Skip to content

RM index, items 400–499

One page of the RM index, which is paged by number: every item numbered 400–499. 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

  • ⏳ RM455 — built on main for 0.8.10 (e63a8d8c, 5799c2c9, 78b8fb3f; 0.9's round-1 fix builder, 2026-11-18), owner enricher, severity low. The older instances of four classes 0.9 fixed in its own code: UnicodeDecodeError escaping three pull-record readers, batch_size checked only at construction in three settings, mitomap_miss reported present over moved parents, a lane keyed on its name. · related RM430, RM431, RM437
  • ⏳ RM454 — open, a patch, after 0.9 merges (the RM430 follow-up, 2026-11-18), owner enricher, severity low. The constraint snapshot compares a gene exactly, both lookup_loci compare an rsID and a contig exactly (RS334, chr11 miss), and SQL's lower(trim()) differs from gene_key on tabs and non-ASCII. No module sends those spellings; a direct lookup can. Every keyed SQL compare goes into KEY_COMPARES, with one walk. · related RM363, RM430, RM431
  • ⏳ RM453 — open, a patch (a declared corrected derivation), built on main 2026-11-18, uncut, for 0.8.10 (the lead's 0.8.9 run, 2026-11-18), owner enricher + compiler, severity medium. Offline, ClinVar files rs80356821's deletion under the retired rs281864900 (merged by dbSNP), so the witness holds nothing under the rsID, both spellings read unplaced, and RM449 fills neither (strict refuses). unplaced conflates no-counterpart with no-record; or the witness could follow merges. Built: RM449's neither arm is gone; no same_event spelling at a mirrored site keeps both, said under the borrowed code. · related RM267, RM398, RM449
  • ⏳ RM450 — a patch, parked (unparks on an incident or a host running two pulls into one cache) (the pm's check from RM437, 2026-11-17), owner enricher, severity low. download.py takes no lock; two pulls into one cache share fixed .part names and the last pull record wins. If kept: a directory flock for the whole pull (@flock-not-a-lockfile), which RM437's groups inherit. · related RM437, RM445, RM231
  • ⏳ RM443 — open, a minor, 0.9's scope (the pm, 2026-11-16), owner enricher, severity low. lane_status calls each lane's resolve() with no argument, so a host that configures lane paths in its own settings gets a report on directories its passes never read; candidate locations= mapping, with load_dotenv_file decided beside it (S154).
  • ⏳ RM441 — open, a minor, 0.10's scope (the pm, 2026-11-16) (one design with RM399), owner format + compiler, severity low. Every release-record axis is compiled output, so RM438's resolution.csv correction left 0.8.5's record a measured zero and the registry's sweep read nothing moved. Asks: a derived-table axis routed to re-derive (not joining RECOMPILE_DRIVING_AXES by default), a requires for its reach, RM438 declared in the next record, and (S156) the newest enricher release whose derivation of a table could differ, readable offline. From S153 and S156. related RM438, RM201, RM120
  • ⏳ RM437 — open, a minor, 0.9's scope (the pm, 2026-11-16) (its doc half cut 2026-11-13 as v0.8.6), owner enricher, severity medium. The ensembl lane pulls data/ only, and every annotation value moved to groups/, joinable only within one release; ask: opt-in group selection attested by the manifest, a public group-directory name, a refusal on a release mismatch. From S150. related RM420, RM339, RM381
  • ⏳ RM435 — open, a minor, 0.10's scope (the pm, 2026-11-16), owner format + compiler, severity low. StudyRow gets effect_direction (increase|decrease, relative to effect_allele), the field and vocabulary GwasEffectRow already carries, with a warning when a stated effect_size disagrees. From S148. related RM91, RM148
