Engine B exceptions + GO/NCBITaxon residual cleanup (48 line-pairs, 28 files)

[realmarcin]

Summary

Part A of the "work on these next" plan (residuals from PRs #90–#107).

• Ontology residuals: 50 → 28 via 9 GO rules and 13 NCBITaxon rules (48 line-pairs across 28 community files).
• Engine B: adds OK_EXCEPTION verdict + per-target exceptions: allow-list so the 28 curator-accepted residuals stop appearing as MISMATCH/ID_NOT_FOUND.
• Validator now reports 5023 OK_CANONICAL, 184 OK_EXCEPTION, 0 errors across communities_yaml and kgx_nodes.

GO (9 rules, 15 line-pairs across 14 files)

7 REPOINTs via OAK replaced_by (IAO:0100001), 1 REPOINT for a bad-map (GO:0051271 → GO:2000146), 1 RELABEL for a rename (GO:0019645 → "anaerobic electron transport chain").

NCBITaxon (13 rules, 33 line-pairs across 15 files)

All replacement ids verified against the kg-microbe ncbitaxon snapshot at data/transformed/ontologies/ncbitaxon_nodes.tsv and re-confirmed in OAK's current sqlite:obo:ncbitaxon adapter. Per [[ontology-term-cleanup]] memory, the OAK snapshot is NOT being refreshed — kg-microbe is the source of truth for current taxonomy.

• Genus renames: Ochrobactrum → Brucella (2 species), Kazachstania → Maudiozyma, Clostridium straminisolvens → Acetivibrio, Rhizobium pusense → Agrobacterium.
• Phylum/clade renames: Asgard group → Promethearchaeati; DPANN group → Nanobdellati; OP3 → Candidatus Omnitrophota; Eisenbacteria → Candidatus Eiseniibacteriota; Dormibacterota → Candidatus Dormiibacterota.
• Spelling/disambiguation: Nitrosotalea devaniterrae; NCBITaxon:2426 → NCBITaxon:43773 ("Syntrophus ").
• 1 RELABEL: Candidatus Micrarchaeota → Microcaldota (same id).

Engine B (scripts/validate_id_label_correspondence.py)

• New OK_EXCEPTION verdict (non-error).
• New exceptions: field on each target (list of {id, label, reason} mappings).
• Loads exceptions per target, applies after canonical/synonym checks.
• 28 communities_yaml entries + 6 kgx_nodes entries (mostly mirrored, plus the exporter-side CHEBI:16793 mercury(2+) cation divergence at src/communitymech/export/kgx_export.py:82).

Cross-repo sync needed

scripts/validate_id_label_correspondence.py is vendored byte-identical across CultureMech / MIM / CommunityMech. The OK_EXCEPTION + load_exceptions() change here must be synced to those repos in a follow-up.

Test plan

• uv run python scripts/term_label_audit.py CHEBI ENVO GO NCBITaxon — residuals match the exceptions list (11 CHEBI + 3 ENVO + 12 GO + 2 NCBITaxon).
• uv run python scripts/validate_id_label_correspondence.py -c conf/id_label_targets.yaml — exits 0; summary shows 5023 OK_CANONICAL / 184 OK_EXCEPTION / 0 errors.
• just validate-all — every YAML validates against the schema.
• just test — 136 passed, 9 skipped.

🤖 Generated with Claude Code

[realmarcin]

Reviewed the Engine B exceptions + GO/NCBITaxon residual cleanup. I independently verified every replacement id against the local OAK adapters (sqlite:obo:ncbitaxon, sqlite:obo:go). All 14 NCBITaxon repoints resolve to the exact replacement labels claimed, and crucially the OLD ids now resolve to unrelated organisms in the current snapshot (e.g. NCBITaxon:1934217 "DPANN group" is now Influenza A virus; :194708 "Ochrobactrum intermedium" is now a plant Adinandra latifolia) — confirming the taxids were recycled and the cleanup is genuinely needed. The genus reclassifications (Ochrobactrum→Brucella, Kazachstania→Maudiozyma, Clostridium→Acetivibrio, Rhizobium→Agrobacterium) and the Syntrophus disambiguation (2426→43773, since 2426 is now Teredinibacter) are all correct. The Engine B script, schema bindings, justfile rc-tracking, and CI workflow look sound. One semantic concern on a GO repoint is inline below.

