Before you trust a number, check the data underneath it.
Every analysis rests on a table somebody else produced. If three rows were loaded twice, a country is spelled three ways, or a price was keyed ten times too large, the query still runs, the chart still draws, and the answer is wrong. Nothing raises an error.
Can I Trust This Data? is a practical guide to the work between receiving a dataset and acting on it: finding out whether the data can be believed, fixing what can be fixed without destroying the original, working out what the numbers actually show, and presenting the result so that someone can act on it.
It is written in SQL, because a check written as a query can be rerun by anyone, on any day. But duplicates, gaps and wrong values are not SQL problems, so every chapter also shows the same check in the tools you meet at work: pandas, spreadsheets and Power Query, dbt, Great Expectations and the cloud warehouses.
Inside: profiling an unfamiliar table and ten checks for any new dataset · duplicates, missing values and missing rows, wrong values that look right, inconsistent names: where each comes from and how to find it · business rules and master data as queries that return the rows that break them · cleaning in layers with an issues log (bronze, silver and gold; dbt's staging models and marts) · joins across different grains and measuring change without being fooled by the base · the one-sentence finding, honest charts, a data-quality dashboard and a write-up someone else can rerun.
The examples use four real-shaped tables of agricultural exports, prices and farm technology, with the faults real data carries: copied rows, gaps, impossible values, spelling variants, a decimal slip, and rows that compete for the same fact (most added for teaching; the competing rows came with the file). The chapters run 211 queries against them, each printed with its actual result, and 60 figures show what the queries find.
The quickest way to run the examples is the LucidPlexus Workbench, a SQL notebook in your browser with the tables already loaded: nothing to install, and the data never leaves your machine. They also run unchanged in any copy of DuckDB. Data: lucidplexus.com/community/dataset.php?slug=agrisphere-production-market-intelligence
Every purchase includes the companion download: the two-page cheat sheet, the workbook, the SQL file for every chapter with a script that checks each result against the book, and the four data tables.
LucidPlexus Premium. First edition, 2026.