Add galvanotaxis with source-bounded flagellar motility evidence

[realmarcin]

Summary

Add traitmech:000581 galvanotaxis (PHYSIOLOGY, CLASS, PROPOSED) beneath
METPO:1000702 motile, with two strain-qualified examples and a five-node,
four-edge flagellar-motility graph. Reserve METPO:1053500 in proposal v458.

The definition requires active movement in response to an electric field.
It does not treat fixed-cell electrophoresis as galvanotaxis or prescribe a
universal anode/cathode direction. The graph does not claim a proven electrical
sensor or infer individual FliC necessity from a fliC/fljB double mutant.

Evidence And Authorities

• Definition: DOI:10.1101/sqb.1988.053.01.006, free publisher Excerpt at
  https://symposium.cshlp.org/content/53/23.short; paywalled full text not accessed.
• K-12 behavior: DOI:10.1128/jb.178.4.1113-1119.1996, PMID:8576046;
  primary abstract inspected, not full text.
• Mechanism and IR715 behavior: DOI:10.1038/s41564-024-01778-8,
  https://www.nature.com/articles/s41564-024-01778-8, Version of Record dated
  August 20, 2024. Publisher full text, Figure 5, Methods, Extended Data,
  Supplementary Table 1 and source workbook inspected. No video was viewed.
• Structural family: https://www.ebi.ac.uk/interpro/entry/InterPro/IPR001492/;
  the issuing API resolves IPR001492 as the Flagellin family, not a terminal domain.
• NCBI resolves NCBITaxon:83333 to E. coli K-12 and NCBITaxon:588858 to
  Salmonella enterica subsp. enterica serovar Typhimurium str. 14028S.
  Green ST is IR715 plus pGFT/RalFc, not the SW473 flagellar mutant or LT2.
• UniProtKB:A0A0F6B2U2 is unreviewed 14028S FliC, entry 42, sequence 1,
  resolved via https://www.ebi.ac.uk/proteins/api/proteins/A0A0F6B2U2.
  https://rest.uniprot.org/proteomes/UP000002695.json confirms its reference
  proteome. UniProtKB REST protein requests failed during source retrieval.
• GO:0071973 grounds the flagellar process, not the organismal trait.
  QuickGO galvanotaxis/electrotaxis searches returned no exact term; broader
  MeSH Taxis Response was rejected as an exact xref. Existing predicates are
  RO:0002211, RO:0002326 and METPO:2007700.

All seven evidence items have directly source-matched contiguous snippets;
aggregate quotations stay within 25 words per source. Europe PMC's abstract-only
checker returned 1 VERIFIED, 3 NOT_IN_ABSTRACT and 2 UNRESOLVED; the latter five
are not proof of paraphrase and were checked directly against publisher sources.

Boundaries And Provenance

• Novelty and allocation searches included ignored and hidden files. Fresh
  seeding returned 399 IDs (344 present, 55 absent); both frozen release-review
  tables were read. A cable-bacteria research mention is not an exact trait.
• Open discussions retain charge/orientation-model uncertainty and the
  differing directedness signs in the source workbook versus published text.
  Direction labels agree; no numeric cosine values or correction are imported.
• The writer defaults to dry run, validates before writes, refuses drift and
  records a MINTED_TRAITMECH_ID event. Repository history:
  history/records/galvanotaxis/2026-10-03T232431Z-codex-3a5ed2.yaml.
• Corpus count is 976. All 975 existing trait pages have footer-only changes
  except motile, which additionally gains the new child link. Reports,
  discussions, priority and QC dashboards were regenerated through maintained
  recipes; shared generator source matched reviewed claw 6d0a6fb (139 files).
• After fixing PR #1642: preserve failed UniProt REST audit result for galvanotaxis FliC #1643, the tracked UniProt report adds the actual HTTP 503 row
  from the maintained REST audit, not an alternate-authority pass. The EBI
  cross-check remains only in the trait's source provenance. All 142 historical rows remain
  byte-identical, including 110 pre-existing findings; they are not claimed as
  freshly revalidated. The full live REST audit exited 1: 143 uses, 112 unique
  accessions, 134 transport failures (114 HTTP 503, 20 timeouts), five older
  entry-version mismatches and four clean uses. The FliC request returned 503.
  Failed lookups also produce empty-metadata comparison flags; those are not
  evidence of biological identity mismatches. This separate temporary report
  did not overwrite the tracked historical observations. No full REST pass is claimed.
• Exact configured embedding inputs and alias path are unavailable. Existing
  embedding artifacts are preserved; rendered coverage is 477/976 (48.9%).
• Protected spore_germination YAML and all older trait YAMLs are unchanged.

