The test cohort of the AssayBench temporal split.
These checks assess whether the evidence supports a reproducible investigation. A source-checked score alone does not meet these requirements.
Release 2026-10-10-7fcc3e48a123 · Evidence verified: Not verified
Exact outcomes, predictions, identifiers and evaluator are connected.
Missing or unresolved evidence
Verified: Not verified
Replay evidence includes annotations and an assessment of dependence. Unknown independence permits descriptive analysis only.
Missing or unresolved evidence
Verified: Not verified
A pinned recipe describes the inputs, environment and resource requirements.
Missing or unresolved evidence
Verified: Not verified
Separate data and exposure records support an independent test.
Missing or unresolved evidence
Verified: Not verified
Readiness describes the evidence in this release. Availability on your computer is checked separately when an investigation runs. Existing data exposure can prevent independent validation even when files are available.
No verified artifact manifest is connected to this record yet. The gaps above identify what is needed before analysis can begin.
16 evaluations · 48 results. Different protocols are not a single leaderboard.
Applied filters: All linked evaluations
| Tested configuration | Protocol and dataset | Finding | Evidence and details |
|---|---|---|---|
| Configuration: Biomni A1 (Claude 4) (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.108 adjusted-ndcg-at-100 unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceBiomni A1 (Claude 4) on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'Biomni A1 (Claude 4)' cohort 'test', column 'AnDCG@100' |
| Configuration: Biomni A1 (Claude 4) (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.0176 directional-false-discovery-rate-at-100 fraction · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceBiomni A1 (Claude 4) on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'Biomni A1 (Claude 4)' cohort 'test', column 'dFDR@100' |
| Configuration: Biomni A1 (Claude 4) (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.17 precision-at-100 fraction · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceBiomni A1 (Claude 4) on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'Biomni A1 (Claude 4)' cohort 'test', column 'Precision@100' |
| Configuration: C2S (Gemma-2B) (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.0355 adjusted-ndcg-at-100 unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceC2S (Gemma-2B) on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'C2S (Gemma-2B)' cohort 'test', column 'AnDCG@100' |
| Configuration: C2S (Gemma-2B) (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.0218 directional-false-discovery-rate-at-100 fraction · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceC2S (Gemma-2B) on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'C2S (Gemma-2B)' cohort 'test', column 'dFDR@100' |
| Configuration: C2S (Gemma-2B) (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.0863 precision-at-100 fraction · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceC2S (Gemma-2B) on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'C2S (Gemma-2B)' cohort 'test', column 'Precision@100' |
| Configuration: Embedding kNN (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.0646 adjusted-ndcg-at-100 unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEmbedding kNN on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'Embedding kNN' cohort 'test', column 'AnDCG@100' |
| Configuration: Embedding kNN (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.0226 directional-false-discovery-rate-at-100 fraction · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEmbedding kNN on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'Embedding kNN' cohort 'test', column 'dFDR@100' |
| Configuration: Embedding kNN (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.106 precision-at-100 fraction · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEmbedding kNN on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'Embedding kNN' cohort 'test', column 'Precision@100' |
| Configuration: Gemini 3 Flash (GEPA) (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.141 adjusted-ndcg-at-100 unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceGemini 3 Flash (GEPA) on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'Gemini 3 Flash (GEPA)' cohort 'test', column 'AnDCG@100' |
| Configuration: Gemini 3 Flash (GEPA) (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.0214 directional-false-discovery-rate-at-100 fraction · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceGemini 3 Flash (GEPA) on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'Gemini 3 Flash (GEPA)' cohort 'test', column 'dFDR@100' |
| Configuration: Gemini 3 Flash (GEPA) (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.199 precision-at-100 fraction · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceGemini 3 Flash (GEPA) on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'Gemini 3 Flash (GEPA)' cohort 'test', column 'Precision@100' |
| Configuration: Gemini 3 Flash (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.145 adjusted-ndcg-at-100 unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceGemini 3 Flash on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'Gemini 3 Flash' cohort 'test', column 'AnDCG@100' |
| Configuration: Gemini 3 Flash (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.018 directional-false-discovery-rate-at-100 fraction · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceGemini 3 Flash on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'Gemini 3 Flash' cohort 'test', column 'dFDR@100' |
| Configuration: Gemini 3 Flash (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.201 precision-at-100 fraction · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceGemini 3 Flash on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'Gemini 3 Flash' cohort 'test', column 'Precision@100' |
| Configuration: Gemini 3 Pro (Few-shot) (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.154 adjusted-ndcg-at-100 unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceGemini 3 Pro (Few-shot) on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'Gemini 3 Pro (Few-shot)' cohort 'test', column 'AnDCG@100' |
| Configuration: Gemini 3 Pro (Few-shot) (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.0213 directional-false-discovery-rate-at-100 fraction · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceGemini 3 Pro (Few-shot) on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'Gemini 3 Pro (Few-shot)' cohort 'test', column 'dFDR@100' |
| Configuration: Gemini 3 Pro (Few-shot) (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.213 precision-at-100 fraction · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceGemini 3 Pro (Few-shot) on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'Gemini 3 Pro (Few-shot)' cohort 'test', column 'Precision@100' |
