144449_libA_T0_input.fastq.gz",
        "published_column": "input",
        "guides_compared": 65383,
        "median_cpm_ratio": 0.9723,
        "fraction_within_25_percent": 0.8885,
        "spearman": 0.9399,
        "zero_guides_splicr": 1748,
        "zero_guides_published": 3013,
        "reads": 12449815,
        "mapping_rate": 0.8863
      },
      {
        "fastq": "SRR11144453_libA_DMSO_rep1.fastq.gz",
        "published_column": "DMSO1",
        "guides_compared": 65383,
        "median_cpm_ratio": 0.9763,
        "fraction_within_25_percent": 0.8401,
        "spearman": 0.9533,
        "zero_guides_splicr": 3186,
        "zero_guides_published": 4713,
        "reads": 5896213,
        "mapping_rate": 0.8143
      },
      {
        "fastq": "SRR11144451_libA_olaparib_rep1.fastq.gz",
        "published_column": "Olaparib1",
        "guides_compared": 65383,
        "median_cpm_ratio": 0.9762,
        "fraction_within_25_percent": 0.8409,
        "spearman": 0.9548,
        "zero_guides_splicr": 4188,
        "zero_guides_published": 5734,
        "reads": 4871170,
        "mapping_rate": 0.871
      }
    ]
  },
  "postscreen": {
    "dataset": "GSE145743",
    "source": "https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE145743",
    "input_sha256": "a14192f5d84a23f08aaaa5dad5e253ffaf5e03e92b4a3a83eeacc628e13edb00",
    "n_guides": 123411,
    "n_genes": 21524,
    "original_min_directional_fdr_hits": 39,
    "corrected_two_family_fdr_hits": 22,
    "statistical_method": "min(1, 2*min(MAGeCK directional FDRs)); two-family union bound",
    "CHD1L": {
      "gene": "CHD1L",
      "n_guides": 6,
      "n_good_guides": 5,
      "lfc": -1.5486,
      "rra_score": 0.00035339,
      "p_value": 0.0037394,
      "fdr": 1.0,
      "rank": 67,
      "direction": "depleted",
      "mle_beta": null,
      "mle_fdr": null,
      "bayes_factor": null,
      "norm_z": -3.92,
      "drugz_fdr": 0.161,
      "guide_lfcs": [
        -3.3551,
        -3.8542,
        -1.618,
        -1.4791,
        -0.85796,
        0.32767
      ],
      "depleted_p_value": 0.0018697,
      "enriched_p_value": 0.91779,
      "depleted_fdr": 0.559153,
      "enriched_fdr": 1.0
    },
    "CHD1L_rank": 313
  },
  "postscreen_qc": "fail",
  "sources_sha256": {
    "research/artifacts/official_references.json": "9d64ca66c4627ce4307aee594d6607885f7ed7271d326f95ada9a3ca23db2e2e",
    "research/artifacts/router_replay_summary.json": "20fd78f34a0ba45dc8cbcea9d24a65b0d415fc6776dcd7afa47b1eb493cd228d",
    "research/artifacts/raw_count_reproduction.json": "3a770f4f996c124466c2579633cb364650d6cf1f8366c8e36a69adace794f226",
    "research/artifacts/postscreen_unpaired_audit.json": "697d272cf7c48ca169b3d6d8c480fc0783e657484353c1eccc132c5034222cc2",
    "research/artifacts/20260928/publication/experiment_D_descriptor.json": "1421beab7740ea245f8f7dd0cb40e28fba2a65c69a2d58bb5e01b4a38daa19eb",
    "research/artifacts/20260928/publication/experiment_E_descriptor.json": "5ab0a8018d58e99b72476c4145137dbc98dc9cb36e2d90ada79b5671eecdcc13",
    "research/artifacts/20260928/publication/experiment_R_descriptor.json": "72b3de11619b0451c454b076ad3a67ebaf2791e49655c6c25277a7d69d27a3a6"
  },
  "limitations": [
    "Public retrospective test; independent prospective validation outstanding.",
    "No demonstrated superiority over the published ensemble.",
    "No calibrated probability of independent validation.",
    "Count agreement and software tests do not certify biological accuracy."
