research-document RR-VE-THM-2026-FB53
Theme Accessibility and Data-Color Evaluation Protocol
Theme Accessibility and Data-Color Evaluation Protocol
Status: active Related: #13, #14
Principle
A theme is not accessible because its body text passes a contrast check. Evaluate the actual semantic relationships used by the interface.
Required evaluations
Text
Measure every foreground/background pair actually used for:
- body text
- muted/supporting text
- links
- button labels
- input text and placeholder text when meaningful
- status text
- text over accent/status fills
Record ratio and criterion. Never infer a pass from palette membership.
Non-text and interaction
Evaluate boundaries and indicators required to perceive or operate the UI:
- input boundaries
- focus indicators
- selected state
- toggles/checks
- chart marks needed for interpretation
- status icons
- adjacent interactive states
A decorative border may be subtle. A boundary carrying necessary state must meet the applicable functional requirement.
Color independence
For success, warning, error, info, selection, and chart categories, verify that color is not the only information carrier. Use combinations of:
- icon/shape
- text label
- pattern/texture where appropriate
- position
- stroke/dash treatment
- direct data label
Color-vision-deficiency robustness
Evaluate at minimum simulated:
- protan-type red deficiency
- deutan-type green deficiency
- tritan-type blue/yellow deficiency
- achromatic/grayscale condition as an information-redundancy stress test
Simulation is a screening tool, not proof of lived accessibility.
For each semantic pair/category set record:
- distinguishable under simulation: yes/no/uncertain
- redundant non-color encoding: yes/no
- lightness separation
- failure notes
Status palette
Every theme intended for applications SHOULD define or inherit:
- success
- warning
- error
- info
- neutral
Status colors MUST be evaluated against their surfaces and against one another. Do not assume red/green is sufficient.
Data-series palette
For data visualization contexts record:
- maximum intended simultaneous series
- ordered vs categorical use
- pairwise perceptual separation
- CVD simulation results
- light/dark surface behavior
- direct-label support
- print/grayscale behavior
Prefer fewer reliably distinguishable series over nominally different colors that collapse perceptually.
Evidence model
Record each result as:
- computed: deterministic calculation from defined colors
- simulated: algorithmic visual simulation
- observed: structured human review
- studied: controlled user evidence
Do not upgrade simulated evidence to observed/studied.
Acceptance gates
A theme may be evaluated when its declared usage surfaces have completed computed checks and rendered review.
A theme may be accepted only for an explicit context. Example: accepted for marketing pages does not imply accepted for dense analytical dashboards.
Controlled experiment requirements
When comparing variants:
- state the intended independent variable;
- measure OKLCH/lightness/chroma/hue differences;
- report unintended differences as confounds;
- use identical reference content/layout;
- keep semantic role assignment stable unless role assignment is itself the experiment;
- report each outcome independently;
- preserve negative/null results.
This protocol intentionally favors falsifiable evidence over theme-ranking folklore.