Validation

Passed: ten focused tests; direct LinkML and strict record validation; history
schema/links and committed-diff provenance; proposal verification, cross-cohort
coverage and ROBOT/ELK reasoning; graph structure, predicate domains and
protein/taxon coverage; source and research-artifact checks; Ruff; whitespace
check excluding only the required trailing ROBOT TSV header cells.

Live NCBI: 698 examples resolved, zero errors and 24 older-record label warnings.
Playwright: 1440px and 390px, five nodes/four edges, protein tooltip, no page
errors, no horizontal overflow or evidence overlap; screenshots inspected.

Committed-tree just qc and identifier-label validation passed on the initial
head. Full local pytest passed there: 1,831 tests, two dependency deprecation
warnings, 938.24 seconds. The #1643 fix changes only the report, one regression
test and skill guidance: all 19 focused tests pass, Ruff and committed history
pass. Final-head local just qc passed. All eight final-head PR checks passed;
CI pytest run 37163196269 reports 1,832 passed, two warnings in 388.34 seconds.
Use the native merge queue; delete branches only after actual MERGED state.

Review Receipts

• Author-side review initially missed the REST-report provenance defect.
  Independent review found it; issue PR #1642: preserve failed UniProt REST audit result for galvanotaxis FliC #1643 and commit d2bb48f address it.
• Copilot review request was accepted; no submitted Copilot review observed yet.
• Independent Claude review: run 37162537461 completed successfully and
  submitted CHANGES_REQUESTED (review 5403506805) on the initial head.
  Re-review run 37163219590 completed successfully and submitted APPROVED
  (review 5403529808) on d2bb48f.
  No new defect or inline finding remained. The reviewer was offline;
  live-source and verbatim checks were performed by the author, not independently
  confirmed through network access by either review round.
• The uncertain InterPro snippet concern was checked again against a fresh
  issuing API response: exact contiguous match in metadata.description text.
• Targeted PR Shepherd: run 37162538898 completed successfully, found the PR
  unstuck and intentionally posted nothing; it did not perform a code review.
• The approving review's optional future generator suggestion is retained in
  the review: suppress empty-metadata mismatch flags after transport errors.
  This PR preserves the actual maintained output and documents its limitation;
  it does not change the resolver's comparison semantics.

Fixes #1643.

Merge And Cleanup

• Enqueued normally at 2026-10-04T00:01:57Z, with an exact-head guard and no
  admin bypass or early branch deletion.
• All eight merge-group checks passed on
  586532f48d5b74c230c125b09411bef5df66c1fb. Queue pytest run 37163659769
  reports 1,832 passed, two warnings in 385.41 seconds.
• Actual MERGED state confirmed at 2026-10-04T00:09:42Z with that commit.
  Its tree 5d33dfd10f4587f272f1112022778c49b1d78c26 matches the approved head.
• Issue PR #1642: preserve failed UniProt REST audit result for galvanotaxis FliC #1643 closed as completed at 2026-10-04T00:09:44Z.
• Remote feature branch deleted with an exact-head lease; remote absence and
  absence of local/tracking refs verified. Local main is clean and 0/0 against
  origin/main. No review-source or service-availability limitation was hidden.

[realmarcin]

Local adversarial review: checked trait novelty (including ignored/hidden files), motile parent scope, K-12 versus IR715/14028S strain identities, double-mutant versus individual-FliC claims, family versus domain grounding, all seven verbatim snippets, proposal/header/allocation parity, and generated artifact scope. No actionable new curation defect found. Explicit limits remain: the graph is not a complete electrical-orientation mechanism; workbook/paper coordinate signs are not reconciled or imported; 1988 and 1996 access is excerpt/abstract only; full UniProt REST audit failed mostly on transport while the new EBI/proteome snapshot passed. All 975 older trait pages are footer-only except the expected motile child link; no older trait YAML changed. This author-side review is not independent approval.

