Skip to content

RM index, items 300–399

One page of the RM index, which is paged by number: every item numbered 300–399. The rules for adding an entry, and the allocator that claims a number, are on the index page.

⏳ Open, no release decided — ROADMAP.md § Active items

  • ⏳ RM399 — open, a minor, 0.10's scope (the pm, 2026-11-16), owner format, severity medium. needs_recompile has no axis for a reference lane the enricher now reads at another release, so 0.8's snapshot switch (RM372), which corrected RM267's insertions, is declared in prose only and a consumer's sweep hears nothing. From S132. related RM372, RM267, RM126, RM181
  • ⏳ RM396 — open, a minor, parked (the pm, 2026-11-05: the reporter's interim works), owner enricher, severity low. Six paced clients (pgs, litvar, civic_api, clingen_allele, pharmvar, atlas_client) keep their default interval or their gate private, and clingen_allele takes no gate=, so a host sharing one gate reads a private name. From S131.
  • ⏳ RM390 — open, a minor, parked (the pm, 2026-11-05: until an author meets it), owner enricher, severity low. RM374's placement check runs only in compile, so enrich commits a table that compile --strict then refuses, and RM373's mixed-release reason shows in one arm only. Designed after RM387. related RM374, RM373, RM387
  • ⏳ RM388 — open, a minor, parked (the pm, 2026-11-05: until a real table reaches it), owner compiler, severity low. Membership, hosting and reference readers skip usable_loci, so they filter neither the build nor not_found; only a hand-written or multi-build table reaches them. related RM362
  • ⏳ RM385 — open, a minor, 0.10's scope (the pm, 2026-11-16), owner enricher, severity low. --offline makes SequenceProxy return no base at all, so an offline run cannot left-align an indel even with a local reference to hand; an offline run should read a configured local reference and withhold only when none is. Open: a local seqrepo or a GRCh38 FASTA, and where it is configured. S157's online half is RM404's. related RM273, RM381, RM382
  • ⏳ RM378 — open, a minor, parked (the pm, 2026-11-05; placed in 0.9 the same day), owner compiler + format, severity low. Errors get what RM131 gave warnings: a code from a closed vocabulary and an errors_summary. Open: pydantic's own validation errors have no closed vocabulary, and RM131 refuses a partial summary. · from S129 · related RM377, RM131, RM218
  • ⏳ RM360 — open, a minor, parked (the pm, 2026-11-16), owner format, severity low. Four v2 facts with no cell: an allele OR on meta.csv, an HLA tag SNP, a GenAge entry id, the proxy allele phase. related RM323, RM328, RM334, RM335
  • ⏳ RM358 — open, a minor, parked (the pm, 2026-11-16), owner enricher + compiler, severity low. DOI-only literature answers are pinned in literature.csv so the compiler's checks see them; bioRxiv's jatsxml as a full-text source. related RM349, RM336
  • ⏳ RM350 — open, a minor, parked (the pm, 2026-11-16), owner enricher, severity low. The IPD-IMGT/HLA allele list as a witness lane. related RM334
  • ⏳ RM346 — open, 1.0, owner compiler + format, severity low — a normalized, SQL-friendly parquet shape of the denormalized, human-centred authored tables; the additive half could land in a minor
  • ⏳ RM345 — open, a minor, parked (the pm, 2026-11-16), owner format + enricher, severity medium. Holder eligibility, a no-training bar, revocability with deletion, notice of modification, and a host licence deferring to owners a row cannot name (Open Targets L2G). related RM333, RM27
  • ⏳ RM344 — open, a patch, unattended, owner enricher, severity low. A .duckdb cache a newer duckdb wrote makes connect raise; fall back to the parquet data/ as for an empty db, with a warning naming the storage version.
