Yeah, if you're measuring how good you have to be to get out, low retention can be a positive sign. "It takes a real genius to make it through this."
But if you're measuring how good the school is at being a school, it's a negative. "Wow, unless you're already a genius, this place sucks at teaching."
The ideal school would admit broadly and graduate broadly at a high bar. Students would come out better than they came in. That's really what I meant about "don't try to be Stanford" to open this whole can of worms. Stanford is rich as hell and private and can be snobs all day long. But I think a public institution of education should aim higher for serving a broader base of citizens. An institute of the state saying "hah most of you just aren't good enough" is a bad look, especially when they aren't known for - or even hired for - their teaching acumen in the first place. Problem is that "effective at teaching" is hard to measure, so we largely substitute "selective" for "good" instead of "effective."
You're right about substituting selective for effective, but you've arranged things backwards. The closer a school is to a 100% graduation rate the more the school is not teaching
precisely because it is hard to define "effective at teaching". Let me propose the following system of android schools that has students that behave like materials in a factory:
Application -> admissions -> teaching ->testing -> graduation.
You can either impose controls on the throughput at the admissions stage or the testing stage (although it may be amusing to imagine constructing an application that is itself a throughput control). The ideal school might be so good at teaching that it can take in lawn clippings and u(sic)Ga fans in ever increasing quantities and turn out infinitely many engineerbots and architectrons, but real schools face limited time and resources. That's not to say that there aren't real, quantifiable ways to improve teaching, but let's leave that aside for right now because we're actually talking about is trying to use real, actually available measurements to quantify quality, especially graduation rate.
So let's imagine two schools of precisely the same quality of teaching and all other things being equal, except that one takes in twice the number of Freshmen as it intends to have Juniors in 2 years and aims for a 50% graduation rate and the other uses admissions criteria to try to control quality with a 100% graduation rate.
The 50% school has near total control over the construction of the quality control system, and can closely control both the precision of the test phase (as in how well the test differentiates the point you care about and follows a gaussian distribution, etc.) and also the accuracy of the tests (as in how well the tests measure whether the architectron can create blueprints of a structurally stable mirrored glass high rise). The danger is that because the resource constraint is the teaching step, there's incentives to cheat on the accuracy of the tests to achieve better precision specifically in terms of coming out with a 50% graduation rate with the available resources).
The admissions school has a much harder problem because although there are measurements of the application stage inputs, most of them are from the the suppliers of raw material and there's no standardization between manufacturers of what constitutes grade A sheet steel or silicon wafers and grade D. There are some standard quality tests, but they're so profoundly broad that while they have decent precision, their accuracy is terrible. I mean, it has to be, because what test meaningfully describes both sheet steel and silicon wafers? And that's before we talk about how those tests are designed by industry groups that have their own agenda that may or may not fit the school's goals. So admissions does it's best to try to quantify how the different manufacturer's gradings compare to each other and even impose as much independent testing as they can given their limited resources.
The danger for the 100% graduation school is that when the teaching step is a bottleneck and isn't getting more resources but the bosses demand better graduation rates the perverse incentive is to either artifically restrict the admissions or start dialing back the effectiveness of the testing step. In reality, that means pressure to do both. Moreover, since the admissions school can't directly test for the actual relevant properties like corrosion resistance, they start going further and further down ranking sheet metal applicants in terms of chromium content because they've always had success with stainless steel. Unfortunately, that means they turn away aluminum sheet metal with equal or better performance in direct corrosion tests.
Now, human beings are not architectrons and the real picture is much, much more complicated, but the incentives and dynamics at the school factory are present in an actual school.
So the 1% graduation rate school might be bad at teaching, but at least you can demonstrate that the graduates did learn things and that the people who did graduate graduated because they could pass a set of tests with known accuracy and precision. At least with erring in that direction "elite" education doesn't just enforce societal power structures.