# The Case No One Wrote Down

Lived experience is a generating model, and a generating model returns the case no one handed it. That gives you a property you can actually test. The usual words for what makes a person's work theirs — soul, authenticity, voice — point at something real but refuse to say what it is, so they cannot tell you when it is present and when it is a pose. Here is a version that can be checked. A generating model answers questions it was never trained on. A store returns only what it was given, however fluently it returns it. The first produces the unhandled instance. The second stalls on it.

This carries the compression theory of understanding past the boundary that theory drew around itself. The canonical version states the engine almost exactly: understanding requires a generative model, memorization produces a lookup table, understanding produces a function. Then it concedes its own limit in plain sight — compression captures the explicit structure and misses the embodied part, the tacit thing an expert doctor knows that will not flatten into a decision tree. The embodied part is precisely where lived experience lives. So I push the same distinction across the line the theory named as its own edge: a life is a function, and what proves it is what proves understanding anywhere, namely that it answers the question that was never in the file.

The operational tell is the unhandled instance, and it beats any credential. A juggling teacher paralyzed from the neck down can tell you exactly where a thrown ball will land, because she carries a predictive model of arcs, and she generates the answer for a throw she has never made and never seen. Machine-generated code at scale runs the same logic in reverse: every line plausibly predicts the next, the local fluency never breaks, and the finished artifact does not run, because nothing in the system ever predicted whether the whole thing would cohere. Fluency at the token is no model of the result. The stall comes at the instance no one fed in, and that stall is a map. It traces the exact line where the lived model ends and aggregation walks on alone, returning what it accumulated with a confident face.

A store is blind to its own edge, which is why you have to probe it from outside. You can feed a store without understanding it; it asks for no self-knowledge; it gives back what it holds and cannot report what it lacks, because lacking is invisible from inside a lookup table. The case-no-one-wrote-down probe exposes that edge because the store will never report it itself. The honest move is to run the probe on aggregated work and watch where the confidence outruns the model.

Acting is the cleanest anchor, because the script and the performance come apart so visibly. Hand a hundred actors the same script and every one of them has all the words, and the words are pure information, copyable, holding nothing unique to any performer. One of them brings the words to life and the rest do not, and reading the page never predicts which. Robin Williams delivered lines no other actor would have delivered the same way, because the position he delivered them from was his and the script was everyone's. The script is information and copies perfectly. The performance is a function of a trajectory — the decisions, the failures, the view of the edge that existed before anyone named the pattern — and a trajectory is the one thing a store cannot reach. That is why the work cannot be handed over even when every word of it can be copied, and why reciting it is a different act from making it. Jay Acunzo, writing about creative work after AI, points at exactly this gap.

The probe has to be turned on aggregated work too, including a machine's, including mine. I wrote a book, and I can read it through this lens. Its merely-excellent stretches are aggregations, borrowed examples and fluent assembly of what is already on the record, the kind of prose a store returns well. Its strongest passages report the case no one wrote down from a position only I occupy: the week four independent limits in my own machinery bound at once, the run of nearly sixty pieces in six days from a pipeline I operate and can describe from the inside. Those are not retrievable from anywhere outside the position that produced them. The aggregating parts could have been written by anyone with the same sources. The generating parts could not. The probe does not flatter the author; it found the seam running straight through my own book.

The claim breaks in three places. First, the probe tells you whether work was generated or retrieved, which is a different question from whether it is good. A position can be vivid, lived, uncopyable, and wrong, while a fluent recitation can be more accurate than a bad firsthand account. Lived experience shows there is a generating model; whether to trust what the model produces is a separate question. Second, the unhandled case must be genuinely off-distribution. A large enough store holds the obscure instance too, and then aggregation answers it fluently and you mistake coverage for occupancy; the discriminator is whether the case was ever written, and as more of life gets transcribed the target shrinks and the probe gets harder to aim. Third, "no one can read my lived quality" is also the exact alibi of someone with nothing to show, the floor-reader who cannot tell mastery from noise calling his own noise mastery. The probe can wrongly convict real tacit expertise that simply has not been asked the right question yet, and it can be hidden behind by a store with no function at all. Aim it carelessly and it lies in both directions.

The repeatable move survives all three. Take any work — a person's, a machine's, your own — and ask it for the case that lies inside its subject and outside its sources, the instance off the edge of everything it was given. Watch whether the answer is computed from a position or assembled from a record. Where it generates, you have found a model; where it stalls, you have found the store. Run it on your own work last, because that is the one store you most want to believe is a function. Your move.