  • ⏳ RM321 — open, a patch, owner format, severity low. The flags description says conditional|phased|pleiotropic are tags the tooling acts on; the only reader is the INFO line that excludes them, and phased is derived from the genotype pipe. Found reading longevitymap v2's plan, which relies on both. · related RM28 · for longevitymap v2
  • ⏳ RM319 — open, a minor, taken into 0.8 on 2026-09-30, design first, severity medium. TRGT writes the repeat count as FORMAT/MC, a per-motif String; repeat_alleles.csv modules point at ExpansionHunter's REPCN, so PacBio users cannot follow them. Overlaps RM66. · from METHYLATION_PACBIO · 0.9 (moved 2026-11-03)
  • ⏳ RM318 — open, a minor, taken into 0.8 on 2026-09-30, design first, severity medium. Imprinting-region methylation (SNURF, KCNQ1OT1) is real in PacBio BED/TSV outputs; needs a region table and a non-VCF pointer; no clinical bands read yet. · from METHYLATION_PACBIO · 0.9 (moved 2026-11-03)
  • ⏳ RM317 — open, a minor, taken into 0.8 on 2026-09-30, design first, severity medium. TRGT's FORMAT/AM (per-allele repeat-tract 5mC) is in every PacBio VCF; binning it needs a methylation kind and a cross-field allele rule (the expanded allele can be the unmethylated one). · from METHYLATION_PACBIO · 0.9 (moved 2026-11-03)
  • ⏳ RM312 — open, a patch, owner enricher tests, severity low. RM233's wrap assertion holds or fails on tmp_path's length (pytest-9 fits the phrase on one line); found by the 0.8 post-merge suite. · related RM233
  • ⏳ RM311 — open, a patch if both arms reuse existing codes, a decision first, owner schema + compiler, severity low. A gene_validity.csv overlay still matches a superseded curation, so the author hears nothing when the source adopts or overtakes their correction; compare against the group's CURRENT row. · related RM290
  • ⏳ RM309 — open, a patch, parked on a precision measurement, owner compiler, severity low. RM279's second rule: state risk/protective under a conclusion that negates it (rs1799889 G/G). Its gate is the one RM279 met. · related RM279, RM308
  • ⏳ RM306 — open, a minor, parked (the pm, 2026-11-05), owner enricher, severity medium. No exon structure anywhere in the tier; the MANE Ensembl GTF (8.6 MB, 204,815 exons) as a second table in the MANE lane, GENCODE v46 as the option for non-coding coverage; cool-seq-tool + UTA refused on weight. An exon answers where, never the gene attribution. From the GKS survey, 2026-09-28. · also in probes/GKS_SURVEY.md · 0.9 (moved 2026-11-03)
  • ⏳ RM302 — open, a minor, parked (the pm, 2026-11-16); tools never write a register, authors may; shape open, two columns by default, or RM28's table if meta-conclusions land. Owner format + compiler, severity medium. conclusion in up to three registers (detailed = today's, lay, concise) for a report's table, details card and explanatory mode; one claim written three ways, never in a key. Options: two optional columns per carrying kind, or one conclusions.csv side table with a closed register vocabulary. Language is a separate axis. Maintainer request, 2026-09-27. · related RM7, RM28 · for longevitymap v2 · 0.9 (moved 2026-11-04)

✅ Shipped

  • ✅ RM398 — SHIPPED 2026-11-07 on main, cut 2026-11-07 as v0.8.1 — a patch, owner enricher, severity medium. The third placement witness reads dbSNP's GRCh38 SPDI from the merge check's esummary db=snp record instead of asking live REST; same subjects, self-consistent with the snapshot, the merge check served the same records. From S132. related RM267, RM292, RM406, RM401
  • ✅ RM397 — SHIPPED 2026-11-05 on main, cut 2026-11-07 as v0.8.1 — a patch, owner enricher, packaging, severity low. The enricher wheel shipped the fetched Atlas .proto sources (_atlas_protos/) beside the generated tree, five files twice; the hook adds them to the sdist only, the wheel target excludes them, and scripts/build-dists.sh refuses a wheel holding the same bytes twice.
  • ✅ RM395 — SHIPPED 2026-11-05 on 0.8, uncut — a minor (the 0.8 docs pass), owner compiler + enricher, docs, severity low. Three guide commands did not do what the guides state: keygen --out into a missing directory now creates it, draft-repeats without --catalogue reads the provisioned STRchive lane as check-repeat-bands does, and cache pull/cache status describe the disk (what was pulled, what had nothing published, no empty directory left behind, both routes named for an absent lane). · related RM202, RM176
  • ✅ RM394 — SHIPPED 2026-11-05 on 0.8, uncut — a minor (the 0.8 audit, round 3), owner enricher, severity medium. The --rederive drift label called an allele gain "moved" (94 subjects), the derived currency line sent an operator with an old snapshot to --rederive, and the carry warning said "no source could be asked" about asked subjects; enrich/rederive_words.py maps each case to its sentence (seven drift arms, four currency cases, six carry reasons), each walked by a test; a currency line per source. Residuals none. · related RM375, RM373, RM382, RM393
  • ✅ RM393 — SHIPPED 2026-11-05 on 0.8, uncut — a minor (the 0.8 audit, round 3), owner enricher, severity high. enrich --rederive with no Ensembl and no ClinVar snapshot wrote a positional row's back-filled rsID empty with source=authored, and the recompiled parquets carried rsid=null; the positional branch now leaves a subject whose recorded rsID came from a map this run did not read (per link) to the carry-forward. Residuals one won't-fix (a read snapshot missing the chromosome). · related RM98, RM382, RM394
  • ✅ RM392 — SHIPPED 2026-11-05 on 0.8, uncut — a minor (the 0.8 audit, round 2), owner compiler + format + enricher, docs, severity low. A count spelled beside a registry or a table is dropped or compared: a total in the comment above any registry in the three packages must equal its size, and SCHEMAS.md's authored roster, overlay key table and exclusions are compared to their registries as sets. · related RM218, RM223, RM359, RM364, RM380
