I think #5 is bad metaethics. You write: "Why punish an improvement?" and "A moral framework should make the good outcome and bad outcome as distinct as possible, not the same."

I think this is a holdover from Judeo-Christian metaethics in which there are distinct classes of good things to do and bad things to do (and morally-neutral things in between) and then clear rewards for doing good and punishments for doing bad. In a world without God, morality isn't about punishing or rewarding us, so a moral framework should provide an ordering over choices rather than distinct classes of good and bad with estimates of how good or bad they are. What's useful to know is "What's the most moral thing to do here?" not "How much will I get punished if I don't do this?" because you simply won't get punished.

Re #2-6, I don't think GiveWell has ever said these. Their argument is simply that if you are going to spend money doing good, they will advise you on how to be optimal at it. This is an argument with 80000hours and (in the case of #5) Peter Singer. And #6 I think GiveWell would explicitly disagree with.