Practice — Medium · The Encoding Audit
Skills focus. Two subtle specialty-visual deceptions, built by your own hands
so you recognize them in the wild: the self-correlated scatter and the stretched color scale.
Ungraded. Both traps are built, examined, and reverted — nothing survives to your working file.
Part 1 — The scatter that proves nothing
- Build a scatter of products: X = [Total Units], Y = [Total Revenue] (not average price). Admire the beautiful upward sweep.
- Write down the conclusion the sweep invites. Then write down what revenue is made of, and reconsider.
- Swap Y to [Avg Selling Price]. Describe what the cloud looks like now, and what the honest chart can support that the sweep could not.
Part 2 — The stretched scale
- On a copy of your return-rate heat matrix, open the color-scale settings and set custom endpoints: minimum 2.4%, maximum 3.9% (instead of automatic). Observe the grid become dramatic — deep saturated cells against pale ones.
- Now set endpoints 0% to 10%. Observe the grid become almost uniformly pale.
- Write one sentence on what the "right" endpoints are, and why neither extreme is neutral. Then delete the copy.
Reflection
- Part 1's trap produced a real correlation. Why is "real" not the same as "informative"?
- Part 2 showed the same thirty numbers telling three different visual stories. Which earlier lesson's lie is this, generalized — and what is the defense when you are the reader?