A hands-on companion to the book. Each chapter opens with a real production failure — then hands you the controls to reproduce it, understand it, and fix it yourself.
The Wrong Data Model Cost Us Six Months
A warehouse stored one row per order. Finance asked for revenue by product line — and the detail had been thrown away at load time. The fix was a six-month rebuild.
How do I know my data model is correct?The Query That Worked Until It Didn't
A revenue report ran green for a year, then reported zero. One NULL order_id in the refunds table, and NOT IN silently excluded every row.
The 300-Line Transformation Nobody Could Debug
A transform mixed pure logic with I/O, mutated data it didn't own, and double-counted on every re-run. Untestable by construction.
Will my code survive production?The Backfill That Corrupted Three Years of Data
A re-run appended instead of replacing. Three years of history doubled, silently, because the load path was not idempotent.
How do I know my pipeline can recover?The Dashboard Was Wrong and Nobody Noticed
One metric, three reasonable definitions, three different numbers. The dashboard rendered confidently — and was trusted for months.
How do I know my data can be trusted?Run the original decision or query and watch it break exactly the way it did in production.
The framework from the chapter, made visual — grain, cardinality, idempotency, in motion.
Make the deliberate choice the chapter teaches, and watch the same question now answer correctly.