  • ⏳ RM433 — open, a minor, 0.9 (the methylation zone), owner enricher, severity low. AlphaGenome v2.0.0's DnaModelServiceV2 computes predictions live and adds a predicted-methylation scorer, tied to RM317 to RM319 by the maintainer; a terms probe, then a trial, then a design. Nothing breaks without it. From the maintainer's ask, 2026-11-12. related RM191, RM345
  • ⏳ RM431 — open, a minor, 0.9's scope (the pm, 2026-11-16), owner enricher, severity low. An rsID-sorted index beside the Ensembl snapshot, a key with pushdown on the whole-table readers, a keyed Open Targets function, and AlphaGenome scores read without a local glob. From S145 asks 4 to 7. related RM430, RM405
  • ⏳ RM430 — open, a minor, 0.9's scope (the pm, 2026-11-16), owner enricher (+ format, a parser at most), severity medium. A remote root (hf://…) collapses in Path, and every resolve, release.json read and connector checks the local disk, so no lane is read remotely though its keyed SQL prunes fine. The format tier fetches nothing (the Constitution's Goals), so a remote release read lives in the enricher. From S145 asks 1 to 3. related RM431, RM432, RM339
  • ⏳ RM429 — open, a minor, parked, owner enricher, severity low. GwasResult gets size_refusals, the count RM428 already keeps on the client and only logs, so an in-process caller sees the Catalog's page ceiling move. Today a caller can inject its own GwasCatalogClient and read size_refusals after the run. From S144. related RM428, RM354, RM100
  • ⏳ RM424 — open, a minor, parked (the pm, 2026-11-09; unparks when a reader asks which checks are current without re-running them, or wants the history RM423 drops), owner format + enricher + compiler, severity low. RM423's residual, S139's candidate 2: a per-record module_hash on VerificationRecord, so a record carried across an edit is read as stale on its own instead of dropped. related RM423, RM45, RM72, RM284
  • ⏳ RM419 — parked (the pm, 2026-11-08; a patch when it unparks on a reported rate-limit refusal or the registry running enrichments as threads), owner enricher, severity low. RM418's class beyond NCBI: the resolver and the LD pass each pace Ensembl REST's per-address budget with a gate of their own, and a dozen other clients build theirs by default. related RM418, RM396, RM404
  • ⏳ RM405 — parked (the pm, 2026-11-07). open, a minor or a patch by design, owner enricher, severity low. The registry measured 9 of an offline enrichment's 10 s in four duckdb queries over the snapshot (two by rsID, two by position), and the snapshot has no rsID index. From S133. related RM372, RM303
  • ⏳ RM406 — parked (the pm, 2026-11-07). open, a minor or a patch by design, owner enricher, severity low. dbSNP's placement, already fetched for every rsID by the merge check, widened past RM398's subjects: report first, withhold if the count says so; no new requests. Split from RM398 on the maintainer's ask. related RM398, RM267, RM292, RM303
  • ⏳ RM412 — open, a minor, 0.9's scope (the pm, 2026-11-16): half 2 and the flag; the per-lane release probe parked, owner enricher. An opt-in --rehearse-download for cache rebuild --dry-run and a per-lane source probe, plus the 0.8 line's private names made public (_ensure_opentargets_snapshot, publish commands for opentargets and clinvar_gkm, RM401's and RM403's). related RM411, RM408, RM404
  • ⏳ RM404 — open, a minor, 0.9's scope (the pm, 2026-11-16), owner enricher, severity low. enrich() builds its own eutils client and SequenceProxy, so a host cannot pace them with its shared bundle; ga4gh's seqrepo proxy has no timeout (upstream). From S133; S157 (a local seqrepo for an online run) folded in. related RM396, RM385
  • ⏳ RM402 — open, a patch, after RM401 and a re-measure (the pm, 2026-11-07), owner enricher, severity medium. An rsID the Ensembl snapshot answered and RM382 rejected is asked again of live Ensembl, the same authority; skip it when REST's release equals the snapshot's. From S133. related RM382, RM373, RM292

✅ Shipped

