Usability evidence

The v0.4 API shape is backed by ten checked fixtures spanning analysts, researchers, reporting engineers, library authors, and users arriving from ggplot2. The fixture gate counts Python statements with the AST, so formatting does not affect the result:

$ python tools/finishing_usability.py
finishing usability: 10 cases, 5 personas, 40 -> 11 statements (72.5% reduction)

Every candidate keeps native axes access, avoids data mutation, and is shorter than its idiomatic Matplotlib baseline. The aggregate exceeds the roadmap’s 35 percent reduction threshold. Geometry, visual, typing, compatibility, and installed-wheel tests remain separate gates because brevity alone does not demonstrate correctness.

External pilot protocol

Before tagging v0.4.0, the release owner must record at least five uncoached pilot sessions against representative fixtures. For each session record the participant’s background, selected task, whether documentation search was needed, completion outcome, misleading or missing API vocabulary, and resulting action. Do not record personal or sensitive data.

Pilot evidence is a human release approval and cannot be manufactured by CI. The tag checklist in CONTRIBUTING.md keeps it explicit alongside the automated gates.