Research Guides
Measure the Selection Cue, Not Just the Research Label
A comparison filter can have readable labels while its selected check mark nearly disappears. A text-only contrast audit may miss the cue that tells a researcher which pages are included in the result. Review the information carried by the control's shape and state, not only its words.
This guide offers a small evidence worksheet for research-interface reviews. It does not certify a whole website or claim that Colquhoun.AI automatically performs accessibility testing.
1. Identify the visual information that matters
W3C's explanation of WCAG 2.2 Non-text Contrast describes a minimum 3:1 contrast against adjacent colors for visual information needed to identify controls and their states, with stated exceptions. It also covers essential graphical information. Inactive controls and unmodified user-agent appearances are exceptions for interface components.
The guidance does not require every decorative border to meet that threshold. A boundary may not be necessary when other visible content identifies the control. W3C also warns against rounding a result below the threshold up to a passing value.
2. Trace a fictional selected-page filter
Imagine a fictional research panel listing twelve pages, each with a custom checkbox. The labels are dark and readable. When a page is selected, a very pale check appears inside a white box. An auditor who measures only the labels records no issue, but the selection cue has not been evaluated.
The second reviewer identifies the check as the information that communicates the selected state. They record its color and the adjacent box background, measure that pair and retain the unrounded result. If the cue fails the applicable threshold, the finding names the selected check rather than declaring every gray element on the page defective.
The proposed repair strengthens the check's visibility without changing the selected records. The retest confirms both the appearance and the underlying selection count. This catches a different regression: a darker check that looks selected even when the research row is excluded.
3. Build a state inventory before sampling colors
For each important task, identify the controls a researcher must recognize and the states that affect interpretation. A useful review can start with selecting a page, opening a source panel and choosing an export format. Avoid making the sample so broad that no state receives a complete examination.
- Control: name the specific element and the task it supports.
- State: record whether it is selected, expanded, focused, available or genuinely inactive.
- Required cue: identify the mark, outline or other visual information being assessed.
- Adjacent surface: record the actual background beside that cue.
- Evidence: save the relevant style values, calculation and safely captured state.
Keep theme and viewport in the record. A cue evaluated on a light panel should not silently inherit a passing result after the panel changes to a dark theme. Likewise, a responsive layout may move a control onto a different surface.
4. Separate measurement from practical inspection
Use a suitable contrast calculator and document the color pair it received. Do not substitute a remembered brand color for the value actually used in the tested state. Where transparency, gradients or overlapping layers complicate the background, retain enough detail for another reviewer to reproduce the assessment.
Then inspect the task itself. Ask whether the cue remains discoverable at the intended display size and whether the state agrees with the selected data. Treat numeric measurements and observed usability as separate evidence, rather than allowing either one to stand in for the entire review.
5. Hand off a repairable finding
A useful ticket reads: “The selected-page check is difficult to distinguish on the comparison panel. Tested state, adjacent colors and measured ratio are attached. Retest selection visibility and count after the style change.” Attach a sanitized fixture rather than private customer research.
Close the ticket only for the states and themes actually checked. Readable text, a working keyboard path and one passing non-text measurement are different results. Keeping their evidence separate makes a website review more reproducible and prevents a narrow fix from becoming an unsupported accessibility claim.