Year two

Where it stands after two years.

Done: foundations, pipelines, warehousing, SQL, quality, statistics. Four tools. The dataset library at fourteen, which is more than we planned and the right call — the datasets turned out to be the thing people actually use.

Not done: experiments, modelling, visualization, production. Three more tools.

The thing we keep getting wrong is scope per lesson. Every first draft tries to be complete. Every good final version covers less and says explicitly what it is leaving out. Deciding what to cut takes longer than writing what remains, every single time, and we still budget as though it does not.

Most-read page: not a lesson. It is the dirty-customers dataset, by a distance. People want something broken to practise on far more than they want to be told how to fix things. We have adjusted — every lesson now ends with an exercise against a specific dataset rather than a summary.

Best decision this year: generating the datasets from fixed seeds. It means a rebuild produces byte-identical files, so when a lesson says "row 4,112 is an outlier", that stays true forever. It also means we can bake a specific teaching trap into the data and know it is still there next year.

Worst decision: trying to write the visualization module before building the visualization tools. Three lessons were written, then the chart builder was built, then all three were rewritten. The tools taught us what the lessons should say.

Still free. Two people have offered to pay. One asked where the real course is. This is the real course.