When your opponent has a perceived weak range like middling pairs, and we are going for value, in general how do you come to a decision about the size to use? do you lean towards 1/3 - 1/2 to keep them in? or charge them more (75%)? or larger? is it ok to change this sizing when bluffing vs value? eg betting larger when we don't have it and smaller when we do have it?

You find the discrepancy between MDF and population and see what sizing is incentivized against the player type you are against.

isnt mdf only on the river

and do you mean, understand what the math dictates to defend, and how much you think the population is going to defend, and find a size from that?

I assumed this question was about the river, but I'll answer it as comprehensively as I can.

I would typically only use one sizing on every street prior to the river as a baseline strategy and only deviate if I have a strong exploitative reason. It simplifies the game tree and is easier to execute without costing EV so there is little reason not to do it. I believe many high stakes players have a similar approach.

For barrelling the turn IP, I would typically use 75% pot or 150% pot depending on the board texture. Use the smaller sizing for turn cards that uncap your opponents range and the larger sizing on cards that do not. Use a similar approach when probing turns OOP vs missed cbets.

So for example, if we have a DB opportunity on a board like 7489 we would not bomb the turn because the 9 gives the OOP caller a lot of two pairs and straights.

However on a board like 738Q we would overbet the turn because our nut advantage from the flop is still preserved when our opponent does not raise and the turn doesn't complete any obvious 2 pairs, sets or draws.

If your hand isn't strong enough to bet for your chosen sizing then you shouldn't bet at all. This has the advantage of protecting your checking range with a lot of robust SDV hands so people can't just run you over because you checked a previous street and "showed weakness" or whatever.

On the river, you generally want to bet as big as you possibly can with your nutted hands, at least in theory. It's better to get called less frequently for a bigger amount than more frequently for a smaller amount. To understand why, let's look at the maths of the situation.

Suppose we decide to bet the river for 33% pot. Let's say we bet $1 into a pot of $3. If we were to bluff for this sizing we would need 25% fold equity to break even. This means that villain needs to call 75% of his range to meet MDF and make us indifferent. So 75% of the time we win an extra $1 and 25% of the time we make no more money. That means the EV of our bet is $0.75.

Now let's suppose that instead of betting 33% pot, we choose instead to bet 200% pot- $6 into $3. Now villain needs to fold 67% of the time and call 33% of the time. 33% of the time we win $6 and 67% we win no further money. 33% of $6 is $1.98: more than twice the EV of the smaller bet size despite the smaller sizing being called more than twice as often.

You can repeat this experiment with any number of betsizes you wish and you will find that it's always more profitable to bet bigger with your nutted hands. This is why you will see solvers use massive all in overbets on the river with invincible hands when you give them the option to do so. Of course a solver is going to contract its range perfectly facing different betsizes so as to make you indifferent. However, real life opponents will not do this and will always have some degree of elasticity vs different sizings meaning there will be a "sweet spot" when finding the value bet sizing that yields the most EV. This is why DDP has advised you to consult MDA (if you have access to it) and compare the MDF to the actual real life defence frequency.