[culturebot-reviewer]

🟡 IMPORTANT — reports/uniprot_grounding_audit.tsv:110 records a UniProt resolution that did not happen

The added row for A0A0F6B2U2 is a clean success across the board: status=unreviewed, primary_accession_match=YES, taxon_match=YES, entry_status_match=YES, entry_version_match=YES, sequence_version_match=YES, proteome_ids=UP000002695, and an empty finding.

That report has exactly one generator, scripts/audit_uniprot_grounding.py, and exactly one authority: API on line 33, pinned to rest.uniprot.org/uniprotkb/<acc>.json. On a 503, fetch() returns status error with uniprot_name set to HTTP 503, every _same() comparison collapses to NO, and result_row() emits UNIPROT_ERROR plus the mismatch findings.

By the account given in this PR, that is what happened: "The FliC request returned 503", and the evidence note in the record itself says "UniProtKB REST protein requests failed during retrieval; the EBI Proteins API supplied the metadata." So the committed row cannot have come from the generator — the generator has no EBI code path. It was assembled from a different authority and written into a generated artifact, which is what the repository contract forbids.

This matters beyond bookkeeping: the tracked report currently contains zero failure rows — grepping it for UNIPROT_ERROR or HTTP 5 returns 0. It reads as a table of verified resolutions, so a clean row there is a positive assertion that this accession was checked against UniProt REST and passed. It was not.

Either drop the row and leave the accession unaudited until a REST pass succeeds, or commit the row the generator actually produces. Do not hand-place a pass verdict the gate never returned. The EBI cross-check is fine as provenance — it already lives in the evidence notes of the record, which is the right place for it.

What I could not verify, and am not scoring

This environment has no network, so I could not check the seven snippets verbatim, resolve the three DOIs, or confirm that NCBITaxon:588858 is the 14028S strain. That is a gap in my review, not a defect in yours.

One snippet is worth a character-for-character re-read before merge: Flagellin is the subunit protein that polymerises to form the flagella (the contributes to edge). The British polymerises and the InterPro register are right, but the clause reads smoothed — that where entry abstracts of this kind usually run which, and an ending that stops short of the usual the filaments of bacterial flagella. If it is a trimmed near-quote rather than a contiguous substring, it fails the snippet policy. I could not establish either way.

What checks out

• just audit-graphs: 0 new findings, 0 blocking; connectivity 129/129 components, 1567/1567 nodes attached. The five nodes are all wired, every edge subject and object resolves to a declared node_id, and the trait node is reachable.
• just validate-history: 1003 records, 0 invalid. History record present under history/records/galvanotaxis/.
• Groundings corroborated against the reviewed mappings already in the repository rather than taken on faith: GO:0071973 = bacterial-type flagellum-dependent cell motility and InterPro:IPR001492 = Flagellin both match mappings/node_grounding.tsv:176 and :62. The flagellar motility node label is an exact match to the mapped label. modulates to RO:0002211, contributes to to RO:0002326, and confers to METPO:2007700 all match mappings/predicate_grounding.tsv, and BIOLOGICAL_PROCESS -> TRAIT is within the declared domain and range for confers.
• Every README count is live-accurate: 976 trait files, 97 physiology, and 461 PROPOSED / 427 REVIEWED / 50 DEPRECATED / 38 SEEDED corpus-wide.
• traitmech:000581 is unused elsewhere; METPO:1053500 does not collide in data/raw/metpo.owl, which also contains no galvanotax or electrotax class, so the novelty claim holds. The block follows 1053400 from v457.
• METPO:1000702 motile is a defensible and notably tighter parent than the one chemotaxis uses (METPO:1000059 phenotype), and METPO has no taxis class to prefer.
• scripts/add_galvanotaxis_trait.py honours the mutation contract: dry run by default, write_validated_trait, record_curation_event, and a refuse-on-drift comparison before --apply.
• The scope discipline is the strongest part of this change. Bounding the graph to the flagellar-motility branch, declining to infer individual FliC necessity from a fliC/fljB double mutant, keeping host CFTR out, and parking the source-workbook cosine-sign disagreement in an open CURATION_TODO instead of importing or silently correcting the numbers — that is the right call on all four, and the kind of restraint that is easy to skip.

