Regression tests
An improvement for one query group should not damage strong existing routes.
Golden queries
Test critical examples for each relevant release.
Exact matches
Check SKU, EAN and brand terms separately.
Relevance metrics
Compare the same judgments and query set with the baseline.
Zero-result searches
Make sure that the number of empty results does not grow unexpectedly.
Performance
Take query latency and index freshness with you as a guardrail.