  • ✅ RM452 — SHIPPED 2026-11-18 on main, cut 2026-11-18 as v0.8.9 — a patch (the RM431 builder, 2026-11-18), owner enricher, severity low. A NULL start raised TypeError at the rsID scan's and the back-fill's int(start); such a row now places nothing (named, next link asked) and such a probe gets no candidates. No real snapshot carries one. · related RM271, RM381
  • ✅ RM451 — SHIPPED 2026-11-18 on main, cut 2026-11-18 as v0.8.9 — a patch (the lead's 0.8.9 run, 2026-11-18), owner compiler, severity low. A coordinate row the table never filed has no resolution_loci, so reverse rebuilds a table row from its cells and lap 2 reads a row lap 1 did not: resolution_signature moves. Not RM449's: a two-row module with no mirror shows it, and one enrich pass removes it. Built: a module that read a table writes [] in resolution_loci for a key with no positioned locus, and reverse writes nothing for it; one example digest moves (cyp2c9_warfarin_grch37). · related RM364, RM380, RM449
  • ✅ RM449 — SHIPPED 2026-11-18 on main, cut 2026-11-18 as v0.8.9 — a patch, a declared corrected derivation (S158, the registry, 2026-11-18), owner compiler + enricher, severity high. An rsID row's C/CT is hosted at both C>CT and CT>C at one position, so the fill puts the insertion's claim on the deletion too: silent alone, a strict Duplicate (variant, genotype) beside an authored coordinate row (21 in cardio). Reopens RM31's "a two-allele genotype carries its frame". Built: the witnessed spelling fills, both-ways keeps both with a warning (borrowing an existing code in 0.8.9; rsid_genotype_reads_as_two_events in 0.9), neither leaves the row unpositioned; the ClinVar drafter writes indel records by coordinate. The sweep moves two pathogenic_clinvar rows. residual RM453 · related RM31, RM267, RM274, RM381
  • ✅ RM448 — SHIPPED 2026-11-18 on main, cut 2026-11-18 as v0.8.9 — a patch (S158, the registry, 2026-11-18), owner enricher, severity high. Online with verify_ref, counterparts pairs Ensembl's A>AT with ClinVar's A>AC at a multi-allelic site, so compare_placements withholds the whole locus and the rsID gets no row. RM267 decided "a different allele is unplaced". Built: a witness allele already the same event as one of the rsID's alleles is no counterpart of the others; S158's four sites as fixtures. · related RM267, RM375, RM381
  • ✅ RM447 — SHIPPED 2026-11-17 on main, cut 2026-11-18 as v0.8.9 — a patch (the lead's RM404 online run, 2026-11-16), owner enricher, severity low. Ensembl REST at release 116 serves dbSNP 156 (rs646776 C/G/T); the same release's VCF dump, the snapshot, carries dbSNP 157 (C>A,G,T, as reference_examples/longevity_studies records). Measured a class: 54 of 454 example rsIDs lack snapshot alleles in REST, never the reverse. A run without the snapshot records the older set, so the --strict refusal on an ensembl-rest row names REST's older build, and the remedy is a re-derivation where the snapshot is provisioned. · related RM374, RM376, RM271
  • ✅ RM446 — SHIPPED 2026-11-17 on main, cut 2026-11-18 as v0.8.9 — a patch (the lead's RM404 build, 2026-11-16), owner enricher, severity low. Withholding.retired is set by a client's first 410 and never cleared, so a client reused across runs (RM404's LookupClients) tells a later run's timeout to upgrade, and that a re-run will not help; RM361 made it sticky on purpose, for one route met twice. Built: every 410 is counted (retired_met), and a run names retired only for one met after the mark it read before asking; the client is never reset. · related RM361, RM404
  • ✅ RM445 — SHIPPED 2026-11-16 on main, cut 2026-11-16 as v0.8.8 — a patch, owner enricher, severity medium. A pull of a cache with no pull record and no manifest.json keeps every whole parquet and then places the hub's release.json, so old bytes state the new release, and every later cache pull reads them as current; this machine's ClinVar cache serves 2026-08 data as clinvar_2026-11-01 (the pm's remote-read trial, finding 8).