Findings not anchorable to diff lines

• NIT (AMD_Acidophile_Heterotroph_Network.yaml:300-301 and Anammox_Bioreactor_DNRA_Destabilization_Community.yaml:110-111): GO:0006807 "nitrogen compound metabolic process" → GO:0008152 "metabolic process" drops the nitrogen specificity that is central to these N-cycling communities. I checked the obvious alternative GO:0034641 "cellular nitrogen compound metabolic process" and it is ALSO obsolete, so there is no cleaner current term — the broad parent is a defensible choice. Flagging only so it is a conscious tradeoff (an exceptions: entry preserving the original specificity would be the alternative).

[realmarcin]

RISK: This repoints GO:0019419 → GO:0000103 "sulfate assimilation", but preferred_term (line 405) is "dissimilatory sulfate reduction" and the surrounding notes (lines 136, 169, 364) all describe dissimilatory sulfate reduction coupled to organic-carbon oxidation. Assimilatory and dissimilatory sulfate reduction are distinct processes. GO obsoleted :0019419 with the note "same meaning as sulfate assimilation", which is why the gate steered here — but GO:0019420 "dissimilatory sulfate reduction" is an ACTIVE term that matches the curator's intent exactly. Concrete fix: repoint to GO:0019420 / dissimilatory sulfate reduction instead of GO:0000103.

[realmarcin]

Review: Engine B exceptions + GO/NCBITaxon residual cleanup

Independently re-verified all 48 changed (id, label) line-pairs (15 GO + 33 NCBITaxon) against the local OAK sqlite adapters (sqlite:obo:go, sqlite:obo:ncbitaxon, snapshots dated 2025-06). Every new pair matches OAK canonical exactly, and every retired NCBITaxon id now resolves to an unrelated organism (e.g. NCBITaxon:2426 -> Teredinibacter turnerae, NCBITaxon:1934217 -> an Influenza A virus, NCBITaxon:221109 -> Oceanobacillus iheyensis), confirming the old pairs were stale id-recycling and the repoints are correct. The validator/config/schema-binding/CI additions are coherent. No correctness bugs or fabricated/mismatched references found. Two low-severity notes below.

NIT

• GO:0006807 -> GO:0008152 in AMD_Acidophile_Heterotroph_Network.yaml:301 and Anammox_Bioreactor_DNRA_Destabilization_Community.yaml:111: follows OAK replaced_by, but collapses "nitrogen compound metabolic process" to the top-level "metabolic process", dropping N-cycle specificity in two nitrogen-focused communities. The GO snapshot has no live nitrogen-specific successor, so this is the only gate-passing option and preferred_term retains the specificity -- acceptable, just flagging the information loss.

Confirmation (not a defect)

• GO:0019419 -> GO:0019420 in Iberian_Pit_Lake_Stratified_Community.yaml:407-408: deviates from the ontology replaced_by (GO:0000103 "sulfate assimilation") but is the correct choice -- the prior label "sulfate reduction (assimilatory)" was itself a mismatch, while preferred_term and context are dissimilatory. Good catch.

Findings not anchorable to diff lines

• Cross-repo sync (already flagged in PR body): scripts/validate_id_label_correspondence.py is vendored byte-identical across CultureMech / MIM / CommunityMech; the new OK_EXCEPTION verdict + load_exceptions() must be synced to the other two copies or they will drift from byte-identical.
• kgx_nodes target depends on a build artifact: conf/id_label_targets.yaml globs output/kgx/nodes.tsv. It exists locally, but if CI does not regenerate it before report-label-drift, the 6 kgx_nodes exceptions (incl. the CHEBI:16793 mercury(2+) cation exporter divergence) and the claimed 184 OK_EXCEPTION count will not be exercised on CI. Confirm the workflow builds the KGX export, or note that the kgx surface is host-only for now.

[realmarcin]

NIT: GO:0006807 -> GO:0008152 follows OAK replaced_by, but "metabolic process" is the top-level root term -- the preferred_term here is "nitrogen compound metabolic process", so the N-specificity is dropped from the bound ontology term. Verified there is no live nitrogen-specific successor in the GO snapshot, so this is the only gate-passing option; acceptable, noting the loss. Same change at Anammox_Bioreactor_DNRA_Destabilization_Community.yaml:111.

[realmarcin]

Confirmation (not a defect): repoint to GO:0019420 "dissimilatory sulfate reduction" deviates from the ontology replaced_by of the obsolete GO:0019419 (which is GO:0000103 "sulfate assimilation"), but is the semantically correct choice -- the prior label "sulfate reduction (assimilatory)" was a mismatch and preferred_term/context here are dissimilatory. Verified GO:0019420 is current (not deprecated) in OAK.