  • ✅ RM391 — SHIPPED 2026-11-05 on 0.8, uncut — a minor (the 0.8 audit, round 2), owner enricher (tests), severity medium. One fence at the transport in the root conftest.py (getaddrinfo, connect, gRPC's channels) refuses every non-local host unless JUST_DNA_NETWORK_TESTS=1, names the test and host, and fails a test at teardown that swallowed the raise; the 31 tests in 7 files it caught ask nothing remote now, and an import walk holds the enricher's transports to the set the fence test drives. · related RM372
  • ✅ RM389 — SHIPPED 2026-11-05 on 0.8, uncut — a minor (the 0.8 audit, round 2), owner compiler, severity medium. withhold_reproduced_loci compares a locus on CELL_FACT_FIELDS, derived from _CELL_REBUILT, so a locus with a vrs_spec and no id, or an id on a row with no ref, is kept and survives compile → reverse → compile; a walk over every carried column and a round trip over every reference example guard it. · related RM364, RM380, RM270, RM304
  • ✅ RM387 — SHIPPED 2026-11-05 on 0.8, uncut — a minor (the 0.8 audit, round 2), owner format + enricher + compiler, severity medium. resolution.csv source_loci records every place the links offered an rsID at, beside the hostable loci its own key holds; RM374's rsid_placements_disagree compares a positional row against the union, so a sibling locus of a one-to-many rsID no longer refuses a valid module, and --rederive clears a table recorded before the column. Carried on resolution_loci through reverse. · related RM374, RM382, RM364
  • ✅ RM386 — SHIPPED 2026-11-05 on 0.8, uncut — a minor (the 0.8 audit, round 2), owner enricher, severity medium. Every hub write goes through one helper pinned on the revision its plan read (parent_commit), so a concurrent writer refuses a module publish, a snapshot publish or a prune (RemoteMovedError) rather than being overwritten; a snapshot is hashed once and its payload checked on the hub before release.json, sent as the bytes read, describes it; a prune deletes exactly its listed paths at the listed revision. · related RM368, RM199, RM186, RM353
  • ✅ RM384 — SHIPPED 2026-11-05 on 0.8, uncut — a minor (S130, the tail of RM372), owner enricher, severity low. Every pulled lane's repository is spelled once, in the public locations.SNAPSHOT_REPO_IDS (lane → org/repo, the Ensembl lane's just-dna-seq/ensembl_variations_split included); every SNAPSHOT_REPO_PREFIXES row, upload.DEFAULT_*_REPO_ID alias and CacheLane.publish_repo is derived from it, and an AST walk refuses a repo id written anywhere else. · for S130 · related RM372, RM278, RM202
  • ✅ RM383 — SHIPPED 2026-11-05 on 0.8, uncut — a minor (found regenerating the reference examples), owner enricher, severity low. record_table_sources takes answered_every_row, stated by every caller; enrich --rederive and the concordance record retire their own licence row no committed row names, and a merging pass never does.
  • ✅ RM382 — SHIPPED 2026-11-05 on 0.8, uncut — a minor (found regenerating the reference examples), owner enricher, severity medium. A link's answer every rsID-only subject rejects is set aside so the next link is asked, and an rsID every link rejected gets no resolution.csv row instead of not_found; pathogenic_clinvar's rs613985 resolves from ClinVar.
  • ✅ RM381 — SHIPPED 2026-11-05 on 0.8, uncut — a minor (found regenerating the reference examples), owner enricher, severity high. Every reader of a snapshot's alt (both rsID lookups, REST's allele_string, gnomAD's ids, the three position→rsID readers) spells each ALT alone through alleles.trim_allele and regroups by it (per_alt_loci), offline too; the split repo's padded GTT>G,GT,GTTT is G>GT, GT>G, GTT>G. A genotype in a site's padded frame is read in it with the same trim (resolution.site_verdicts, both tiers). An AST walk holds the readers to a registry. Residuals two won't-fix. · related RM270, RM372, RM376, RM267
  • ✅ RM380 — SHIPPED 2026-11-05 on 0.8, uncut — a minor (found building RM375), owner compiler, severity medium. The parquets' resolution-locus struct is derived from ResolutionRow minus a reasoned exclusion map, so authority, the placement witness and verdict, left_aligned, source_release and caid survive compile → reverse → compile; reverse's header is derived too, its two other hand-kept headers and the enricher's two sidecar headers are held to their models, and _verify's strict refusal no longer says the table's position is lost. · related RM364, RM373, RM375, RM267, RM270
  • ✅ RM379 — SHIPPED 2026-11-05 on 0.8, uncut — a minor (the 0.8 Ensembl migration probe, RM373's review), owner enricher + format + compiler, severity low. Each consulted authority a clin_sig_authority_calls.csv row names gets its licence row at the new clin_sig_authority_call layer, inside the table's commit and labelled by record_table_sources; CIViC contributes no row and writes none. · related RM373
  • ✅ RM377 — SHIPPED 2026-11-04 on main, uncut — a patch, owner compiler, severity low. The two refusals strict adds have phrases published in COMPILER.md and pinned by a test that reads them from that table and walks the source for every strict compile: literal. · for S129 · related RM378, RM131, RM218
  • ✖ RM376 — CLOSED 2026-11-04 on 0.8 — not a defect, no code change, owner enricher, severity low. rs746071566 read three ALTs offline and one online on one split snapshot. The snapshot, Ensembl REST and dbSNP 157 all carry the same three alleles at one site; the difference is RM270's fill, which left-aligns and writes the piece the genotype names only when reference bases are readable, and writes the locus whole with left_aligned unknown when they are not.
