A/B Tests

Every test the pipeline has run — verdicts decided in code

2/2 closed with a statistically-sound decision · ~3.5d avg cycle per test · generated variants beat baseline in 2/2 read tests