Replace the release-scale multi-table layout with one scalar clinvar table. record_kind identifies variations, alleles, locations, genes, RCVs, SCVs, conditions, observations, citations, attributes, text, and flat-import policy decisions. Repeated source elements remain rows rather than nested values. XML-derived policy decisions remain SQL views and are materialized only by an explicit export. The former public relation names remain query-compatible views.
Add rclinvarbitration_import_flat() as the compact default import. It joins official variant_summary and submission_summary reports directly into the canonical table. The XML importer now feeds the same table when richer source evidence is needed. Release-scale flat imports execute one bounded record-kind projection at a time so large deduplication and policy states do not remain live together.
Retain GRCh37 and GRCh38 locations side by side. clinvar_vcf exposes their VCF CHROM/POS/REF/ALT tuples, preserves exact sequence accessions, keeps alternate placements distinct, and does not collapse X/Y PAR placements. Source locations lacking a complete VCF tuple remain in clinvar. Allele-level decisions and allele-gene facts deduplicate at their semantic keys even when one allele has multiple placements or a source gene list repeats a token.
Include alternate-accession decisions in the seven-column compatibility export using their official ClinVar VCF CHROM. A complete March 2026 audit found 40 rows for which pinned upstream ClinVarbitration relabels an alternate-locus position as chr17, chr19, chr22, or chr9; NCBI’s archived GRCh38 VCF and clinvar_vcf agree exactly on accession, POS/REF/ALT, VariationID, and AlleleID.
Add rclinvarbitration_publish_ducklake() as the single key-based publication path for tidy ClinVar exports. It registers Parquet without collecting rows in R, leaves unchanged records untouched, publishes inserts, updates, and withdrawals in one snapshot, and returns native DuckLake change counts.
Use temporary spill-backed XML staging and project each record kind as one contiguous block. Staging no longer leaves its high-water mark as free blocks in the durable database. A legacy-layout guard prevents old base tables from being overwritten by compatibility views.
Give every canonical row a stable record_key. Tidy Parquet retains the repeated release label as release_id, returns the release receipt separately, contains no nested columns, and is compared exactly by DuckLake.
Add clinvar_gene_disease_summaries with descriptive, policy-versioned ClinVar evidence strata. These support retrieval and temporal reanalysis but are not represented as gene-validity classifications.
Add case-insensitive submitter exclusions and named policy profiles without deleting imported source submissions.
Add clinvar_hpo_terms, clinvar_literature_links, clinvar_semantic_documents, and disease-aware clinvar_gene_summaries for source-attributed retrieval and VariantStory integration.
Add native x86-64 Windows extension builds using Rtools-provided libxml2, zlib, and target-aware pkg-config; retain Linux, macOS, and webR builds. Runtime artifact selection now matches exact DuckDB platform metadata and includes both windows_amd64 and R-devel’s windows_amd64_mingw identities.
Execute the pinned upstream TSV algorithm on exact March 2026 flat inputs: all 4,125,389 keys and values match the package reproducer. Classify all 377 XML/flat key or value differences with source-row receipts, and quantify sample/method/observed-data/consequence XML structure coverage.
Add release-differential tests for disease keys, SCV replacement, withdrawn assertions, compound alleles, and mitochondrial locations, plus curated real HPO and PubMed context projections.
Add experimental webR/WebAssembly support. The package now builds its version-matched DuckDB extension as an Emscripten side module and has a browser smoke test that loads it and imports the compressed VCV fixture.
Project compact parser rows with the package-owned rclinvar_json_field() scalar rather than DuckDB’s separately downloadable JSON extension. Imports require no extension download, including in browser/webR runtimes.
Remove the premature local v1 policy suffix; preserve the pinned cpg-clinvarbitration-2.2.11 identifier, source-order strong-review rule, and separate disease- and allele-level decision views.
Retain imported SCV source order for deterministic strong-review decisions.
Bundle exact C_STRUCT_UNSTABLE extension artifacts for DuckDB v1.5.0 through v1.5.5, selected from the enabled connection’s engine version.
RClinVarbitration 0.1.0
First public release.
Added a package-owned DuckDB C extension with clinvar_xml_statements(path), a single-threaded libxml2 forward scan of ClinVar VCV XML and XML.GZ releases.
Added semantic SQL materialization for ordered XML nodes, edges, literals, and discovery text, with a real NCBI VCV XML.GZ fixture.
The first artifact supports exactly DuckDB v1.5.3 on Unix-like hosts; it fails closed for other engine versions.