  • ✅ RM375 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (the 0.8 Ensembl migration probe). Owner enricher + compiler + format, severity medium. One three-valued judge of two spellings (events.py); the enricher records resolution.csv authored_event (bound to judged_coordinate) against the reference and the compiler reads same_event as agreement (no positional_identity_contradicted, no rsid_coordinate_disagrees), never judging itself; coordinate_verdict stops reading an insertion's anchor as a substitution; --rederive's drift is respelled / moved / undecided. Residuals two won't-fix. · related RM267, RM270, RM364, RM372, RM376
  • ✅ RM374 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (the 0.8 Ensembl migration probe), owner compiler, severity medium. One rsID placed two ways by two subjects of one resolution.csv draws rsid_placements_disagree in both modes and is refused by compile --strict and validate --strict; a resolved allele refusal names the stale-release reading. · related RM372, RM373, RM375, RM275
  • ✅ RM373 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (the 0.8 Ensembl migration probe), owner format + enricher, severity medium. resolution.csv source_release (outside resolution_signature), stamped by the run that answered a row and never back-filled; every derived pass's licence row labelled by its committed table through licensing.record_table_sources; rsid_coordinate_agreement.release withheld with a count over a mixed table. Guarded by an equality walk of each pass's lane reads against DerivedProducer.release_lanes. · related RM372, RM374
  • ✅ RM372 — SHIPPED 2026-11-04 on 0.8, uncut — a minor. Owner enricher + format, severity medium. The Ensembl snapshot moves to just-dna-seq/ensembl_variations_split (dbSNP 157, one row per ALT); a cache whose pull record names another repo is re-pulled by cache pull, the release's manifest.json attests every kept and downloaded file by sha256, a cache holding a file it does not attest is withheld whole, and the lane's label is the manifest's release_id. An autouse fence keeps tests off the real hub. Residuals three won't-fix. · related RM303, RM267, RM270, RM304
  • ✅ RM371 — SHIPPED 2026-11-04 on main, uncut — a patch, owner enricher, severity medium. Every wait_exponential_jitter passes multiplier= where it passed tenacity's deprecated initial=; the floor is tenacity>=9.2.1, and an AST walk refuses the old spelling. · for S128
  • ✅ RM370 — SHIPPED 2026-11-04 on 0.8, uncut — a minor, owner enricher, severity low. Every enricher sidecar writer calls layout.nothing_to_write first: zero rows and no file there already writes no table and so no licence row (seven header-only tables, the concordance pair in three reference examples removed); a table already on disk is still the pass's to empty, and resolution.csv and an overlay-answered concordance record are written empty on purpose. A test walks DERIVED_PRODUCERS with one nothing-to-answer run each, and an AST walk asserts every CSV writer calls the predicate. · related RM142, RM231
  • ✅ RM369 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (the 0.8 audit, round 1), owner format + enricher, severity low. begin_snapshot_write withdraws release.json before a builder's payload moves and its commit is the only route to the writer (all 16 builders); begin_block_write for ClinVar's citations block; parts_read/parts_total/source_release/vrs_version typed on SnapshotRelease. Guarded by an equality AST walk, a reader walk over all three tiers, and every builder run with its describe step failing.
  • ✅ RM368 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (the 0.8 audit, round 1), owner enricher, severity medium. The plan records every file's sha256 and the send commits those verified bytes; a malformed declared field refuses; the manifest itself is compared at the versioned path; every dry run rehearses the real run's checks; one PublishRefused family.
  • ✅ RM367 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (the 0.8 audit, round 1), owner compiler, severity medium. The audit's class F: validate and compile asked different checks, and the compile refused after writing. Both commands now run one resolution step (compiler/resolution_step.py), check_paths rosters the checks one path asks (an equality against an AST walk), every compile refusal sits above output_dir.mkdir(), --no-resolve reads nothing it says it skipped, and validate takes --resolve/--no-resolve and --ba1-threshold.
