I set my distribution to be 50/50 - why isn't it perfectly split?

We handle distribution by distributing based on your chosen probabilities at an independent level for each device.

Even if you set your distribution to 50/50, and you try with 10 different devices, there’s still a possibility, however small, that all 10 devices will be bucketed into one variation while the other gets zero devices.

The bigger the sample you’re distributing the experiment to, the closer the distribution will get to 50/50.

If you've changed your distribution at any point of the experiment, that may affect your distribution as well, so you should be cognizant of that.