Experiments
Controlled tests that prove which changes actually move search performance, with significance recorded before anything is rolled out further.
- 2
- 1
- 0
- 1
- +12.2%
- 100%
How significance is treated
An experiment is only promoted to a site-wide change once significance reaches 0.95 and the lift holds for a full weekly seasonality cycle. Anything below 0.80 is marked inconclusive and either extended or retired — it never becomes a recommendation on its own.