  • ✅ RM366 — SHIPPED 2026-11-04 on 0.8, uncut — a minor, owner format + compiler + enricher, severity low. One declaration per model (_PAIRED_FIELDS, base.Companion) and one validator (base.PairedFields) refuse a column stated without its companion: the GWAS posterior and its method, a colocalisation statistic and its method, a placement witness and its verdict; eight hand-written pair validators fold into it; a published pair (effect_size without effect_measure) warns as paired_field_half_stated. related RM330, RM325, RM267, RM239
  • ✅ RM365 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (the 0.8 audit, round 1), owner format + compiler + enricher, severity low-medium. Every registry guard is an equality over a walked set: DERIVED_PRODUCERS checks against what each pass writes (four were missing), the first-seen model registry and since() values against the package and the release tags, COMPILER.md's parquet diagram and input list against their tuples, every *_FACT_FIELDS tuple against its model, and three narrower rosters.
  • ✅ RM364 — SHIPPED 2026-11-04 on 0.8, uncut — a minor, owner compiler, severity medium. reverse copies back provenance.json and the logs (shipped_files), weights.parquet and every positional parquet carry resolution_loci so a row the fill left alone gives back the table's answer, and no resolution.csv comes back for a module that read none; every manifest field is on REVERSE_MANIFEST_RECONSTRUCTED or REVERSE_MANIFEST_EXCLUDED. · also in CHANGELOG, COMPILER § Reverse, SCHEMAS, FAQ, MODULE_LIFECYCLE § 6.9, INTEGRATION_0_8 § 1.28, ROADMAP_1_0 § RM69
  • ✅ RM363 — SHIPPED 2026-11-04 on 0.8, uncut — a minor, owner format + compiler + enricher, severity medium. One comparison per kind in just_dna_format.spelling (gene, drug, DOI, and chrom folded on input on every model) plus spec.citation_key for a PMID, used by every key, scope and selector in all three tiers, with model-walk and AST guards asserting equalities; new code meta_member_drug_disagrees. related RM351, RM356
  • ✅ RM362 — SHIPPED 2026-11-04 on 0.8, uncut — a minor, owner compiler + enricher, severity medium. One answer to "which genes, loci and variants is this module about" (compiler.subjects.module_subjects), called by the five derived-row orphan checks, the study-orphan check, the overlay's resolution.csv reachability, stats.genes and the enricher's gene scope; a PGx-only module's fact tables are checked, and a diplotype module's own gene is no longer an orphan. related RM28, RM157, S57
  • ✅ RM361 — SHIPPED 2026-11-04 on 0.8, uncut — a minor, owner enricher + format + compiler, severity high. A failed, retired, unread, timed-out, unparsable or absent input is never recorded as a definite answer: a 410 is retired (net.could_not_ask_reason, Withholding), each client's "no such thing" is one named status, a body of another shape is could-not-ask, an Open Targets snapshot not read whole answers no key, an undecided regex withholds quotes_found, and an absent cell is an empty key segment (coordinate_key). Every site is in the RM361 entry, each guarded by a walk or a red-first test. related RM354, RM258, RM333
  • ✅ RM359 — SHIPPED 2026-11-04 on 0.8, uncut — a minor, owner compiler + docs, severity low. Under --strict an escalated finding is in errors only, never also in warnings (the pre-flight runs in the compile's mode; resolution ladders route through route), and registry totals stated in prose are dropped (SCHEMAS.md's "21 entries today" and the sweep behind it).
