Only can be determined experimentally. Gather up the top 20 firearm model candidates, shoot each one 1000 rounds for familiarization and break in. Then shoot an easy, medium, hard and long stage each a minimum ofthree times (better would be 10+) with each gun in random order. Determine overall mean time for each stage then calculate mean and range for each gun by stage. Determine deviation from grand mean for each gun by stage.
Evaluate the best guns to determine why they are best, mod all guns to emulate those "best" features and perform a typical slick up / tune up / tweek and repeat the stage runs and calculations / evaluations.
Then (finally) perform a best known set of mods to each and repeat the tests.
Naturally, all failures during all runs would be tracked and incorporated into the final evaluation.
Oh, each gun would need to be provided in calibers 9, 38s and 40 with ammo provided in major and minor. That would tell you what model is "best" in each caliber in box stock, typical modded and fully modded condition.
But, that result would only be totally valid for the shooter that performed the evaluation. So, the whole thing would need to be repeated with 30+ different shooters.
But nobody wants to know that bad! Do they?