I asked the same question. It seems the answer is this:
If you start with HiLo running count/pen with probabilities based only upon that, with no knowledge of anything else, a set of rank probabilities can be computed. Since at that point no aces (or tens) have been specifically removed prob(T) = 4*prob(A). Now specifically remove an ace. RC=RC-1, pen=pen-1 and compute with 1 ace specifically removed. For 6 decks in the new values prob(A) = 96/23*prob(T). In fact you can continue removing aces and until 24 of them have been specifically removed there is a prob > 0 that an ace can be drawn as long as RC/pen is such that prob of drawing a high card is greater than zero. You are dealing with prob(rank) and not how many of rank are present in this model.
An integral representative composition might be used to somewhat reasonably estimate a starting composition. From that point the dynamics of removing cards relative to a count is different from removing cards from an integral composition.
If you start with HiLo running count/pen with probabilities based only upon that, with no knowledge of anything else, a set of rank probabilities can be computed. Since at that point no aces (or tens) have been specifically removed prob(T) = 4*prob(A). Now specifically remove an ace. RC=RC-1, pen=pen-1 and compute with 1 ace specifically removed. For 6 decks in the new values prob(A) = 96/23*prob(T). In fact you can continue removing aces and until 24 of them have been specifically removed there is a prob > 0 that an ace can be drawn as long as RC/pen is such that prob of drawing a high card is greater than zero. You are dealing with prob(rank) and not how many of rank are present in this model.
An integral representative composition might be used to somewhat reasonably estimate a starting composition. From that point the dynamics of removing cards relative to a count is different from removing cards from an integral composition.