Nice question.

The duration of the experiment is calculated from sample-size and the current throughput received by control group `(sample-size*no.of_variants/(daily throughput on control)`). Hence "sample size needed for seeing the MDL is unlikely to be met for the duration of the experiment" is very unlikely to happen.

But even if that happens I don’t think there is a correct answer for that. It will depends on couple of things.

1. Experiment (like where is it running, if it is running on the payment portal(mde is generally 1-2%) people would not like to increase the MDE as it will be little ambitious, if it is home page landing you can increase the mde)

2. How many days extra you will have to run.

3. How small is the MDE value, if it very small you can increase it little bit, but still that is subjective and will depend on the case.

I hope it clears the doubt your Rajath