  • ✅ RM444 — SHIPPED 2026-11-14 on main, cut 2026-11-14 as v0.8.7 — a patch, owner enricher, severity low. recover_rsid's text: a none beside overlapping records says dbSNP records nothing (F508del sits one base right of VCF's spelling), the note names alleles never given, a position-only recovered advises authoring the anchor base's SNV, and ambiguous names CLI flags to library callers. Text only; a separate outcome declined (S155).
  • ✅ RM442 — SHIPPED 2026-11-14 on main, cut 2026-11-14 as v0.8.7 — a patch, owner enricher, severity low. Three strings in licensing.py say what the saved terms do not: the Atlas row's notice gives the June 2025 training bar where Output from 2026-10-09 bars training any model, its license dates the Output Terms with the Additional Terms' date, and the AVI row cites prohibition 4, which exempts Permissive artifacts. No licence axis moves. From the liaison's re-vendoring, 26db64cb. related RM433, RM195, RM191
  • ✅ RM440 — SHIPPED 2026-11-13 on main, cut 2026-11-13 as v0.8.6 — a patch, owner enricher, severity low. RM438's residual: an enrich --rederive that writes a coordinate-only row without the back-filled rsID it carried, because the row's alleles refute every rsID at its position, logs one warning per run with the count and the first five variant keys; a log line, no warning code, the table unchanged. related RM438, RM393
  • ✅ RM439 — SHIPPED 2026-11-13 on main, cut 2026-11-13 as v0.8.6 — a patch, owner format, severity low. RM434's residual: VariantRow.stat_significance's description says the level is the module author's verdict across the variant's evidence, with no threshold of its own, and points at each study's grade in studies.csv and its p_value_num. related RM434, RM148
  • ✅ RM438 — SHIPPED 2026-11-12 on main, cut 2026-11-12 as v0.8.5 — a patch, owner enricher, severity high. One private rule, resolver._hosting_rsids, narrows a coordinate-only row's back-filled rsIDs by the row's alts, else its genotypes, on both paths: resolve_variants attaches one, withholds several with rsid_ambiguous and says when none can host; enrich keeps pick-and-flag over what is left, a corrected derivation of resolution.csv declared with two exposure queries. From S151 and S152. residual RM440 · related RM382, RM381, RM271
  • ✅ RM436 — SHIPPED 2026-11-12 on main, cut 2026-11-12 as v0.8.5 — a patch, owner enricher, severity medium. The enricher's floor is tenacity>=9.0.0 again, below google-genai's <9.2.0 cap; net._jittered_backoff picks multiplier= or initial= by the installed constructor's signature, and the floor is the release a suite run proved. From S149. related RM371, RM254
  • ✅ RM434 — SHIPPED 2026-11-12 on main, cut 2026-11-12 as v0.8.5 — a patch, owner format, severity low. StudyRow.stat_significance's description says the level is the authors' verdict at their own test's threshold, and that p_value_num is the column to threshold on one scale; a fifth member declined. From S147. residual RM439 · related RM148, RM150
  • ✅ RM432 — SHIPPED 2026-11-11 on main, cut 2026-11-12 as v0.8.5 — a patch, owner enricher, severity medium. gene_metrics.lookup_snapshot and civic_refutation.civic_snapshot_rows raise FileNotFoundError on a reference they cannot read, where they answered {}/[]; the three enricher callers report could-not-read. From S146. related RM361, RM430
  • ✅ RM428 — SHIPPED 2026-11-11 on main, cut 2026-11-11 as v0.8.4 — a patch, owner enricher, severity high. The GWAS pass asks the Catalog v2 for a page of 500 (it sent 1000, which the Catalog has refused with a 400 since some time after 2026-11-04, so the pass filled nothing on any module), asks a refused first page once more with no size, warning once per run, and an opt-in live test asks with the pass's page size. Ask 4, an enricher-only v0.7.4 reading v2, declined by the maintainer (desk #252), so 0.7.3's pass stays broken against the retired v1 API. From S143. related RM354, RM361
  • ✅ RM427 — SHIPPED 2026-11-10 on main, cut 2026-11-11 as v0.8.4 — a patch, owner enricher, severity low. Both clinical-assertion skip lines name cache pull --only clinvar and clinvar build, where they said clinvar pull, which no app registers; a test walks every quoted group command in src/ against both apps' command trees, with a reasoned list of deliberate absences. Ask 3 (a remedy key on findings) declined by the pm. From S142. related RM131, RM111, RM414
