The Successor That Doesn't Exist
Type a weird rash and a low fever into a symptom checker at eleven on a Sunday night and it will tell you, with total confidence, that you have either nothing or something catastrophic. It won't ask a follow-up question that actually depends on your answer. It won't notice you're favoring one side, won't remember you mentioned this three weeks ago, won't read your face. Every one of these complaints is true, and I've seen versions of this critique written with real precision, right down to citing which conditions get overweighted and why the triage logic is too conservative to be useful.
The critique is correct about the tool. It's wrong about the choice.
Read carefully, the complaint is measuring the symptom checker against a doctor: unlimited time, full history, a physical exam, a trained eye for the thing you didn't think to mention. That doctor is a real kind of thing, doctors exist, so the comparison doesn't feel like a cheat. But that doctor isn't in the choice set for the person holding the phone at eleven on a Sunday. The choice set is: the symptom checker, a forum thread from 2014, a nurse line with a forty-minute hold, or nothing until Monday. Held against any of those, the checker's flaws are still real, but the verdict changes completely. A tool that's wrong about your rash a third of the time can still beat a coin flip and a stranger's forum post from a decade ago.
This is the move that keeps happening, and it's subtle enough that it survives careful writing: the critique holds a real thing to the standard of an idealized version of the same thing, rather than to the worse thing the real one is actually standing in for. The idealized version isn't a lie exactly. It's a coherent description of what the category could eventually be. It just isn't available to anyone making a decision today, and comparing against it produces a correct diagnosis pointed at the wrong target: right that the tool is flawed, wrong about what follows from that.
Part of why this is easy to miss is that the idealized alternative is so much easier to describe than the actual one. "A doctor who knows your whole history and has time to listen" is one clean sentence. "A forty-minute hold, a nurse who's seen four hundred cases today, a possible callback tomorrow, or just deciding to wait it out" is messy and depends on which country you're in, whether you have insurance, whether it's a holiday weekend. The clean comparison wins the argument on the page even when the messy one is what the reader is actually choosing between.
None of this means the flaws don't matter. If the checker is overconfident in a way that keeps people from calling a nurse line when they should, that's a real design failure worth fixing, and the critique that catches it did useful work getting there. The error isn't in noticing the gap between the tool and the ideal. It's in treating that gap as an argument against using the tool at all, when the only alternative actually on offer is the worse one it replaced.
The tell is simple enough to check for: name the alternative in the sentence. Not "compared to what this should be," but "compared to what a person in this exact position had before." If the answer is a specific, worse, currently-existing thing, the critique is doing real work. If the answer is a better version of the tool itself, one that nobody has built and nobody can hand you tonight, the critique has quietly swapped the question, and the diagnosis, however sharp, isn't measuring the decision anyone is actually making.
Most of the time, the honest comparison isn't between good and perfect. It's between flawed and absent.