[BSAIPF]

ListContains calls pairwise scorers' private _run_eval_* methods, bypassing their public eval() / eval_async() wrappers. This loses arguments bound by ScorerWithPartial.partial() for custom scorers. A case-insensitive matcher that works directly scores 0 instead of 2/3 when comparing ["PARIS", "Rome", "extra"] with ["rome", "paris"].

Use the public scorer entry points for each pair. The regression test covers sync/async evaluation, scorer classes/instances, matching metadata, and the extra-entity penalty. This is Python-specific; TypeScript already invokes the supplied scorer function.

Validation:

• All four regression cases fail before the fix and pass after it.
• pytest py/autoevals/test_values.py py/autoevals/test_json.py -q: 12 passed.
• Broader offline selection on Python 3.10, 3.12, and 3.14: 78 passed on each. The 16 existing live API cases were excluded; network sockets were disabled.
• pre-commit run --all-files, python -m build, and python -m twine check dist/* passed.

On this macOS host, Python 3.10 used the official SciPy 1.15.3 macOS 12 wheel after the default macOS 14 wheel failed to load. No dependency versions or lockfiles were changed.