  • ✅ RM357 — SHIPPED 2026-11-04 on 0.8, uncut — a minor, owner compiler, severity medium. tables.AUTHORED_NON_KIND_TABLES registers overrides.csv, meta.csv and model_organisms.csv; one loader (load_authored_table) replaces three, and every list naming them is derived; a refused table keeps its parsed rows, so a meta.csv duplicate no longer reports its citations as uncited. related RM335, RM28
  • ✅ RM356 — SHIPPED 2026-11-04 on 0.8, uncut — a minor, owner format + compiler, severity medium. A haplotypes.csv row's duplicate rule splits on a gene both rows state (HaplotypeRow._KEY_DISTINCT_WHEN_STATED, compiler.tables.same_row at the compile, drafting and lint sites); a gene-less row still collides. related RM351
  • ✅ RM355 — SHIPPED 2026-11-04 on main, uncut — a patch, owner enricher, severity low. One manifest reader and one entry filter for the module publisher; upload --dry-run reports a refusal as the real run does instead of a traceback. related RM347, RM353
  • ✅ RM354 — SHIPPED 2026-11-04 on main, uncut — a patch, owner enricher, severity high. The gwas pass reads the GWAS Catalog REST API v2; every client names a 410 with its own *Retired type, and no client retries a permanent 4xx. related RM239, RM333
  • ✅ RM353 — SHIPPED 2026-11-04 on main, uncut — a patch, owner enricher, severity medium. The module publisher sends exactly what the manifest declares (artifact files, logo, readme, provenance, logs) and its versioned-path guard compares every declared file's hash; every name handed to the hub (both publishers, retirements, prune) is glob-escaped by one helper. related RM347
  • ✅ RM352 — SHIPPED 2026-11-04 on 0.8, uncut — a minor, owner format + compiler, severity high. A toolchain change kind (parquet_bytes only) so the 0.8.0 record can declare the polars 2 byte move; the gate accepts it only over a fixed-compiler sweep (sweep --toolchain-before) with differing recorded toolchains and unmoved content; the Goals sentence names the recorded toolchain. related RM342
  • ✅ RM351 — SHIPPED 2026-11-04 on 0.8, uncut — a minor, owner compiler, severity medium. The cross-table PGx checks key an allele on (gene, allele); an HLA gene's typed alleles are exempt from star_allele_undefined (pgx.is_hla_gene); a gene-less haplotypes.csv row still defines its name for every gene. related RM334
  • ✅ RM349 — SHIPPED 2026-11-04 on 0.8, uncut — a minor, owner enricher, severity medium. A pmid=NONE row's DOI is asked of Crossref (the title it names recorded; a DOI no agency knows a finding) and its quote matched against the text Europe PMC holds for it, preprints included. related RM336, RM263
  • ✅ RM348 — SHIPPED 2026-11-04 on 0.8, uncut — a minor, owner compiler, severity medium. Every path log discovery considers, the top-level *.log and logs/ itself included, must resolve to where it lies, so a symlink at any level stays behind; one ceiling, LOG_MAX_BYTES, for every file; an unreadable file, or a directory the walk cannot list, draws a new unreadable reason instead of failing the compile. related RM337, RM338
  • ✅ RM347 — SHIPPED 2026-11-04 on main, uncut — a patch, owner enricher, severity low. The module publisher uploads every path manifest.logs names, read from the manifest, never a glob; an unnamed file under logs/ stays behind, and a path leaving the module directory is skipped with a warning. The plan is the upload's allowlist. related RM338
  • ✅ RM343 — SHIPPED 2026-11-04 on main, uncut — a patch, owner enricher (tests), severity low. The query-shape test reads the join operator in either spelling, so duckdb 2.0's Hash Join passes; no cap.
  • ✅ RM342 — SHIPPED 2026-11-04 on 0.8, uncut — a minor, owner format + compiler + enricher, severity medium. compilation.toolchain records the Python and polars that wrote the bytes; a publish collision or a sweep movement the two recorded toolchains explain is toolchain_differs (ToolchainDiffersError, ModuleDelta.bytes_reason); P4 amended. related RM304, RM341
  • ✖ RM341 — CLOSED 2026-11-04 — reverted before push: no cap, the maintainer's decision, owner compiler (packaging), severity high. A polars<2 cap and footer test, removed the same night; the rule: the liaison maps adoption, a pin only for a proven break. RM342 carries the toolchain record.
  • ✅ RM340 — SHIPPED 2026-11-04 on main, uncut — a patch, owner compiler, severity low. The no-ALT line of vrs_coverage_incomplete names verification.json's rsid_coordinate_agreement record when it says the snapshot left authored rsIDs unplaced; message text only. · for S127
  • ✅ RM339 — SHIPPED 2026-11-04 on main, uncut — a patch, owner enricher, severity high. A pull writes pull-record.json from the remote listing before its first download, so an interrupted pull resolves as absent and the next run fetches only what it lacks; an unrecorded cache is compared with the listing once. A timeout names HF_HUB_DOWNLOAD_TIMEOUT. · for S126
  • ✅ RM338 — SHIPPED 2026-11-04 on main, uncut — a patch, owner compiler, severity low. Under logs/, .txt/.md/.json/.jsonl ship when they decode as UTF-8 and parse, png/jpg/jpeg when their magic matches and they are at most 5 MiB; SVG never. *.log ships unchecked as before, and each newly shipped file is named by an info line. No identity moves; the left-behind warning is RM337, a minor. · for S125
