Loading and Inspecting Data

The one idea: Look at the data before you compute on it. Fifteen minutes of the same simple checks — row count, column types, the first and last rows, date ranges — catches the errors that otherwise ride all the way into a final report: numbers stored as text, a missing week, a footer row masquerading as an order.

⚠️ The trap: "The first ten rows look clean, so the file loaded fine — I can get on with the metric query." That feels right and it's wrong. The head is the most polished part of any file; corruption hides at the tail (footer rows, truncation, trailing nulls) and in the middle (a format change halfway through a year of appended exports) — check the last rows and the distributions, not just the top.

The statistician John Tukey spent a career arguing one simple point: look at the data before you compute on it. Fifty years later, most bad analyses still trace back to skipping this step. Someone loads a CSV, immediately writes the metric query, and ships a number built on data they never actually saw. That data had revenue loaded as text, dates parsed as strings, and half of March missing.

The fix is not talent. It is a ritual: the same fifteen minutes of inspection, run on every dataset before any analysis begins. Pilots don't skip the preflight checklist because they're experienced; they run it because they're experienced. This lesson is your preflight checklist.

The first-15-minutes ritual

Run these checks, in roughly this order, every time you receive data:

1. Row count. How many rows? Does that match expectation? (You were told "about 50k…