  • ✅ RM426 — SHIPPED 2026-11-09 on main, cut 2026-11-11 as v0.8.4 — a patch, owner enricher, severity low. GeneLocusConflict's closing sentence names both readings (an invented variant beside a real symbol, or a real variant beside the wrong gene) and the lookup that separates them; the fact sentence is unchanged. Ask 2 declined by the pm. From S141. related S24
  • ✅ RM425 — SHIPPED 2026-11-09 on main, cut 2026-11-11 as v0.8.4 — a patch, owner format, severity low. ExpressionEffectRow.effect_size's description names the single strongest tissue track, sign kept, which may disagree with effect_direction (the majority sign); it said "aggregated over the tracks". From S140. related RM111, RM200
  • ✅ RM423 — SHIPPED 2026-11-09 on main, cut 2026-11-09 as v0.8.3 — a patch, owner format + enricher, severity high. Over moved authored bytes merge_records keeps this run's records only, so record_verification no longer signs an earlier check's record under a binding it was never computed over (a finding on a removed row, or a clean pass over rows the check never read), and it logs each check it drops by name. Declared in INTEGRATION_0_8 § 4. From S139. residual RM424 · related RM45, RM72, RM73, RM284, RM292
  • ✅ RM422 — SHIPPED 2026-11-09 on main, cut 2026-11-09 as v0.8.3 — a patch, owner compiler, severity low. validate and compile name a resolution.csv row whose variant_key or rsid no authored table names and no study proxies, as one more derived_row_orphan arm, a warning in both modes. From S138. related RM43, RM337
  • ✅ RM421 — SHIPPED 2026-11-08 on main, cut 2026-11-09 as v0.8.3 — a patch, owner enricher, severity low. Every bulk download has a 120 s timeout unless its caller names its own (stream_to_file's default was None, and eleven of fourteen call sites named none), so a server that accepts and goes silent is retried and then reported as the lane's error instead of blocking forever. related RM409, RM403
  • ✅ RM409 — SHIPPED 2026-11-08 on main, cut 2026-11-09 as v0.8.3 — a patch, owner enricher, severity low. An interrupt stops the Open Targets mirror's in-flight part downloads at their next chunk, and the AlphaGenome AVI build's in-flight contigs at their next block, instead of waiting for each to finish; nothing half-written lands. related RM333, RM408
  • ✅ RM420 — SHIPPED 2026-11-08 on main, cut 2026-11-09 as v0.8.3 — a patch, owner enricher, severity medium. cache pull and cache prepare re-pull a cache that came from a pull of its repo when the published release label differs, and leave a lane built here alone with both labels; enrich still never reads the hub for a whole cache. related RM410, RM372, RM339, RM411
  • ✅ RM418 — SHIPPED 2026-11-08 on main, cut 2026-11-09 as v0.8.3 — a patch, owner enricher, severity low. An EutilsClient built without a gate takes the process's NCBI gate (net._service_gate), so two in one process share one budget. related RM404, RM419, RM398
  • ✅ RM417 — SHIPPED 2026-11-07 on main, cut 2026-11-07 as v0.8.2 — a patch, owner enricher, severity medium. Live-API gene-metrics rows are labelled gnomad_api_constraint; the API serves 4.1.1+, so the old gnomad_v2.1.1_constraint label was false, declared for existing rows and kept on a re-run. related RM410, RM373
  • ✅ RM416 — SHIPPED 2026-11-07 on main, cut 2026-11-07 as v0.8.2 — a patch, owner enricher, severity medium. --pin is taken by the seven lanes whose adapter reads it and refused for the rest, before the run; nine adapters had ignored it. related RM410, RM411, RM412
  • ✅ RM410 — SHIPPED 2026-11-07 on main, cut 2026-11-07 as v0.8.2 — a patch, owner enricher, severity low. The constraint lane builds gnomAD 4.1.2 (bgzip, labelled from the file), takes --pin, and a re-run over 4.1 rows adds nothing. related RM229, RM408, RM416, RM417
  • ✅ RM415 — SHIPPED 2026-11-07 on main, cut 2026-11-07 as v0.8.2 — a patch, owner compiler, severity low. The bins_ungrounded clause is chosen by whether any bin cites, so it no longer says "no bin pmid" beside a bin that has one. From S137.
