P4 · Python & DS Engineering Craft
idioms → vectorization → pandas/polars → scipy; testing, packaging
Writing Python that reads well and scales: idioms, vectorisation, dataframes and testing — the craft layer between knowing the language and engineering with it.
There is a stage after "I know Python" and before "I can be trusted with this codebase", and most self-taught practitioners spend years in it without knowing it has a name. The symptoms are recognisable: loops where a vectorised expression belongs, a dataframe transformation written as five steps that should be one, code that works and that nobody else can safely change. This pillar is that stage, worked deliberately.
It is a single graded ladder rather than a set of topics, because the skill is cumulative. It begins with idiom — the constructs experienced Python programmers reach for automatically, and why they are reached for, since "idiomatic" is otherwise indistinguishable from fashion. Then vectorisation, where the mental shift is from telling the machine each step to describing the whole operation, which is what makes numerical Python fast enough to be useful. Then dataframes, where the same shift applies to tabular work. And finally the engineering layer: code with tests that prove it does what you claim, packaged so somebody else can install and use it. Each rung is practised on its own, so the progress is visible rather than assumed.