Machine learning lab
Box Office Verdict
Type in your clicks. Get the verdict, the odds, and where to send tomorrow's traffic.
How it worksHonest limits
Peeking still costs you
Checking every day and stopping the moment it says Ship raises the odds of a false win. Pick your bar before you start, and treat early verdicts as provisional.
Yes or no outcomes only
It counts clicks, sign ups and callbacks. It does not weigh revenue or order size.
Run whole weeks
Weekend visitors behave differently. Stop on the same weekday you started.
Keep audiences comparable
Each version should get the same kind of traffic over the same dates. Changing the audience halfway breaks the comparison.
Your numbers never leave this page. The only copy is the link in the address bar.
Why this works
The split above is Thompson sampling. Here it runs live against five posters with hidden click rates, racing two simpler rules. Lower regret means fewer visitors wasted on weaker versions.