My mentor's mentor used to say that there should be four reserve categories: Proved, Probable, Possible and Impossible. He had a good point.
In short form, Proved means "reasonably certain," and Probable means "more likely than not." My personal favorite is Possible which means "less likely than not." That is to say, the volumes probably are not there.
Following the progression a little further. . .If we estimate "reasonably certainty" to mean 90% confidence, and if we place the limit of Possible reserves at 90% confidence on the other end of the scale, then what should we call that last long, high side tail of 10%? Though there may be estimable scenarios, that long high side tail is, by simple logic, reasonably certain not to exist. I support the proposal of Clark Gillespie; let's talk about a category for Impossible reserves.
Reserves quantified probabilistically are precisely, if not always accurately, demarcated. The the top 10% of the range can be ignored without discussing a label. The problem arises when, as is the wide majority of cases, the quantification is not probabilistic, and there is no computer program to draw the line and give an often illusory sense of confidence.
The main advantage and central challenge of probabilistic analysis is defining the width of the input uncertainties which, in turn, dominate the width of the output uncertainty. The limit of that input width is, by definition, the boundary between possible and impossible. A central question of probabilistic analysis, and that which controls the 10% percentile of outcomes, is defining what is possible and what is not. And that is why probabilistic analysis is best done by someone with a good feel for low probability events, someone who has seen a wide range of scenarios.
But engineers mostly produce singular answers instead of ranges, particularly in ubiquitous decline curve analysis. Many people, even engineers, mistake decline curve analysis for a curve-fitting exercise, and they reason implicitly or explicitly that any backwards curve-fit is approximately equally valid as a forecast. It is not.
Bringing to bear an understanding of reservoir dynamics and observations of similar, older wells can refine the range of likelihood. And experience and expertise from a wider range of samples, as before, help to define that limit.
Less rigorous and more subject to bias, the debate over appropriate curves can devolve into a swearing match. (For an unbiased approach to forecasting, I suggest you look into BetaZi's forecasting algorithm.) But a limit must exist somewhere. At some point, a forecast crosses over from merely aggressive to pie-in-the-sky, to the Impossiblereserve category.
To paraphrase a rebuke my professor once delivered to a his client, "I can make the cash flows say whatever you want, but it isn't going to change what the reservoir is going to do."
(For more on the philosophy of probabilistic analysis, you can read more here.)