If operating with the best model you have is intuitive and not logical, then everything in science it intuitive and not logical, so we might have to redefine your use of "logical" to be more useful.
Peanut exposure caused anaphylaxis in some kids. Many kids in the world are never exposed to peanuts with no ill effects (in fact, a large portion of the globe went all of recorded history up until the Columbian exchange without exposure to peanuts). If we still want peanuts in anyone's diets, waiting until the child is older for controlled exposure is a logical response. Yes we didn't know how that would affect developing allergies, but as I mentioned in my comment, we still don't understand how allergies develop, so the only difference is we conducted a massive peanut exposure experiment with other countries operating as incidental controls.
> I don't know if this increased the total number of people who died. That'd be an interesting question
In fact, this (plus quality of life + cost of prevention/treatment) is the only measure on which the old guidance could be "exactly wrong", which hopefully gives some insight on why talking about it being "backwards" is itself wrong. It's not Thalidomide. Exposure at 6 months isn't the opposite of exposure at age 3. And whatever harm metric you can define is going to be a result of a mixture of effects, including allergic kids that weren't exposed and kids that (probably) developed an allergy because of the lack of exposure. It's almost certainly on the cost side that it's come out negative, because treatment has improved and exposure prevention has become so widespread.
> It's based on a wrong model of how the world works. Yes, that wasn't known. But I feel like we have to account for the fact that we might not know everything. (Admittedly, this drives me particularly nuts because I think this fallacy pervades so much public health advice, especially for kids.)
I mean, all models are wrong, some are useful. We'll never know everything and you have to work with the evidence you have. Delaying exposure to something that could kill 1% of kids is categorically different than some new study saying "we detected a small magnitude but statistically significant result on speech acquisition due to magnesium supplementation".
If you instead mean we need to better communicate uncertainty in developmental and health recommendations, I completely agree. You can see it in this thread, for instance, assuming that early exposure prevents all peanut allergies. Even if you assume exposure is the only causal variable here (almost certainly wrong), we can observe a baseline level of peanut allergy incidence, so, no, early exposure is not a panacea, but that doesn't seem to have been communicated well.