Position

This file says what is known, what is not, and what this repository adds about figures and diagrams.

Known

Fact	Grade	Source
Fewer edge crossings help path finding in a node-link diagram	measured	Purchase 1997, Ware 2002, Störrle 2016b
When crossings remain, wide angles hurt less than narrow ones	supported with caveat	Huang 2008, Huang 2008b, Huang 2014
Quantities are read more accurately by position on a common scale than by length, angle or circular area, in charts; angle was not worse than length	measured, for charts	Heer 2010 (Cleveland 1984 for the method)
Items inside one bounded region group together and can override proximity and similarity (not verified: whether measured or demonstrated; connectedness by a line not read)	convention	Palmer 1992
Spatial compatibility between the data region and the legend shortened the time to read a graph (not verified: what compatible meant; not shown for direct labels)	supported with caveat	Huestegge 2011
Beauty gives a small, uneven gain in task performance of interfaces	measured	Thielsch 2019, Schlamann 2026
Holmes-style embellished charts were described as accurately and recalled better two to three weeks later (10 participants per recall condition, charts by one designer)	supported with caveat	Bateman 2010; Few 2011 restates it and criticises it
WCAG 2.2 treats ASCII art as non-text content, so it needs a text alternative; box-drawing characters fall under it only by extension of the glossary definition	convention	W3C 2023a, W3C 2023e
Mermaid can write an accessible title and description into its SVG	convention	Mermaid 2026c
GPT-4o answered most structure questions better from graph text than from an image alone (not cycle checks), on small random graphs	supported with caveat	Zhong 2024

Not known

Whether text art is read worse, or slower, than a drawn diagram by any reader. No study tests it.
How screen-reader users fare on box-drawing diagrams; no test found.
Which visual channels carry node type, link type and group best in a diagram.
Whether alignment, proximity or a bounded region helps a node-link diagram, outside laboratory dot displays.
Whether a flow direction (left to right or top to bottom) matters; one pilot found readers adapt.
Whether appeal changes trust in a diagram; the trust studies use bar charts and infographics.
Whether a language model reads a diagram better from source or from an image when the same authored diagram is given both ways (tested only on generated graphs, in Zhong 2024 and Das 2024).
The gain from a diagram drawn from source over a typed one, for maintenance and for drift.

Where practice runs ahead of the evidence

Every style guide read (Urban 2025, FT 2026, Europa 2023, Rougier 2014) states rules on titles, labels, colour and decoration with no experiment behind most of them.
"Remove decoration" is near-universal in guides, while the controlled studies are mixed on it.
Diagrams as code are adopted widely (held: Mermaid and PlantUML counts), while the accessibility of what they emit is a documented option, not a tested one.
Text art in code blocks is common in repositories; no source here measures its cost or benefit.

Where the evidence is ignored by practice

Wide crossing angles are measured results that none of the four guides read names; a common scale for quantities is named only through one Cleveland and McGill link in FT 2026.
The accessibility standard already covers ASCII art; of the three tool pages read, GitHub 2026 diagrams and Quarto 2026 are silent on alt text for diagrams and Mermaid 2026c documents accTitle and accDescr.

What this strand adds

It joins four fields that are read apart: graph drawing aesthetics, graphical perception, appeal and trust, and accessibility of diagrams.
It states that the case which triggered it, text art against a drawn diagram, has an accessibility argument (a standard and two advice pages) and no reading-speed evidence.
It keeps what a program can check (contrast, alt text, crossings, text art) apart from what needs a reader.
"Nobody has measured this" means none found in 31 searches on 2026-10-29. It does not mean no one has.