Four charts I deleted from this site

Writing the caption is a lie detector for the chart. Four times now I have written a caption, realised I could not honestly finish the sentence, and deleted the chart.

One. A line chart of monthly active accounts by plan, four series. Accurate. The caption wanted to say "enterprise is growing fastest", which was true in percentage terms and invisible on the chart because enterprise is the smallest line. Replaced with an indexed chart, all series starting at 100.

Two. A stacked area of revenue by channel. Beautiful. Completely unreadable for any series except the bottom one, because in a stacked area only the bottom band has a flat baseline — every other band's shape is contaminated by the ones below it. Replaced with small multiples.

Three. A scatter of tenure against spend, coloured by six segments. The colours were validated, and it still did not work: in a point cloud every pair of colours appears adjacent somewhere, and six was too many. This is why the chart builder caps scatter at three. I found the cap by needing it.

Four. A correlation heatmap without the values printed in the cells. Fine for spotting the pattern; useless for anyone who cannot separate two of the hues, and useless for reading an actual number. The version that shipped prints every coefficient in its cell, so colour carries the pattern and text carries the value.

The general lesson is not about chart types. It is that the caption is a test. If you cannot write the sentence you want the reader to leave with, the chart is not making that argument, and no amount of restyling will fix it.

I now write the caption first. It takes thirty seconds and it has changed the chart about a third of the time.