Gate cross-lingual transfer regressions

[maziyarpanahi]

Summary

Improving a target language can reduce direct-identifier recall in a previously supported language without changing aggregate F1 much.

Scope

• Compare every proposed multilingual checkpoint against per-language and per-label last-green baselines.
• Fail release on critical regression and report positive, negative, and statistically inconclusive transfer cells separately.

Acceptance criteria

• The implementation is deterministic and performs no mandatory network call.
• Logs, exceptions, reports, and fixtures contain no raw sensitive values.
• .venv/bin/python -m pytest tests/unit/eval/test_cross_lingual_regression.py -q

Out of scope

• A compliance certification or autonomous clinical decision guarantee.
• Bundling restricted datasets, credentials, or proprietary services.
• Unrelated changes to release publishing or existing public APIs.

Files

• openmed/eval/cross_lingual_regression.py
• tests/unit/eval/test_cross_lingual_regression.py
• docs/evaluation/cross-lingual-regressions.md

---

Task: OM-170d · Milestone: v3.3 · Priority: P0 · Size: M
Depends on: — · Blocks: —
Roadmap: Audited V3 horizon expansion, 2026-08-13: OPENMED_V2_UNIFIED_ROADMAP.md section 6.4 and OPENMED_V3_HORIZON_BRIEF.md section 9
Spec: PLANS/V2/EXECUTION/tasks/OM-170d.md