[realmarcin]

Addressed review 5403506805 in d2bb48f (issue #1643). The tracked A0A0F6B2U2 row now exactly matches the completed maintained REST audit: status=error, HTTP 503, and its emitted findings. The 142 pre-existing rows are byte-identical; the EBI cross-check remains in evidence notes only. Added a regression test exercising the actual fetch() HTTP 503 path through audit_uses() and write_report(), plus an add-trait skill rule preventing alternate-authority success substitution. All 19 focused tests, Ruff and committed-history checks pass; full QC/CI are rerunning.

The InterPro snippet concern is not a defect: I re-fetched https://www.ebi.ac.uk/interpro/api/entry/interpro/IPR001492/ and verified the stored snippet is a literal contiguous substring of metadata.description[0].text, including the word that and the stated ending. No snippet rewrite is needed. This is the InterPro family description, not the differently worded UniProt annotation.

Requesting a fresh independent review of the corrected head; no merge while the changes-requested verdict remains outstanding.

[culturebot-reviewer]

Approving. I found no defect.

What I checked, and how

Ontology grounding — all six CURIEs resolve correctly and the labels match:

• METPO:1000702 resolves in data/raw/metpo.owl to motile ("A motility in which an organism has the ability to move independently using metabolic energy"). This is the right parent, and notably a tighter one than chemotaxis/magnetotaxis, which are still parented to the generic METPO:1000059 phenotype.
• GO:0071973 → bacterial-type flagellum-dependent cell motility, RO:0002326 → contributes to, RO:0002211 → regulates (the documented closeMatch target for modulates in mappings/predicate_grounding.tsv:66), METPO:2007700 → confers with BIOLOGICAL_PROCESS in subject_types and TRAIT as the gated object. All already in the mapping tables and therefore covered by the blocking label-correspondence gate, which passes.
• InterPro:IPR001492 is ignored_prefixes in conf/id_label_targets.yaml so it is not machine-resolved, but Flagellin matches the reviewed entry in mappings/node_grounding.tsv:62 and the existing use in spirochete_shaped.yaml.

Causal-graph structure — 5 nodes, 4 edges, every edge endpoint declared, the trait node reachable from all four others. just audit-graphs reports 74 findings, 74 baselined, 0 new, 0 blocking, and 1567/1567 nodes attached. The new connectivity row (5/1/5/5) is consistent.

Numbers — every README edit is exactly right against the live corpus: 97 physiology records, 53 carrying graphs, 51 PROPOSED / 45 REVIEWED / 1 SEEDED, 976 total, 718 graphs corpus-wide.

Identifiers — traitmech:000581 is unique and the next free ID after 000580; METPO:1053500 follows v457's 1053400; proposal v458 follows v457. No pre-existing galvanotaxis or electrotaxis record exists; the only other mention in the repo is the cable-bacteria research lead the proposal already discloses.

Provenance — history record present under history/records/galvanotaxis/, writer defaults to dry run and refuses drift against an existing target, and both record_curation_event and write_validated_trait are used. The file shape matches the thermotaxis PR merged immediately before this one, path for path.

CI: qc, validate-strict, label-correspondence, canonical-example-taxonomy, history, pr-sanity, vendored-sync all pass.

One disclosure about the scope of my review

Network access was unavailable to me in this environment, so I could not independently confirm that the seven snippets are verbatim, nor re-resolve IPR001492, A0A0F6B2U2, or the two NCBITaxon ids against their issuing APIs. The snippet-length budget, contiguity, and the confinement of interpretation to notes all look right, and audit-snippets passes, but verbatim-ness itself rests on the prior review round rather than on anything I did. I am not raising it as a finding — just naming what my approval does and does not cover.

One optional observation, not blocking

The preserved REST-audit row is the right call under the playbook, and hand-editing it would have been wrong. Worth noting for a future generator change rather than this PR: galvanotaxis is now the only status=error row in reports/uniprot_grounding_audit.tsv, and it carries TAXON_MISMATCH|ENTRY_STATUS_MISMATCH|ENTRY_VERSION_MISMATCH|SEQUENCE_VERSION_MISMATCH|PROTEOME_MISMATCH alongside UNIPROT_ERROR. Those five come from comparing against empty metadata after the 503, not from any biological disagreement. A reader scanning the finding column has no way to tell that apart from five real identity mismatches. Suppressing the comparison flags when status=error would make the report say only what it actually observed. That is a change to the resolver, not to this record.