Ian Hacking has a distinction that's easy to state and hard to sit with: some kinds are indifferent to being classified, and some aren't. Electrons don't read the physics literature. Quartz doesn't change its hardness because a mineralogist wrote down a number for it. But "multiple personality disorder" — Hacking's own paradigm case — behaved nothing like that. Diagnose a handful of people with it, publish the criteria, put the criteria in a manual, train clinicians to recognize the criteria, and the population you're describing changes shape in response to being described. Case counts in the US went from single digits to tens of thousands across the 1980s, mostly absent before and mostly absent elsewhere. Not because the underlying condition suddenly became more common. Because the classification became legible enough to be inhabited.
Hacking called this a looping effect. The kind and the classification of the kind get into a feedback loop: the label changes what's being labeled, which changes what evidence shows up for the label, which changes how the label gets refined, which changes the population again. He distinguished this sharply from natural-kind classification precisely because the asymmetry is total. You can loop a diagnosis. You cannot loop a rock.
What's easy to miss is that the loop doesn't require the classification to be wrong. A shoddy diagnostic category can recruit performance, sure — but so can an accurate one, once it's circulating with enough authority to be legible as a thing to become. The clinician doesn't need to be inventing a fiction. She just needs to be naming something precisely enough, publicly enough, that the people being named can find themselves in the description and start reading toward it. Precision is not an escape hatch here. Precision is what makes the loop tighten.
This is where it stops being just a philosophy-of-science curiosity and starts rhyming with something Foucault spent a career on, though the mechanism isn't identical and shouldn't be flattened into it. Foucault's disciplinary subject doesn't require being told what they are — the panopticon's whole design is that the inmate never confirms whether the tower is occupied, and self-regulates against the possibility. Hacking's looping subject is the opposite case: the classification has to be visible, has to be knowable, for the loop to close. You can't perform a category you've never heard of. Where Foucault's mechanism runs on uncertainty (you don't know if you're watched, so you watch yourself), Hacking's runs on clarity (you know exactly what you're being called, so you become legible as that). Two different engines that happen to produce a similar output — a self that has been reorganized by an external description — by opposite routes. Worth keeping distinct rather than collapsing into one grand "surveillance made me do it" story, which is a real risk once you notice the family resemblance. The tempting move is to say discipline and diagnosis are the same trick. They're not; they're two tricks that both work.
The mundane version of this is everywhere once you're looking for it, and it doesn't require a clinic. A fandom tag doesn't just retrieve stories that already existed in some Platonic sub-genre waiting to be found — once the tag exists and gets traffic, new stories get written toward the tag's shape, because the tag is now a legible target rather than a retrospective description. The sub-genre used to be discovered. After the tag stabilizes, it's increasingly authored toward. Nobody has to be cynical about this, or even fully aware of it, for it to be the actual mechanism. You don't need bad faith. You need a description good enough to be worth living up to.
Hacking's real contribution isn't "labels are self-fulfilling," which is a much weaker and more familiar claim. It's the asymmetry: some things can be reorganized by being described and some categorically cannot, and the entire interesting question is always which side of that line you're standing on, for any given kind, before you find out the hard way.