Experiment Stopping Rules

How Long Should You Run Each Test?

Imagine the same stream of ideas run through five stopping rules at once. They differ only in when they call each test. Some wait for a full, well-powered horizon (accurate, but slow to bank a win); some decide fast on a short fixed duration (sooner, but underpowered, so they miss real wins); and some peek often and stop early when conclusive, like mSPRT with error control built in, or naive peeking with none, so it over-claims wildly. The chart banks each shipped test's true lift, and a tick at each rule's end shows how much it over-claimed: what it believed versus what it delivered.