Statistics Core

Roll two dice. Everyone at the table knows 7 is the smartest bet — but could you say exactly why, and exactly how much smarter it is than 10? Now suppose you win six rounds in a row and the table goes quiet: is the die loaded, or are you just lucky? You can feel both answers. This track is where you learn to compute them — the first is a distribution (count the ways each total can happen), the second is a hypothesis test (how surprising is six in a row if the die is fair?). The whole of statistics lives in the gap between feeling and computing.

Six modules cross that gap in order. You start by describing data — centre, spread and shape, computed on word counts small enough to check by hand — then watch those counts settle into distributions, the shapes chance keeps producing. The middle of the track turns sceptical: you learn to doubt a difference you can see (is model A really better, or did the coin land well?), and when the textbook formulas run out, you resample — shuffle and re-draw the data itself until the confidence interval falls out. The last two modules bring it home to language: the statistics behind every NLP benchmark score, and the strange, reliable laws — Zipf above all — that every human text obeys.

Every later block of the platform quietly leans on what happens here. "The model improved by 2 points" is a claim about a difference; whether to believe it is a statistics-core question, and you will meet it again in Traditional ML, in evaluation dashboards, and in every paper you read. Each module ends in an interactive lab where the numbers are yours to poke — no calculus anywhere, nothing beyond careful arithmetic and the patience to ask "could chance alone have done this?" one more time than feels polite.