The category that eats the chart

variant on ab-test-checkout has 3 values, and one of them is 34.8% of the data.

• variant_b — 2,089 rows, 34.8%
• control — 1,974 rows, 32.9%
• variant_a — 1,937 rows, 32.3%

The top three take 100.0% between them. With only 3 values there is no tail to worry about, which makes this a genuinely easy column to chart.

select variant,
       count(*)                                      as rows,
       round(100.0 * count(*) / sum(count(*)) over (), 1) as pct,
       round(avg(revenue_usd)::numeric, 2)             as avg_revenue_usd
from ab_test_checkout
group by 1
order by rows desc;

The second column is the one that matters. Share of rows tells you what is common; avg_revenue_usd tells you whether the common thing is the important thing. They disagree more often than not, and a chart that shows only the first is answering the easier question.

Decide what happens to the tail before you plot it. "Other" as an explicit bucket is honest; twelve slivers is not, and neither is silently taking the top eight.