Currently, when running the TI steps, it fetches the data from the last execution, and based on that, it identifies what tests are run and what are not. The next builds that is executed takes the previous build test timings into consideration and splits the tests accordingly.

1) Instead of considering only the last execution of pipeline, we want this data to be checked by taking an average of all executions, and based on that, do the splitting.

Or,

2) Should have a bench-mark build set so that a new build can referr to bench-mark build execution timings for splitting tests to parallel splits accordingly ensuring the overall build time is not increased.