  ],
  "research_addendum": {
    "measurement_date": "2026-09-28",
    "split": "validation_2021 (218 screens, 34 publications)",
    "note": "Development results on the 2021 validation split, not the 334-screen public test. Neither experiment established an improvement and neither model is promoted.",
    "experiments": {
      "20260928_experiment_D": {
        "title": "Gene-level LambdaRank over 57 historical-context features",
        "selected_model": "E2_history_lambdarank",
        "split": "validation_2021",
        "evaluation_kind": "validation_selection",
        "input_contract": "metadata_only",
        "promotion_status": "research_only",
        "mean": 0.18098516821887162,
        "ci95": [
          0.1133776654908188,
          0.25951374682060274
        ],
        "n_screens": 218,
        "n_publications": 34,
        "comparison": "Published frontier ensemble on the same 218 screens: 0.186437. Paired difference -0.005452, 95% publication-cluster interval [-0.016147, +0.010805] over 10,000 resamples.",
        "finding": "No improvement established: the interval contains zero and the point estimate is below the ensemble. Selection multiplicity 14."
      },
      "20260928_experiment_E": {
        "title": "Requested-effect polarity conditioning",
        "selected_model": "P0_baseline_D3",
        "split": "validation_2021",
        "evaluation_kind": "validation_selection",
        "input_contract": "metadata_only",
        "promotion_status": "research_only",
        "mean": 0.09945010274065483,
        "ci95": [
          0.03114703090902138,
          0.217460825450562
        ],
        "n_screens": 218,
        "n_publications": 34,
        "comparison": "Cross-validation selected the unconditioned baseline. The polarity prior scored -0.007931 [-0.023172, +0.004531] against it; the polarity-plus-transfer variant -0.010680 [-0.024030, +0.001069].",
        "finding": "The parser separates all 47 metadata-matched increase/decrease comparisons that previous features collapsed. That is a representation improvement, not an established ranking improvement."
      }
    }
  },
  "post_screen_replication": {
    "measurement_date": "2026-09-28",
    "benchmark": "SplicR cross-screen replication benchmark: 124 held-out screen pairs from different publications, first authors and libraries, same cell line, unperturbed proliferation",
    "note": "Given a screen a lab has already run, rank its candidates by how likely each is to be called a hit in an independent screen. This is a different task from the pre-screen prediction table and the two are not comparable.",
    "cohort": {
      "benchmark_artifact_sha256": "d525ec0a8dd30ac95a61acf732ec34e7ed75c6070898b248e4c045c67701558a",
      "n_screen_pairs": 124,
      "n_units": 248
    },
    "primary_model": "lr_effect_freq",
    "comparator": "effect_score1",
    "metric": "average precision, non-common-essential gene space",
    "primary_mean": 0.3538029148717251,
    "comparator_mean": 0.24636521237071196,
    "paired_difference": {
      "ci95": [
        0.10197492297346228,
        0.11278569643377552
      ],
      "mean_difference": 0.10743770250101319,
      "n_screen_pairs": 124,
      "n_units": 248,
      "resampling_unit": "pair_key",
      "significant": true,
      "wilcoxon_p": 4.330109895931127e-22
    },
    "precision_at_10": {
      "comparator": 0.6963709677419354,
      "paired_difference": {
        "ci95": [
          0.2004032258064516,
          0.2383064516129032
        ],
        "mean_difference": 0.21895161290322582,
        "n_screen_pairs": 124,
        "n_units": 248,
        "resampling_unit": "pair_key",
        "significant": true,
        "wilcoxon_p": 5.725557880076196e-22
      },
      "primary": 0.9153225806451614
    },
    "non_hub_paired_difference": {
      "ci95": [
        0.04736208500111694,
        0.09793373313036204
      ],
      "mean_difference": 0.07212556753352574,
      "n_screen_pairs": 11,
      "n_units": 22,
      "resampling_unit": "pair_key",
      "significant": true,
      "wilcoxon_p": 0.0009765625
    },
    "promotion_status": "research_only",
    "limitations": [
      "Cross-screen replication is a proxy for reproducibility, not proof that a candidate passes an independent biological validation assay.",
      "The target screen's hit call depends on its own significance criteria; a real effect can fail to replicate because that threshold was stricter.",
      "113 of the 124 held-out pairs are one library comparison (Behan 2019 against Meyers 2017); the 11 remaining pairs are reported separately.",
      "Unperturbed proliferation screens only. Nothing is established for drug-modifier, host-pathogen or reporter screens.",
      "This is a ranking, not a calibrated probability that any individual gene validates."
    ]
  }
}