Find each fault in a table, count it, fix it on purpose, and keep a record.
Two A4 pages on cleaning data with pandas, written on the Verdant Union tables of Green on Paper: emissions, air quality and water. Every snippet was run (pandas 3) and its real output is printed under it.
It covers why a number column arrives as text and how to find the cells behind it, pd.to_numeric and counting what failed, text dates in four formats (three shapes) parsed one stated format at a time and tested against a rule, blanks versus "n/a", na_values per column, what a gap does to a total, duplicates by row and by key, labels with stray spaces and spellings, map versus replace, ordered categories, outliers flagged against their own group, and a cleaning log that counts the rows and cells each step changed.
It shows the faults that give wrong answers without an error: one unreadable cell that turns a 15.5% fall in CO2 into a 20.8% rise, format='mixed' reading 66 dates month first, and an na_values list that wipes out 57 real ratings.
PDF to print, and an HTML version to read on screen. It goes further than the two-page cheat sheet that comes with the book.