Pretty much this. 3, 5 or really even 10 can't show you how big the extreme spread of the group (group size) is going to be but it can tell you that it will ATLEAST that size so if its worse than what you want a 30 round group to look like at 3 shots adding more data is irrelevant as i dont care HOW bad it it just that its bad at that stage. It doesn't give any hard data on what teh final result will be.
It's not contrary as there are 2 seperate statements. Low samples sizes cant prove how precise a load is. And a low sample size can show how precise a load could be best case. It takes alot of shots to get an accurate measure of precision (mean radius stabalises around 30 and is a better metric @gimp has done lots of posts on how to get this. There are a few apps but I digress. It is true in the case it can prove that a load won't ever be better as its already shown it can be atleast that bad. But it one 5 shot group is under the precision required more data is required to confirm it will remain under with time but if its already too bad then no amount of shots will make it smaller ie a 2.5 inch 3 shot group at best will be a 2.5 inch 30 shots group. But a 0.1 inch 3 shots group could also end up a 2.5 inch 30 shot group. The more shots the more likely it will and the smaller the likely deviation.
I generally work on as soon as a shots thrown outside 2 inchs I stop testing the load because im looking for a bear minimum of 1.75inchs at 100 for 30 rounds. But obviously ideally want something better. In field conditions the hit probability difference between a 0.5 and 1.5moa system with an average wind caller is very minuscule as other factors come into play.

