Data and code for: Circularity in automated ACMG variant classification

# Data and code for: Circularity in automated ACMG variant classification Deposit accompanying the manuscript submitted to *Genetics in Medicine*. Every quantitative statement in the manuscript is recomputed from these files by the scripts under `scripts/`. The verification scripts each carry a positive and a negative control and report both on every run. ## What is here | Path | Records | Contents | |---|---|---| | `data/erepo_reference_set.jsonl` | 9653 | ClinGen Evidence Repository extract, queried 2026-08-01: expert-panel verdict plus the ACMG/AMP codes each panel applied | | `data/erepo_nonsnv_resolved.jsonl` | 992 | Non-SNV records whose reference/alternate representation was resolved against the ClinVar VCF (GRCh38, release 2026-07-28) | | `data/classifier_output.jsonl` | 9343 | GeneBe public batch API output, 2026-08-03, reduced to the fields required to reproduce the analysis | | `data/tool_survey.json` | 13 | Documentation survey, one record per classifier: retrievability from three independent sources and per-term hits. The count is the number of classifiers, held under the "tools" key | | `data/erepo_curation_dates.json` | 12950 | eRepo curation dates used by the temporal stratification | | `data/clinvar_first_directional.json` | 6436 | First directional ClinVar classification per variant, from ten annual archival releases 2017-2026 | | `data/clinvar_scan_20260818.log` | - | Scan log for the archival-release sweep | | `data/clinvar_review_status.json` | 9339 | ClinVar review status per variant, as resolved | | `data/second_engine_subset.jsonl` | 230 | The 230-variant balanced subset on which the second rule-based engine was assessed (in-sample with respect to its own tuning; see SR5) | | `data/second_engine_benchmark.jsonl` | 230 | Second-engine benchmark output on that subset | | `scripts/` | 32 files | Acquisition and scoring (12), screens and per-panel tests (11), display items (3), verification (6) | | `PREREGISTRATION_temporal_stratification.md` | - | Written before the temporal data were retrieved | ## What is deliberately not here The classifier output is reduced to the fields required to reproduce the analysis: the applied criteria with their strengths, the returned verdict, the scores, and the coordinates needed to join records back to the reference set. Anyone can regenerate the full response with `scripts/fetch_genebe.py` and an API key. No patient data of any kind is included. The parent repository is a clinical pipeline; this deposit was built from an explicit allowlist and scanned before packaging. ## Sources - ClinGen Evidence Repository, queried 2026-08-01 - ClinVar VCF, GRCh38, release 2026-07-28, plus ten annual archival releases 2017-2026 - GeneBe public batch API v1, queried 2026-08-03 ## Licence Code under `scripts/`: MIT. Data files: CC BY 4.0. The underlying ClinGen and ClinVar records are public-domain US government or ClinGen-licensed content and are redistributed here unmodified except where a column reduction is noted above.