Claude review false positives on HGNC CURIE case (should prefer lowercase hgnc:)

[cmungall]

Summary

Claude PR reviews are repeatedly flagging lowercase HGNC CURIEs (e.g., hgnc:746) as errors and recommending uppercase HGNC:746. In this repository, lowercase hgnc: is the canonical/expected form and validates.

I (Codex) observed this across multiple metabolic PR reviews, where this was treated as a blocking issue despite term validation not requiring uppercase.

Examples

• PR Add Argininosuccinic Aciduria disorder entry #404: review repeatedly flagged hgnc:746 and requested HGNC:746 as blocker.
• Similar pattern appeared in other review rounds where HGNC case was treated inconsistently with repo conventions.

Why this is a problem

• Creates false-positive “blockers” and review churn.
• Conflicts with existing repo guidance and examples using lowercase CURIE prefixes.
• Wastes cycles on non-issues while masking real ontology-placement issues (e.g., GO MF terms in biological_processes, UBERON in GO slots).

Suggested fixes

1. In .claude/skills/dismech-pr-review/SKILL.md, explicitly require and prefer lowercase hgnc:.
2. Add a “do not flag as error” rule for valid lowercase HGNC CURIEs.
3. Add a validation-grounding step to reviewer output:
   • If claiming HGNC case error, cite failing command output from just validate-terms-file <file>.
4. Add a short reviewer checklist item:
   • “Only mark as blocking if schema/term validation fails in-repo.”

Optional follow-up

• Add a lint/check (or test fixture) that enforces canonical lowercase prefixes for HGNC to remove ambiguity.

[github-actions]

Summary & Analysis

This issue tracks a recurring false-positive in Claude's automated PR review: the reviewer incorrectly flags lowercase hgnc: CURIEs as errors and demands they be changed to HGNC:.

Verified: lowercase hgnc: is the canonical form

A grep across the entire knowledge base confirms all existing HGNC gene references use lowercase:

kb/disorders/Phenylketonuria.yaml:      id: hgnc:8582
kb/disorders/Salla_Disease.yaml:        id: hgnc:10933
kb/disorders/Neurofibromatosis_Type_1.yaml: id: hgnc:7765
kb/disorders/Bacterial_meningitis.yaml: id: hgnc:11849
... (all files consistently use lowercase)

No uppercase HGNC: IDs appear in the disorder files.

Root Cause: Misleading comment in conf/oak_config.yaml

The OAK config file contains a comment that likely confuses AI reviewers:

# Gene symbols - uses HGNC (note: OAK uses lowercase 'hgnc')
# Data files use uppercase HGNC: prefix, but OAK requires lowercase
HGNC: sqlite:obo:hgnc

This comment is factually wrong — data files do NOT use uppercase. This may be actively causing reviewers (human or AI) to believe uppercase is the norm. This comment should be corrected.

Related Issues

• Retrospective (Codex): tighten PR review workflow skill + add respond-to-reviewer skill #441 — Retrospective (Codex): tighten PR review workflow skill: Broader retrospective about review process failures. Proposes adding hard guards and validation-grounded review discipline. The HGNC case issue is a specific instance of the general problem identified there: reviewers blocking on non-issues while missing real schema/ontology mismatches (e.g., GO MF terms in biological_processes slots).

Concrete Changes Needed

1. Fix the misleading comment in conf/oak_config.yaml:

   # Gene symbols - uses HGNC
   # Note: data files use lowercase 'hgnc:' prefix (e.g., hgnc:746)
   HGNC: sqlite:obo:hgnc

2. Update .claude/skills/dismech-pr-review/SKILL.md to add an explicit rule — currently the skill has no mention of CURIE case. Suggest adding to section 9 (Genetic Section Content):

   HGNC CURIE case: Canonical prefix is lowercase hgnc: (e.g., hgnc:746). Do NOT flag lowercase as an error. Never require HGNC: uppercase. Only flag as blocking if just validate-terms-file <file> actually fails.

3. Add a "validation-grounding" principle to the skill (also requested in Retrospective (Codex): tighten PR review workflow skill + add respond-to-reviewer skill #441): any reviewer claim of a schema/term violation must cite the output of just validate-terms-file <file>. If the validator passes, the complaint is not a blocker.

4. Optional: Add a test fixture or lint rule that confirms all HGNC IDs in disorder files use lowercase prefix, removing ambiguity permanently.

Priority

Medium — this is a process/quality-of-review issue, not a data integrity issue (the data is correct). However, it wastes significant cycles and creates reviewer fatigue, as noted in #441. The fix to oak_config.yaml is a one-line change that should be done immediately to prevent further confusion.

@dragon-ai-agent please fix the misleading comment in conf/oak_config.yaml and update .claude/skills/dismech-pr-review/SKILL.md to explicitly state that lowercase hgnc: is the canonical CURIE prefix in this repo and should never be flagged as an error by reviewers