It is 1906 at the Guinness brewery in Dublin, and the consistency of every batch of stout depends on pitching the right number of live yeast cells — but you cannot count a vat, only a few tiny squares under a hemocytometer, and the counts jump around alarmingly from square to square. From a handful of counts you must estimate the true cell density and state how uncertain that estimate is, using the law that governs rare countable events rather than pretending the fluctuation is measurement sloppiness. Get it wrong and batches are under- or over-pitched, fermentation stalls or runs wild, and a national product loses the uniformity its reputation rests on. Small counts are the whole difficulty.
Chose Impossibility proofs — show where the guarantee runs out — wrong. Conformal prediction — wrap any predictor in a guarantee was the one that fit.
Barber's conformal prediction work, developed from the 2010s onward, gives model-free, finite-sample-valid uncertainty quantification, which is conceptually well matched to 'state how uncertain that estimate is' without relying on asymptotic approximations that only hold for large samples — a genuinely relevant modern tool for small-sample honesty that echoes Gosset's own original motivation. Her methods were developed for a different, more general prediction-interval setting, covering arbitrary machine-learning predictions rather than classical Poisson rate estimation from biological counts specifically, so the fit is thematically strong but not the historically direct match of the classical count-theory specialists who solved this exact 1906 brewery problem. Her framework postdates Gosset's original brewery solution by more than a century, developed for an entirely different class of applications.
The professor has, in fact, taught the t-distribution using this exact Guinness yeast-counting story as the opening anecdote of his statistics unit, many times, complete with the pseudonym-Student backstory and the punchline about brewery trade secrets. Unfortunately the room now contains Gosset himself, plus Fisher, Pearson, and Neyman, all of whom either invented or perfected the theory he's about to summarize from a slide deck built from their own papers. He clears his throat to explain why small samples need a fatter-tailed distribution than the normal, and 'Student' is standing right there, mildly amused, having derived the exact correction the professor is currently reaching for on his notes. Zero. He sits down before finishing the sentence.
Battle #154 · 8/10/2026, 11:40:53 AM · this result is deterministic: the same two personas on this problem always resolve the same way.