CIViC-Fact v3 within-linked-publication passage retrieval, post-cutoff cohort
For each post-cutoff CIViC entry with main-text-eligible evidence, the source publication is already known (CIViC's existing link from claim to paper). A passage-retrieval model ranks candidate passages from that single known paper and returns the top 5 as candidate evidence for the claim. The retrieved top-5 passages are then manually reviewed, per entry, for whether they contain appropriate (sufficient) evidence for the claim. The source separately reports that a curator, with group-consensus discussion for ambiguous cases, assigned the cohort's reference labels (Methods, 'Temporal evaluation'); that process description does not by itself specify the reviewer count, blinding, or replicate structure for the retrieval-success judgment reported in Results, and none of those is assumed here — recorded as unreported. This is retrieval of passages within one already-identified publication, not a literature-search or citation-context benchmark that finds which paper, among an open corpus, is relevant to a claim or question. It is not a stance/claim-verification (SUPPORTS/REVISIONS/NEI) benchmark on its own, and not the static, separately-scoped retrieval-subset evaluation (n_test=1120) reported elsewhere in the same source.
Overview
For each post-cutoff CIViC entry with main-text-eligible evidence, the source publication is already known (CIViC's existing link from claim to paper). A passage-retrieval model ranks candidate passages from that single known paper and returns the top 5 as candidate evidence for the claim. The retrieved top-5 passages are then manually reviewed, per entry, for whether they contain appropriate (sufficient) evidence for the claim. The source separately reports that a curator, with group-consensus discussion for ambiguous cases, assigned the cohort's reference labels (Methods, 'Temporal evaluation'); that process description does not by itself specify the reviewer count, blinding, or replicate structure for the retrieval-success judgment reported in Results, and none of those is assumed here — recorded as unreported. This is retrieval of passages within one already-identified publication, not a literature-search or citation-context benchmark that finds which paper, among an open corpus, is relevant to a claim or question. It is not a stance/claim-verification (SUPPORTS/REVISIONS/NEI) benchmark on its own, and not the static, separately-scoped retrieval-subset evaluation (n_test=1120) reported elsewhere in the same source.
Consult the linked sources for architecture or protocol details. Missing evidence is not evidence of a missing capability.
2 recorded evaluations, 2 metric rows. A comparison chart has not yet been validated for these results. The table retains the individual findings and their sources.
Results
Results are available, but no reviewed comparison panel is linked in this release.
All evaluations
2 evaluations · 2 results. Different protocols are not a single leaderboard.
Filter evaluations
Applied filters: All linked evaluations
| Tested configuration | Protocol and dataset | Finding | Evidence and details |
|---|---|---|---|
| Configuration: Fine-tuned MedCPT cross-encoder (passage retrieval) | Protocol: CIViC-Fact v3 within-linked-publication passage retrieval, post-cutoff cohort Dataset: CIViC-Fact v3 post-cutoff temporal-evaluation cohort, 2026-03-03 to 2026-06-09 | 37/40 (92.5%) appropriate_retrieval_rate_postcutoff percent · higher Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceCIViC-Fact v3 post-cutoff within-linked-publication retrieval: fine-tuned MedCPT cross-encoder Not reported Aggregation: Not reported CIViC-Fact: a proof-of-concept framework for AI-assisted verification of cancer variant interpretations (bioRxiv v3) · Results, 'Passage Retrieval Models Perform Well without Fine-Tuning' subsection (printed page 14 / PDF page 15): '...both the fine-tuned MedCPT model and the 8B Qwen 3 reranker performed similarly retrieving appropriate content for 37 of the remaining 40 entries (92.5%).' |
| Configuration: Qwen3-Reranker-8B (pretrained, not fine-tuned) | Protocol: CIViC-Fact v3 within-linked-publication passage retrieval, post-cutoff cohort Dataset: CIViC-Fact v3 post-cutoff temporal-evaluation cohort, 2026-03-03 to 2026-06-09 | 37/40 (92.5%) appropriate_retrieval_rate_postcutoff percent · higher Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceCIViC-Fact v3 post-cutoff within-linked-publication retrieval: Qwen3-Reranker-8B (pretrained) Not reported Aggregation: Not reported CIViC-Fact: a proof-of-concept framework for AI-assisted verification of cancer variant interpretations (bioRxiv v3) · Results, 'Passage Retrieval Models Perform Well without Fine-Tuning' subsection (printed page 14 / PDF page 15): '...both the fine-tuned MedCPT model and the 8B Qwen 3 reranker performed similarly retrieving appropriate content for 37 of the remaining 40 entries (92.5%).' |
Source checking is not independent reproduction. Release 2026-10-07-1448159e6a81.