| Configuration: Gemini 3 Pro (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.157 adjusted-ndcg-at-100 unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceGemini 3 Pro on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'Gemini 3 Pro' cohort 'test', column 'AnDCG@100' |
| Configuration: Gemini 3 Pro (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.0164 directional-false-discovery-rate-at-100 fraction · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceGemini 3 Pro on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'Gemini 3 Pro' cohort 'test', column 'dFDR@100' |
| Configuration: Gemini 3 Pro (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.223 precision-at-100 fraction · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Independent external evaluation · Source checkedMethods, coverage and sourceGemini 3 Pro on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'Gemini 3 Pro' cohort 'test', column 'Precision@100' |
| Configuration: Gene-frequency (by phenotype) (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.133 adjusted-ndcg-at-100 unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceGene-frequency (by phenotype) on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'Gene-frequency (by phenotype)' cohort 'test', column 'AnDCG@100' |
| Configuration: Gene-frequency (by phenotype) (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.0292 directional-false-discovery-rate-at-100 fraction · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceGene-frequency (by phenotype) on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'Gene-frequency (by phenotype)' cohort 'test', column 'dFDR@100' |
| Configuration: Gene-frequency (by phenotype) (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.22 precision-at-100 fraction · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceGene-frequency (by phenotype) on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'Gene-frequency (by phenotype)' cohort 'test', column 'Precision@100' |
| Configuration: Gene-relevance predictor (AssayBench) | Protocol: AssayBench: rank 100 candidate genes for a described CRISPR screen Dataset: AssayBench test cohort of human CRISPR screens | 0.0565 adjusted-ndcg-at-100 unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceGene-relevance predictor on the AssayBench test cohort tgtval-20261009-protocol-debrouwer2026-screen-gene-ranking Aggregation: Not reported AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents · Table 3, row 'Gene-relevance predictor' cohort 'test', column 'AnDCG@100' |
Source checking is not independent reproduction. Release 2026-10-10-7fcc3e48a123.
A dataset supplies biological observations. The evaluation protocol defines how those observations are split, used and scored.
Source checking verifies the cited claim or transcription. It does not establish independent reproduction.
Trace each statement to its source and review. A context-only reference supports the record generally; it does not verify an individual field. Source checking does not reproduce an experiment.
One row per statement and cited source. Multiple citations are not independent evaluations. Shared locators are labelled explicitly.
7 evidence rows matching the loaded filters
| Property and statement | Original source and location | Review and provenance |
|---|---|---|
| attributes.assay Pooled human CRISPR screens with the screen's own significance criterion Context-only references | AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents Sections 2.1 to 2.3; Table 1 (split statistics) Version: arXiv:2605.10876 version 1, posted 2026-05-11; not peer reviewed | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: Hash scope: pdftotext -layout text layer, parsed by extract/extract_target_validation.py |
| attributes.population 334 benchmark entries from screens published after 2021 Context-only references | AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents Sections 2.1 to 2.3; Table 1 (split statistics) Version: arXiv:2605.10876 version 1, posted 2026-05-11; not peer reviewed | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: Hash scope: pdftotext -layout text layer, parsed by extract/extract_target_validation.py |
| attributes.scope_note Each entry is one screen cast as a gene-ranking task. Hits are the genes meeting that screen's significance criterion; non-hits are assayed genes that do not meet it and are given relevance 0; for AnDCG@100, genes the screen did not assay are removed from the ranked list rather than counted as false positives; the printed Precision@100 and dFDR@100 formulas do not state that step. In screens with a bidirectional criterion, genes significant in the opposite direction get negative relevance. Context-only references | AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents Sections 2.1 to 2.3; Table 1 (split statistics) Version: arXiv:2605.10876 version 1, posted 2026-05-11; not peer reviewed | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: Hash scope: pdftotext -layout text layer, parsed by extract/extract_target_validation.py |
| attributes.source_locator Sections 2.1 to 2.3; Table 1 (split statistics) Context-only references | AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents Sections 2.1 to 2.3; Table 1 (split statistics) Version: arXiv:2605.10876 version 1, posted 2026-05-11; not peer reviewed | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: Hash scope: pdftotext -layout text layer, parsed by extract/extract_target_validation.py |
| attributes.version AssayBench, arXiv:2605.10876v1; built from the 2025 BioGRID ORCS release and 19 recent publications Context-only references | AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents Sections 2.1 to 2.3; Table 1 (split statistics) Version: arXiv:2605.10876 version 1, posted 2026-05-11; not peer reviewed | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: Hash scope: pdftotext -layout text layer, parsed by extract/extract_target_validation.py |
| description The test cohort of the AssayBench temporal split. Context-only references | AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents Sections 2.1 to 2.3; Table 1 (split statistics) Version: arXiv:2605.10876 version 1, posted 2026-05-11; not peer reviewed | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: Hash scope: pdftotext -layout text layer, parsed by extract/extract_target_validation.py |
| name AssayBench test cohort of human CRISPR screens Context-only references | AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents Sections 2.1 to 2.3; Table 1 (split statistics) Version: arXiv:2605.10876 version 1, posted 2026-05-11; not peer reviewed | not individually reviewed No individual claim review recorded Audit detailsField: Source artifact SHA-256: Hash scope: pdftotext -layout text layer, parsed by extract/extract_target_validation.py |
Release 2026-10-10-7fcc3e48a123 · Record review: source checked
Stable ID: tgtval-20261009-data-assaybench-test