Correct the Multiple sclerosis OMOP concept, restore six lost IPEX judgments

[KrishnaTO]

Answers the question "were any curated mappings removed?" — yes, six were, on one disease. It also corrects an error #78 introduced, which is the more urgent half.

1. Multiple sclerosis has the wrong OMOP concept on main, and that is my fault

OMOP	Concept	SNOMED code
374919	Multiple sclerosis	24700007
4027727	Systemic sclerosis, diffuse	128460000

The SSSOM rows for ARI:0001135 had these two verdicts swapped. ari.equivalencies.tsv had them the right way round. #78 found the two exports disagreeing, resolved it in favour of the SSSOM side, and did not check either concept's label — so main now stores the systemic sclerosis concept on Multiple sclerosis and has dropped the correct one.

The curator's own SNOMED verdicts settle it beyond doubt: they confirmed SNOMED 24700007 and flagged SNOMED 128460000. Those are exactly 374919 and 4027727. Their SNOMED judgments were right; only the OMOP pair was inverted, and #78 propagated the inversion instead of catching it.

Both exports and ARI_OMOP now say 374919.

The generalisable bit

A mapping row is not self-validating. I re-checked every stored OMOP concept against the SNOMED code it carries and the SNOMED codes its disease stores. Five more diseases hold an OMOP concept broader or narrower than their SNOMED one:

Disease	Stored OMOP	Concept
ARI:0001057 Bullous pemphigoid	139899	Pemphigoid
ARI:0001117 Juvenile Rheumatoid Arthritis	4259507	Juvenile idiopathic arthritis
ARI:0001138 Myocarditis due to autoimmune disease	321578	Acute rheumatic myocarditis
ARI:0001144 Neuromyotonia	46272949	Autosomal recessive axonal neuropathy…
ARI:0001196 Systemic lupus erythematosus	255891	Lupus erythematosus

These are pre-existing curation questions, not errors introduced here. Left for a curator — narrowing them is a judgment call, not a repair.

2. Six confirmations were lost on ARI:0001106 (IPEX)

AnjaliRH recorded nine judgments on 2026-08-03. 02938dd (4 Aug) wiped every row for the disease, and her next publish on the 7th restored four. These six never came back:

MONDO:0010580 · OMIM:304790 · ORPHA:37042 · mesh:C580192 · ncit:C131009 · umls:C0342288

Every one of those ids is still stored on the disease, and her changelog entry still names her confirming them — so the review page has been showing six cells as never reviewed when they had been. Restored with her original authorship and date.

This is the first confirmed loss of mapping rows rather than ontology records. #78 concluded the mapping set was additive, but only checked back to PR #69 (16 Aug); this loss predates that window.

3. umls:C0398650 on ARI:0001107

Stored, and named confirmed by AnjaliRH in a changelog entry from 2026-08-10, but no row was ever written. This creates the record that entry implies rather than restoring a deleted one — flagging the distinction because it is a different act.

What remains unjudged is deliberate

I audited every disease for the same shape — an id stored, or a changelog entry naming it, with no judgment in the mapping set. What is left is intentional: 11 ICD-9 codes whose rows the ICD-9 retirement removed on purpose, and one ARI:0003 entry whose row exists under the repaired MONDO:0014523 spelling.

One thing for a curator, not a fix

ARI:0001143 (Neuromyelitis optica) carries a changelog entry from 2026-08-27 confirming OMOP 4027727 — systemic sclerosis again, on a third disease. The mapping set holds the correct omop:380995, so the data is right and only the note is wrong. But the same wrong concept reaching two diseases in one session looks like a mis-click worth knowing about.

Verification

python .github/scripts/validate_mappings.py                     # 0 errors, 5 warnings
python .github/scripts/validate_mappings.py --since main        # 0 errors, 0 warnings

check_deletions passes: removing 4027727 from MS is justified by the Not row this PR flips it to.

Advances #23.

🤖 Generated with Claude Code