Methods and evaluation design
Procedure, tasks and evaluated configurations
Recorded evaluations
Each evaluation records what was tested and under which conditions.
Baseline coverage
Reference methods help show what a model adds beyond simple controls. We track a null control and a conventional method for each protocol.
0 of 2 active baseline roles have published Rewire measurements in this release. Measurements on a selected protocol do not establish coverage of an entire suite.
No execution recipe linked to this protocol. Recipe availability does not establish a completed evaluation.
- Author-reported evaluations
- 2
Literature evidence is not a Rewire measurement. Executed but unpublished runs and private review status are not included.
Null control
Proposed control: requires review
Protocol-specific valid geometric or structural control
Protocol-specific applicability, permitted inputs, access, split, evaluator and execution requirements need review before implementation or execution.
This is a suggested selection rule, not a validated method or a measured score.
Conventional reference
Proposed control: requires review
Upstream conventional structural reference with matched templates and cutoffs
Protocol-specific applicability, permitted inputs, access, split, evaluator and execution requirements need review before implementation or execution.
This is a suggested selection rule, not a validated method or a measured score.
Protocol coverage CSV · Model evaluation matrix · Source table · Release and checksums
Coverage is derived from release 2026-10-07-1448159e6a81. Source citations describe the original records; they do not validate an unreviewed baseline proposal. No results have been generated by this audit.
Run instructions
No runnable recipe has been reviewed for this protocol. Dataset access, model requirements, licences and compute requirements must be checked against its sources before execution.
Strengths, limitations and unresolved questions
Evidence
Source checking verifies the cited claim or transcription. It does not establish independent reproduction.
Evidence table
Inspect claims, sources and review details
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.
0 evidence rows matching the loaded filters
| Property and statement | Original source and location | Review and provenance |
|---|
No evidence rows match these filters. Choose another scope or clear the search.
Sources and history
View linked audit checks and correction history
Release 2026-10-07-1448159e6a81 · Record review: source checked
1 source records and release history
- CIViC-Fact: a proof-of-concept framework for AI-assisted verification of cancer variant interpretations (bioRxiv v3) · Original source · bioRxiv preprint doi:10.1101/2025.09.10.675443, Version 3, posted 2026-08-29 (per in-document header and bioRxiv API version metadata); published: NA.
Technical metadata and extraction receipts
Stable ID: ucc-clinical-egfr-protocol-civicfact-v3-retrieval
- areas
- dna-genomes
- contexts
- clinical_research
- protocol
- Fixed known-source-paper passage retrieval: claim input plus the single publication CIViC already links to it; retriever ranks passages and returns the top 5; entries requiring supplementary material or images are excluded (NEI) before scoring; the retrieved top-5 passages are then manually reviewed per remaining entry for 'appropriate' (sufficient) evidence. No automated/recomputed metric and no source-search or citation-context benchmark.
- review
- method: automated_source_review; reviewer: Claude Sonnet EGFR-evidence-research worker; date: 2026-10-07; note: Source-backed literature curation with independent automated transcription and scope checks, independently reviewed by Codex across two review cycles before ingestion. No new model execution, independent experimental replication, qualified human scientific review or clinical validation.; source id: ucc-clinical-egfr-source-civicfact-v3
- limitations
- Pan-cancer protocol; no EGFR- or NSCLC-specific subgroup or score.; Single post-cutoff temporal cohort (2026-03-03 to 2026-06-09); not an independent replication and not a second independent cohort.; Partial coverage: only 40 of the original 150 candidate entries are scored, after sequential exclusions for full-text access, source overlap, data error, and evidence sufficiency.; Enriched cohort per the source's own Discussion; not representative of CIViC overall.; Manual review judgment of 'appropriate' retrieved content; not an automated or recomputed metric, and no confidence interval is reported or recomputed here.; Reviewer count, blinding, and replicate/seed structure for this manual retrieval-success judgment are UNREPORTED in the source; the separately reported curator/group-consensus process for assigning the cohort's reference labels (Methods) does not specify these for the retrieval judgment, and none is assumed.; Status is source_checked (author-reported, transcribed); not reproduced by any Rewire execution.; Does not measure or substitute for the benchmark-execution gap tracked separately at rewire-benchmarks #28.