Representative Run for User Flow

[johnbenz13]

Feature request summary
We've been using Lighthouse Navigation for the past years as part of our CI flow.
As recommended to reduce variability we run LH 5 times and we use the computeMedianRun function to retrieve the most representative report.

This works well for navigation mode but now we want to run User Flow tests and we don't know what is the best approach to find the most representative report. In navigation mode we use the normalized distance based on FCP and TTI but is it still relevant for other modes ? Multiple steps reports ?

• What is the recommended way to choose the most representative report ?
• Also why specifically FCP and TTI were the web vitals chosen to compute the median which serves as a base to retrieve the most representative report ?
• Which web vitals should we choose to compute the median for timespan mode ?
• How to handle the case of multiple steps ?