  • ✅ RM414 — SHIPPED 2026-11-07 on main, cut 2026-11-07 as v0.8.2 — a patch, owner format, severity low. RM55's and RM56's findings name the module's declared source_field and speak conditionally when it is not VCF's own field, instead of typing it as RUC/CN. From S136. related RM55, RM56
  • ✅ RM413 — SHIPPED 2026-11-07 on main, cut 2026-11-07 as v0.8.2 — a patch, owner enricher (tests) and the lead (RELEASE_CYCLE), severity medium. Cache-lane prerelease gates: stage parity, idempotency of the commands that promise to change nothing, and a published release in every lane's repository, which blocks a cut when red. related RM408, RM411, RM412
  • ✅ RM411 — SHIPPED 2026-11-07 on main, cut 2026-11-07 as v0.8.2 — a patch, owner enricher, severity medium. cache rebuild --dry-run takes nothing: per lane it prints the plan, this machine's release and the published one, and rehearses an upload only against a snapshot already built. related RM412, RM408, RM176
  • ✅ RM408 — SHIPPED 2026-11-07 on main, cut 2026-11-07 as v0.8.2 — a patch, owner enricher, severity medium. The Open Targets lane gets a repository (just-dna-seq/opentargets) and a private pull route, so cache rebuild --only opentargets --publish publishes it and cache pull pulls it instead of every deployment downloading ~22 GB to build it. related RM333
  • ✅ RM407 — SHIPPED 2026-11-07 on main, cut 2026-11-07 as v0.8.1 — a patch, owner enricher, severity low. An indel's VRS mint reads one 1,024-base window each side of the locus through the run's SequenceProxy and serves every slice normalize takes from it, where each slice was an HTTP read (about ten per indel); the run's minter keeps its ids, so an allele is normalized once. related RM404, RM385, RM267, RM406
  • ✅ RM401 — SHIPPED 2026-11-07 on main, cut 2026-11-07 as v0.8.1 — a patch, owner enricher, severity medium. The live link and the REST witness ask 20 rsIDs per POST /variation/:species; a left-out rsID is asked by GET, a merged one is matched by synonyms, a failed batch is could-not-ask. Its names are private in the patch; 0.9 publishes the knobs with RM404. From S133. related RM402, RM403, RM267, RM361
  • ✅ RM400 — SHIPPED 2026-11-07 on main, cut 2026-11-07 as v0.8.1 — a patch, owner enricher, severity low. The REST rung drops a mapping off the GRCh38 primary assembly, counted, as the tier's other readers do; a re-derivation leaves a recorded off-primary locus out of the comparison and counts it, so it no longer reads as lost_allele. From S132. related RM268, RM362, RM376, RM394
  • ✅ RM403 — SHIPPED 2026-11-07 on main, cut 2026-11-07 as v0.8.1 — a patch, owner enricher, severity low. Every retry policy names net._log_retry and net._log_give_up: one WARNING per retried attempt (client, subject, attempt n of m, the error, the wait) and a give-up line with the attempts and seconds; an AST walk refuses a silent policy. From S133. related RM401, RM42