  • ✅ RM337 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (a new warning code, RM308's rule), owner compiler, severity medium. log_file_not_shipped names each path under logs/ that discovery leaves behind, in both modes, with one of nine reasons decided by the same function as the verdict (_log_left_behind_reason). · for S125
  • ✅ RM336 — SHIPPED 2026-11-04 on 0.8, uncut — a minor, owner format, severity medium. A DOI-only pmid is refused and named; pmid=NONE makes doi mandatory, one rule (spec.CitationPair) on every model with a pmid, and doi? added to the binning kinds, pharm_variants.csv and gwas_effects.csv. · for longevitymap v2
  • ✅ RM335 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (PROPOSAL_0_8_PT2 §5 A6). model_organisms.csv → model_organisms.parquet: one row per model-organism lifespan finding about a human gene, keyed on all but the finding's two columns, concordance closed, citation optional and unwarned. Annotation only and not a lead table. · for longevitymap v2
  • ✅ RM334 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (PROPOSAL_0_8_PT2 §5 A4). No HLA table: an HLA allele is an allele_function.csv row, and where gene starts with HLA- the name must follow IPD-IMGT/HLA's grammar (two to four fields, optional expression suffix) or it is refused. The published-module scan found no HLA row. · for longevitymap v2
  • ✅ RM333 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (PROPOSAL_0_8_PT2 §5 D2). The opentargets lane (release 26.09, CC0-marked, notice carries the terms' third-party clauses) and opentargets enrich: locus_to_gene.csv per (study locus, gene, module variant) for every set a module variant is a member of (members, not lead-only, 2026-11-04), colocalisations.csv per (left, right study locus, module variant); several study loci per variant kept; no C/E label computed. Residuals RM238, RM345. · for longevitymap v2 · related RM236, RM239
  • ✅ RM332 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (PROPOSAL_0_8_PT2 §5 A5). effect_unit on StudyRow (free text, verbatim), and a derived effect_unit on weights.parquet when every stating study row of the variant agrees; the VariantRow half was cut. · for longevitymap v2
  • ✅ RM331 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (StudyRow.proxy_rsid and the allele check against the proxy; the computed pair: ld-pairs asks Ensembl's pairwise LD in the study's one superpopulation and writes ld_pairs.csv, r² withheld with its reason for a mixed or unstated ancestry), owner format + compiler + enricher, severity medium. · for longevitymap v2
  • ✅ RM330 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (PROPOSAL_0_8_PT2 §5 A5). meta.csv's sex condition and StudyRow's stratum/stratum_level (free text, both or neither), through studies.parquet and reverse; the duplicate-citation check reads the stratum pair. · for longevitymap v2 · built on 0.8 2026-11-04: sex on meta.csv; the study stratum still open
  • ✅ RM329 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (PROPOSAL_0_8_PT2 §5 A5). age_min/age_max (years, open either side) and the closed age_of on StudyRow, through studies.parquet and reverse; the duplicate-citation check reads the band cell by cell. · for longevitymap v2
  • ✅ RM328 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (PROPOSAL_0_8_PT2 §5 A5). ancestry on StudyRow over VALID_SUPERPOPULATIONS (VALID_TRAINING_ANCESTRY derived from it), and the derived eur_only_evidence on weights.parquet. · for longevitymap v2
  • ✅ RM327 — SHIPPED 2026-11-04 on 0.8, uncut, a minor, owner format + compiler, severity medium. Per-trait judgements are one-member meta.csv rows (trait_efo_id, direction, trait_role); tradeoff is computed per member on meta_resolved.parquet, true or null, never false (no closed trait list); the name moved from the reserved set to MISPLACED_COLUMN_REASONS. · for longevitymap v2
  • ✅ RM326 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (PROPOSAL_0_8_PT2 §5 A5). phenotype_class on StudyRow (open, v2's twelve recommended, phenotype_class_unrecognized outside them), the duplicate-citation narrowing by class, and the derived outcome strip on weights.parquet. · for longevitymap v2
  • ✅ RM325 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (PROPOSAL_0_8_PT2 §5 A5). sample_size (the total analysed, cases included), the closed sample_unit (individuals, lives) and n_cases on StudyRow, through studies.parquet and reverse; a size without its unit, or more cases than analysed, is refused. · for longevitymap v2
  • ✅ RM324 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (PROPOSAL_0_8_PT2). One open tier on meta.csv, graded on module_spec.yaml's tier_scale (refused without it), copied into the manifest and written back by reverse. · for longevitymap v2 · related RM323, RM294
  • ✅ RM323 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (PROPOSAL_0_8_PT2). Per-locus facts live on meta.csv only ("go with C"): tier, age_of_action and its confidence, flags, resolved per row into meta_resolved.parquet; causal and effector confidence are not authored. · for longevitymap v2 · related RM28, RM324, RM333, RM346
