Look at distributions and real rows, not only averages

Google's guide warns that means and medians hide things like multimodal behaviour, and recommends histograms, cumulative distribution plots and Q-Q plots. It also says to look at examples: sample rows from each class, and look at the extremes as well as the middle.

A quick first pass on any new column:

• count, nulls, distinct values;
• min, max and a few percentiles;
• a histogram;
• ten random rows, plus the ten largest and ten smallest values.

import polars as pl

df.select(pl.col("order_total")).describe()   # count, nulls, mean, percentiles
df.sample(10, seed=42)                          # real rows, reproducible
df.sort("order_total", descending=True).head(10)

Outliers often point to a real problem, such as a test account or a unit error. Google's guide says excluding them is fine, but only once you know why they are there.