If the model says to take between X and Y units ("but I don't know what's causing the harms outside the range"), then it may be a shallow understanding but it's not failing the SCH, and it avoids a common failure mode.
Exactly! And yet for centuries it was nevertheless an extremely valuable model to have (I should have said "lemons" rather than "vitamin c").
"Hey -- bring lemons on your ships because it stops scurvy"
'Oh, so why not a ton of lemons? Two tons? Ten tons?'
"Well, to stave off scurvy, you only need x units per sailor per day. Beyond that, it's just expensive deadweight."
In contrast, there are policy advocates who want an X to be higher and yet who haven't met the "tradeoff development threshold". Instead of being able to articulate a model which tells you when X is high enough be a net negative, they will show inability to understand the core challenge: "Strawman, no one's advocating 2X". "That would just be absurd." "I didn't say 2X, I said 1.4X."
It's true that if you posit a scenario in which it's physically impossible to steer far enough right to hit Charybdis, then it will looks successful to have the model "steer as far right of Scylla as you can"; but this isn't the general case, and it wouldn't count as an understanding of tradeoffs.
"If I'm going to be late, I should walk faster" isn't modeling velocity — it's modeling punctuality.
I don't see it that way, I see it modeling Vitamin C consumption. To me, your assertion that it's modeling scurvy instead feel contrived in order to fit the top-level comment's princple.
If someone says "take vitamin C", you don't point blindly at a log chart and thus consume a kilogram of it (which, based on a rat model, probably would kill you).
This is because the practical algorithm for "take Vitamin C" has the grocer, the government, and a team of scientists sign off on the size of a dosage and how many pills are even in a bottle. So while that may not be part of your model, it is most definitely part of the model. The model tells you how much Vitamin C not to take, and so it passes the heuristic.
And without that cap, Vitamin C is unsafe, etc., so the heuristic holds.
However, in that example, I'd say that modern medicine certainly has a "too much vitamin C" threshold, and thus might be a sensible model of human vitamin C needs.
I had never seen this idea in writing, but I certainly remember thinking along those lines about retiring age and similar policies.
The point was that the cited model, "scurvy symptoms => too little vitamin C", is useful (in some situations, very useful, if you aren't in possession of a better model with which this one would agree) while being in violation of the maxim given upstream.
I think a much stronger rule - which wouldn't deserve the same catchy name - is that your model should at least be able to say "X is not to low".