Empirically there will be a better choice between players, but I don't think anything exists yet that can identify that choice to the point where it should rule all of a GM's choices, or hold ultimate sway among closely ranked players. Draft models like PCS help because they give you more information for your draft guesses. The variables used in PCS -- how many are used? 9 or 10? -- might be more than the tip of the iceberg, they might even be most of the iceberg, but they aren't all of the iceberg. You can probably go back a few hundred years and find scientists, kings and popes who could give you 9 or 10 reasons why they knew definitively -- not guessing, but knew definitively -- that the earth was flat. Apart from obvious choices like McEichel and Matthews there's still going to be an amount of guesswork for GMs using PCS and whatever in-house methods they use. In many cases even you can't say that according to PCS, player X is better than player Y; you are limited to saying player X gives a higher probability of being better than player Y -- and that's only according to the criteria PCS has deemed worthy of consideration, and in proportions PCS has deemed appropriate. So passing up an organizational need among a group of closely ranked players because one of them is ranked marginally higher, in a system you admit yourself is imperfect, is too slavish an adherence to rote.
The argument was: does organizational depth/need ever supersede BPA.
Theoretically you use need for tie breakers, but really I have never had, nor have seen anyone else have, two players considered identically equal.
You want to draft the best player available. Period.
You want this because draft picks are weighted lottery tickets. Not only is the best player likely garner you more value, but they are also more likely to actually give any value, since ceiling and safe is a related factor (exception: KHL factor).
In addition, so much can change in organizational needs 4-6 years down the road, when the average prospect enters the NHL as a regular. Look at the "your team in 5 years" thread in the main boards, no one is even close.
The perfection (or really lack thereof) of any model or even scouting is another discussion altogether really.
Once you combine all your information (qualitative factors from scouts and quantitative factors from models like PCS) you try to determine who are the best players. When you do that you are trying to maximize the chance that you are right. You are assuming player A is BPA and not player B. That's what you have decided. You do not go for B saying "hey all this energy and resources I put that assumes A>B may be wrong and B plays D and I need more D".
Yes, your assumption may be wrong, because there are factors not accounted for yet and ones that would be impossible to know yet (freak accidents, etc.). Going for B is considering yourself wrong.
Knowing that you COULD be wrong is different than assuming you ARE wrong.
Going for the need over what you believe to be BPA is doing the latter, you don't want to undermine your scouts and your analysts.