The gates¶

This page lists every row the two checkers return, the threshold each one measures against, and what makes it fail.

audit(fig) returns (ok, rows) with one row per figure gate. check(colors) returns the same shape for a palette. Each row is a (name, status, detail) triple, and status is True, False, or "warn".

For what to type when a row fails, see the how-to guides. For why a threshold sits where it does, see the figure style guide.

Figure gates¶

check_figure.py renders the figure through an Agg canvas at MEASURE_DPI = 150, measures the result, and hands the figure back on the dpi it arrived on. audit() returns these 21 rows in this order.

Advisory rows¶

Eight of the 21 rows are advisory. An advisory row can return "warn" but never False, so it never fails a build. ADVISORY_GATES in check_figure.py is the list:

• Overplotting
• Banking
• Ink coverage
• Identity channel
• Style sheet
• Contour dash
• Fonts
• Alt text

Type size is the one row that does both. It fails under the floor, and warns on a figure placed under PLACED_FRAC_WARN = 0.35 of the content width.

Detail strings¶

A row's detail carries up to two marks:

[FIX]

introduces an action. Nothing else uses this mark.

[WHY]

introduces the reason the row fired: the published floor, the perceptual fact, or what a reader loses.

A detail may carry both, in that order:

under 1.0pt on page at scale 0.50: ['a stroke at 0.40pt']
  [FIX] set linewidth to at least 2.00 at this scale
  [WHY] SIAM: lines thinner than one point break up or disappear in print

Every gate except check_collisions names a fix. That gate names the two colliding strings and stops, because which of the pair is free to move is a fact about the layout it cannot see.

Palette gates¶

check_palette.py takes hex strings on the command line or through check(), and imports nothing outside the standard library. Distances are CAM02-UCS dE. The lightness and chroma rows stay in OKLab.

check(colors) returns these five rows:

Ink tokens passed through ink= are exempt from the lightness and chroma rows.

The separation rows are named CVD separation (adjacent) or CVD separation (all-pairs) after the comparison they ran. all_pairs=False is the default and compares neighbours, which is what a line chart needs. all_pairs=True compares every pair, which is what a scatter needs.

Protanopia and deuteranopia are gated. Tritan separation is measured and printed in the detail string, but not gated.

Ordinal rows¶

ordinal=True swaps the five categorical rows for four that apply to an ordered ramp:

What each row tells you to do¶

Eleven rows carry a remedy in suggest_fixes.py, and eight of those come with a runnable snippet. The other ten name their fix in the detail string and stop, mostly because the answer is to draw something else.

Thresholds¶

Every threshold is a module-level constant. To change one, see Change a threshold.

Thresholds cite a published floor where one exists: SIAM's one point, WCAG's 4.5:1, and the Nature, Science, and PNAS type minima. The rest were measured. The figure style guide records the measurement and the figure that motivated each one.