Based on the information provided in the knowledge sources, I can offer some guidance on your SRM (Sample Ratio Mismatch) issue and how to handle it in GrowthBook. Here's what you should consider: 1. SRM Detection: GrowthBook detects SRM errors when the actual allocation of participants deviates from the intended allocation ratio. The default P-value for SRM detection is set to 0.001, indicating a very low probability that the mismatch was caused by random chance [Source](https://linen.growthbook.io/t/23127380/u07fj89lffb-i-am-running-an-experiment-implemented-using-url). 2. Adjusting Split Percentages: Generally, you should not adjust the split percentages in GrowthBook to match an unintended distribution. As Graham from GrowthBook states, "you shouldn't need to (nor should you) adjust the split to make the traffic 'more equal'" [Source](https://linen.growthbook.io/t/16429595/hello-i-seem-to-have-a-sample-ratio-mismatch-in-one-of-our-r). 3. Impact on Statistical Analysis: The split percentages defined in GrowthBook do affect its statistical analysis. If you change the split, you may get SRM warnings, and a new phase might be needed for the experiment [Source](https://linen.growthbook.io/t/13253623/i-have-created-an-experiment-and-i-want-two-groups-to-be-add). 4. Handling Setup Mistakes: In your case, since the issue is due to a setup mistake, the best approach would be to: a. Stop the current experiment. b. Fix the assignment logic in your setup to ensure each group receives the intended 5% of the population. c. Start a new experiment with the correct configuration. 5. Starting Over: As suggested by Graham in another context, you can reset the experiment to draft, adjust the weights (in this case, fix your assignment logic), and start again to avoid creating a new phase [Source](https://linen.growthbook.io/t/438432/hello-i-ve-got-a-few-questions-about-the-experimentation-fea). In conclusion, it's not recommended to adjust the split in GrowthBook to match an unintended distribution caused by a setup mistake. Instead, fix the underlying issue in your assignment logic and start a new experiment. This approach will ensure the integrity of your statistical analysis and prevent ongoing SRM warnings.