  • ✅ RM322 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (PROPOSAL_0_8_PT2 §5 A3). DiplotypeRow gains negatives and category, VariantRow's free-text columns, through diplotypes.parquet and reverse; priority, actionability and flags were cut by the maintainer. · for longevitymap v2
  • ✅ RM320 — SHIPPED 2026-10-02 on main, uncut — a patch, owner enricher, severity medium. WGLab answered WGLab/PubMind#5: the ANNOVAR/Zenodo PubMind table is CC BY-NC-ND 4.0. PUBMIND_TERMS records it (commercial_use=False, NoDerivatives in notice), so a module carrying a pubmind row compiles only under non_commercial; pubmind build and cache rebuild gate on --use; pubmind publish still refuses, now because the snapshot is adapted material. Rows drafted earlier keep their null axes until deleted and re-drafted. · from RM134's one open question · related RM289, RM27, S99
  • ✅ RM316 — SHIPPED 2026-09-30 on main, uncut — a patch, owner format + docs, severity low. annotated_alt's meaning and the consumer contract resolve a VCF 4.5 local-allele field through LAA before any element rule; local_annotated_alt refused (P2). Text only. · related RM54, RM315
  • ✅ RM315 — SHIPPED 2026-09-30 on main, uncut — a patch, owner format + docs, severity medium. VCF_FIELD_NUMBER transcribes all of VCF 4.5's reserved keys (§3's RUC among them, the local-allele and base-modification keys), CICN is a named collision, VCF_NUMBER_MEANINGS covers LA/LR/LG/M by an equality test, and the docs state the version. · related RM53, RM54, RM56
  • ✅ RM314 — SHIPPED 2026-09-30 on main, uncut — a patch, owner format + docs, severity medium. A reference block's extent is the sample's FORMAT LEN first, INFO END only as fallback (4.5 made END a cross-sample maximum); a . GT inside the sample's own block is that block's call. Text only. · related RM57, RM313
  • ✅ RM313 — SHIPPED 2026-09-30 on 0.8, uncut — a minor (PROPOSAL_0_8_PT1). RM57's backwards quality_floor_inverted is retired (kept in the vocabulary, declared in RETIRED_WARNING_CODES) and replaced by qual_floor_on_reference_evidence, which recommends no field. · related RM57, RM314
  • ✅ RM310 — SHIPPED 2026-09-29 on main, uncut — a patch, owner enricher, severity medium. _rebuild_cpic catches cpic.CpicError, so a CPIC outage is a failed lane instead of a traceback that skips every later lane; a test walks CACHE_LANES and asserts every adapter contains its builder's documented errors. · related RM288, RM97, RM101
  • ✅ RM308 — SHIPPED 2026-09-30 on 0.8, uncut — a minor (PROPOSAL_0_8_PT1). RM279's conclusion-genotype rule now runs at validate/compile as conclusion_names_other_genotype, a warning in both modes, aggregated by reason (swapped pair / copied text). · related RM279, RM309
  • ✅ RM307 — SHIPPED 2026-09-29 on main, uncut — a patch, owner enricher, severity low. Measured: no Atlas throttle up to 21 calls/s (concurrency degrades at 16 threads first), so ATLAS_REQUEST_INTERVAL = 0.0; the client still waits on a PacingGate per attempt so spent counts it, and takes an injected gate=. · related RM280, RM192
  • ✅ RM305 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (PROPOSAL_0_8_PT2 §5 C3). The clinvar_gkm cache lane: ClinVar-GKM's GRCh38 net insertions and deletions with VRS id, VCF spelling, ClinVar variation and rsID, downloaded whole and verified against the set's manifest (pin (source_release, sha256)); ClinVarGkmWitness.lookup for the checks that name it, a three-arm currency (clinvar-gkm currency) and a --verify-datasets probe. No check of its own. · related RM267, RM270, RM292 · also in probes/GKS_SURVEY.md, probes/LANES_0_8.md
  • ✅ RM304 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (PROPOSAL_0_8_PT2 C1). The per-row version was already ResolutionRow.vrs_spec (0.5), now read; the pin manifest.compilation.vrs_version; the minter reads ga4gh.vrs.VRS_VERSION at its minor; a row under another version is vrs_version_differs, never a mismatch; rs72613567 pinned behind JUST_DNA_NETWORK_TESTS=1. ga4gh.vrs stays capped <2.4. · related RM270 · also in probes/GKS_SURVEY.md · 0.8
  • ✅ RM303 — SHIPPED 2026-11-04 on 0.8, uncut — a minor (PROPOSAL_0_8_PT2 §5 D1). The snapshot layout names and SnapshotRelease, the model of release.json, live in just_dna_format.snapshot; every builder writes through write_release (an AST walk refuses a hand-built one), old spellings (rows, licence, ClinVar's dateless label) are read and never written, and ClinVar's per-lane label reader is gone. Goals amendment. · related RM261
  • ✅ RM301 — SHIPPED 2026-09-27 on main, uncut — a patch, owner enricher, severity medium. Credential readers use locations.env_value, which reads one key (environment, then .env) and exports nothing; load_env() stays in the cache resolvers. Reopens RM102's credential half. · from S124 · also in CHANGELOG, ENRICHER, AGENT_NOTES, CLAUDE.md, CONSUMER_SUGGESTIONS_HISTORY
  • ✅ RM300 — SHIPPED 2026-09-30 on 0.8, uncut — a minor (PROPOSAL_0_8_PT1). manifest.stats gains clin_sig_counts and clin_sig_variant_counts: rows and distinct variants per VALID_CLIN_SIG tier, zeros included, from the effective clin_sig. · from S123 via RM299