The first run of this test had two problems:
- The wrong UI message was displayed, telling users that they were seeing rewritten query results (they were not) instead of telling them we have a query suggestion
- The logging didn't record which requests had results provided by glent. If the impact was large this wouldn't matter, but M0 in particular provides low volume but high precision suggestions. To measure any changes here we need to know exactly which requests had a glent suggestion.
Both of these have been addressed and deployed, re-run the test.