Suppose I calculate EV of population folding considering 3 bet size. (in bb/100 in PokerTracker4 by example)

As example, based on a 100bb pot :
- I bet 33% of the pot, I win 20 EV.
- I bet 50% of the pot, I win 10 EV.
- And I bet 100% of the pot, I win 0 EV. (population don't overfold here)

On the other side, in my database, even if I bet for value or bluff, the best size for the highest EV (bb/100) is when I bet 100% of the pot, 50% bet in the middle and bet 33% as the worst EV, but positive.

Which is the best strategy based on this data ?

(consider any cases like delay Cbet, River Probe, Turn cbet, etc...)
(consider I have mass data, no sample data issues)

---- Inverted case now---

Suppose another spot.
Now, B33 has the best bb/100 and B100 has the best overfold.

How to manage that spot ?

If you're looking in your database at your winrate when you are betting a specific size, then that is not going to be particularly useful because the std dev and sample size for these situations is such that it means very little.

Considering what you said... Should I only use EV of folding to build my size strategy ? (If I exclude GTO line for the moment.)

It implied that I will add some value in the most overfold line I suppose.

By the way, what you said, made a lot of sense now. I don't think about that in that way. Thanks !

I would mostly just look at the folding frequencies for the given sizing(s).

If we are primarily looking at river bets, because that is the easiest to interpret since there aren't future streets, then the EV a bet can approximated. When you bet there are roughly 4 things that can happen:

1. you bet and they fold and you win the amount in the middle

2. you bet and they call and you win

3. you bet and they call and you lose

4. you bet and they raise --> other lines vs. the raise

If you assume you are betting a ~0 equity hand, then your EV of betting is basically (1) + (3) where every time they call you lose.

EV = (% fold)*(pot in middle) - (1 - % fold)*(your bet)

Let % fold = F
Let (pot in middle) = P
Let (your bet) = B

Then:

EV = F% * (P) - (1 - F%)*B

Input whatever folding frequencies you have for given bet size, and just let P = 1 and B = a percentage of P for the bet size and find out which maximizes EV. I'd mostly be building my strategy around the situations where they are overfolding and under folding and just always bluff and never bluff for those spots and for the sizings that maximize the variable "EV".