Data Quality & SQL
A track of P12 · Data Engineering.
SQL and polars as first-class analyst tools, then the quality machinery - validation rules and lineage - that makes a number trustworthy.
This is where a number earns trust or quietly loses it. The track is hands-on: a real in-browser database and a real columnar engine, worked on the same tiny retail dataset you can total by hand (Paris 40, Lyon 30), so every query's answer is one you can check.
It opens with SQL from the ground — SELECT, WHERE, GROUP BY — then the two things that turn SQL from reporting into engineering: joins, whose silent row-multiplication is the single most common way a dashboard lies (a duplicated dimension row and your revenue doubles), and window functions for running totals and latest-record dedup. polars brings columnar thinking and lazy evaluation for analytics at scale, cross-linking the Python-craft pillar rather than re-teaching pandas. Then the quality machinery that catches the lies before they ship: validation rules that fail loudly on nulls, out-of-range values, duplicates and orphaned foreign keys — the four checks run on a deliberately corrupted copy of the dataset so you see each one fire. The track closes on lineage: tracing a single reported number all the way back to the rows it came from, because "where did this come from?" is the question every trustworthy analyst can answer.