No GPS-device or running pod is 100 % accurate in measuring distance covered. Please add a sensitity margin to estimated running bests detection, such that efforts that are marginally shorter than the reference distances count as PB.
It is really annoying if you PB in a, say, 5 km race or trial, ending up with a recorded distance of 4.99 km. This difference is well within the error margin of any GPS-watch or running pod, see Fellrnr's excellent research documentation: https://fellrnr.com/wiki/GPS_Accuracy-statistics. Yet, given the hard cutoff Strava is currently using, this effort is not detected as estimated 5 km run best. Even if we assume that 4.99 km is indeed the true distance covered, it is odd to assume that I could not have run the additional 10 meters at approximately the same speed.
Given the evidence, 2 % seems a reasonable and still very conservative sensitity margin, that is, any difference to the reference distance that falls within this margin should be ignored. For example, for the estimated 5 km best, any effort of 5. km x .98 = 4.90 km up to 5.00 km counts.
There are two methods to proceed: (1) Let the recorded time count as PB as it is. Or (2) take the average speed of the slightly short effort to scale the time linearly up. The difference between the two methods should be small (given the sensitivity margin is small), but in my opinion the second method is fairer. In any case, given that the running bests are "estimated" anyway, both should be okay.