(caveat - based on my observations and experience from a previous life)
Considerations
(a) you need a sufficient sample size to satisfy the Regulator as to validity
(b) subject to (a), you might discard "spare" runs.
Off speed, as you cite, would be good starting point for cause. Gradient didn't appear to be a significant problem .. such trials are done in good wx conditions and the TP generally runs a good gradient to the 50ft point.
At the end of the day, the Regulator has to be satisfied that the data set is reasonable and representative .. but, within that, the OEM is going to skew the set to the low side of the distance distribution to the extent one can get away with ...
Generally the crew will discard off target handling runs while the engineers will be looking more in a global sense to skew towards the low side while preserving a sufficient sample size.
The D/0.6 captures this variation such that the line pilot doesn't have to worry too much, providing that he/she doesn't waste runway for those cases where the available approximates the required.
I am particularly concerned, however, by those who appear to have a belief that using QRH unfactored minimum data is a good strategy.
If that's all you have available to you, fine .. but expect to go off the end in the real world on most occasions if you are on speed and on slope. While you might toy with the idea of coming in below speed and below slope, such strategies come with a new set of problems .. and you are generally better served to go off the end at a lower speed than crash on the approach at whatever speed ..
In all cases, the WHOLE picture needs to be examined and a global assessment of strategy made .. runway distance is just one part of the jigsaw. There is not much point focussing in on one aspect of the solution and then get clobbered by something out of left field that you didn't bother to consider.
In my view, the line pilot's pucker factor should start rising as soon as we go below 1.67 and increase exponentially as the (actual) available factor reduces further ...