Clear the last 25 mapping validation errors

[KrishnaTO]

Stacked on #62 (which is stacked on #61). Review order is #61 → #62 → this. GitHub retargets automatically as each merges.

With this, the audit goes to 0 errors, 17 warnings.

1. Nine null cross-reference rows removed from each export

Four were flagged-wrong rows — a curator recorded rejecting nothing at all. Five were confirmations: a mapping asserted to exist against an identifier that was never written.

They are removed rather than repaired. Supplying the real id is a curation judgment I should not make on a curator's behalf, and recording NoTermFound instead would assert the vocabulary has no term for the disease — which is demonstrably false in at least one case below.

These nine pairs need re-review:

Disease	Vocabulary	Was
ARI:0001005 Acute lichenoid pityriasis	NCIt	confirmed (also has ncit:C37871)
ARI:0001010 Alopecia areata	NCIt	flagged wrong
ARI:0001017 ANCA positive vasculitis	DOID	flagged wrong
ARI:0001094 Female infertility due to antisperm antibodies	DOID	flagged wrong
ARI:0003 Fulminant type 1 diabetes	DOID	flagged wrong
ARI:0001105 Immune checkpoint inhibitor-induced autoimmunity	MeSH	confirmed
ARI:0001110 IgG4 related ophthalmic disease	MeSH	confirmed
ARI:0001113 Insulin autoimmune syndrome	MeSH	confirmed
ARI:0001108 Immunoglobulin A vasculitis	OMIM	confirmed

The ARI:0001105 case is worth a look

#49 supplied mesh:C580192 for that disease by review suggestion on 2026-08-03. A mesh:null row dated 2026-08-07 has since replaced it — the corrected value is gone from the file entirely.

A hand-fixed mapping was overwritten by the same defect four days later. That is precisely the regression the check in #63 exists to stop, and it is the strongest argument for landing that PR.

2. Four mechanical fixes

Each was settled by checking the ontology rather than by judgement:

• MONDO:MONDO:0014523 → MONDO:0014523 in both exports. The prefix was concatenated onto a value that already carried it.
• DOID:111157 → DOID:0111157 in the equivalencies export — the last cross-file drift. The ontology stores 0111157 for ARI:0001011, so the leading zeros were lost on the equivalencies side, not invented on the SSSOM side.
• Two ARI_MONDO values de-prefixed in the ontology: ARI:0001080 and ARI:0002 stored MONDO:0005147 and MONDO:0011027 where the other 55 MONDO values are bare digits.
• ICD-10 range I00-I02 removed from Rheumatic fever (ARI:0001182). Nothing is lost — the disease already records I00 as a single code alongside it, and a range is not an exact match.

Verification

The full diff is 5 insertions and 24 deletions: 3 rewritten rows, 18 deleted rows, 3 ontology lines. Every ontology change is listed above in full.

The 17 remaining warnings are unchanged and non-blocking — 11 confirmed mappings not stored on their disease, 6 diseases holding a DXCODE with no SNOMED counterpart.

Advances #23 ("Fill all missing values from database xrefs" is what the nine pairs above now need); referenced rather than closed.

🤖 Generated with Claude Code