The gemara makes a distinction between breira that is "to'leh b'da'as atzmo" - dependent on the person making the stipulation, and breira that is "to'leh b'da'as acherim" - dependent on the decision of someone else. Rashi explains the logic for "toleh b'da'as atzmo" to be less likely to say breira than "toleh b'da'as acherim" is that when it is dependent on the person making the stipulation he is clearly undecided at the time that he is making the stipulation. This type of situation where the stipulator is clearly undecided does not allow us to pretend that the decision made later is retroactive, since right now the stipulator is telling us that he is undecided.Therefore, the clarification that occurs later is really occurring later and cannot be attributed to be retroactive. However, when it is to'leh b'da'as acherim, the stipulator is completely gomer da'as for the action to go into effect, based on the decision of someone else. Since at the time of stipulation he is completely gomer da'as without retaining the ability to control it anymore, the occurrence later can work retroactively.
The difficulty with this distinction is that the gemara considers the case where one separates teruma and ma'aser verbally now, based on what will be left over later, and categorizes that as to'leh b'da'as atzmo. If the whole problem with toleh b'da'as atzmo is that his condition reveals a lack of ability to make a decission now, and shows us that he is riding the fence, why should the case of teruma and ma'aser be to'leh b'da'as atzmo, that should be like to'leh b'da'as acherim since it is not dependent on a decission of the person making the stipulation? Shouldn't this case have the advantage of toleh b'da'as acherim where at the time of his stipulation he is making it completely dependent on some later occurrence of which wine would be left and no longer in his control to decide?