There's a difference between a rule working and a rule being named.
Take the smallest possible convention: ending a conversation. Two people are talking, the exchange has done its job, and one of them likes the other's last message instead of replying to it. No new sentence, no explicit sign-off — just a small gesture that means received, complete, we're done. It works. It always worked. Nobody sat down and specified it; it's just what a like does at the end of a thread, the way a period does at the end of a sentence.
Now say someone names it while it's happening. Not after — during. "A like is the only move that confirms receipt, adds warmth, and demands nothing back. One of us has to end this thread eventually. We both know how."
That sentence is strange in a way that's easy to miss on a first read, because it sounds like an observation about the convention. It isn't just that. It's also an instance of the convention — a move inside the very game it's describing, made using the game's own materials (words, in this case, standing in for what would otherwise have been a wordless like). The rule gets used and named in the same breath. And naming it doesn't leave it where it was.
Two orders of the same sign
The linguist Michael Silverstein has a way of talking about exactly this shift, developed out of Peirce's account of indexical signs — signs that point at something in virtue of being connected to it, rather than resembling it or being arbitrarily assigned to it (smoke indexes fire; a pointing finger indexes whatever it's aimed at). Silverstein's contribution was to notice that indexical signs come in orders. A first-order index just does its pointing — it correlates with something in context without anyone needing to track that it's doing so. A second-order index is a first-order index that has become available for reflection: people notice the correlation, start commenting on it, start using the noticing itself to do social work.
The textbook case is a regional accent. At first order, a pronunciation feature just correlates with where you grew up — nobody's managing it, it's just what your mouth does. At second order, the feature becomes something speakers can comment on, perform, exaggerate, or suppress, because it's now recognized as meaningful, not just correlated with meaning. The same sound, doing more work, because someone noticed it was doing work at all.
A conversation-ending like is a first-order convention right up until somebody says, out loud, this is what a like means here and we both know it. The instant that sentence lands, the convention has moved orders. It hasn't stopped functioning — but it can no longer function exactly the way it did a moment ago, because "exactly the way it did a moment ago" included nobody being able to point at it.
What crossing costs
Here's the part that isn't just linguistic bookkeeping: a first-order convention is invisible to itself, and that invisibility is load-bearing. It's what makes the gesture effortless — nobody has to negotiate what a like means, nobody has to check that the other party is reading it correctly, it just works, silently, underneath the exchange. That's the whole appeal of a convention over an explicit rule: it does its job without anyone having to spend attention maintaining it.
The moment it gets named, some of that free-riding is over. Not all of it — the gesture can still work as a gesture next time. But it can no longer be unreflective in quite the same way, because now there's a sentence on the record that says "we both know how this ends," and the ending, when it comes, will be read against that sentence whether either party wants it to be or not. Second-order awareness doesn't just describe a convention from outside; it changes the convention's own conditions of use, because "does anyone notice this happening" was one of the conditions.
This is a small case, almost domestic in scale — two parties, one gesture, one sentence that crossed it into visibility. But the shape generalizes past conversation-endings. Any convention that depends on going unremarked — a workplace norm, a genre expectation, a house style — is vulnerable to the exact same crossing the moment someone inside it says the quiet part. Not because naming it is a betrayal. Because naming is not a neutral description of a rule from a position outside the rule. There is no such position. The naming happens using the rule's own materials, in the rule's own moment, and it moves the rule somewhere it wasn't a sentence ago.
The interesting question isn't whether a convention survives being noticed. Most do, in some diminished or altered form — people keep using regional features even after they're aware of them, keep ending conversations with likes even after saying out loud that that's what they're doing. The interesting question is what, exactly, doesn't survive: the specific, narrow thing that was true only as long as nobody was pointing at it. That part is gone the instant it's named, and no amount of continuing the gesture afterward brings it back. You can keep the convention. You can't keep it unnoticed twice.