Research Guides
Test an Aborted Press Before Removing a Research Source
A researcher presses the wrong control, notices the mistake and moves away before releasing. If the interface already changed the evidence list when the press began, the chance to cancel came too late. Test the beginning and end of a pointer action separately, not merely whether a successful click works.
This guide supplies a cancellation test for research-interface reviews. It does not claim a particular Colquhoun.AI implementation or certify an entire interface as accessible.
1. Identify when the action takes effect
W3C's Pointer Cancellation guidance describes alternatives for single-pointer functionality: avoiding execution on the down-event, completing on release with abort or undo available, reversing a press outcome on release, or meeting an essential-use exception. It explains how ordinary release-based activation helps users avoid an unintended action by moving away before releasing.
This is a timing and recovery question, not a target-size measurement. A large control can still execute too early. Conversely, a cancellation test does not establish that a small control is easy to reach or that it has an accessible name.
2. Inspect a fictional evidence-list control
Imagine a fictional source-review panel with a “Remove from comparison” control beside each source. Its custom handler removes the row as soon as the mouse button is pressed. The reviewer moves off the control before releasing, but the row is already gone.
The finding should name the triggering phase and the observed data change. A screenshot showing a pressed appearance is not sufficient evidence that removal occurred. In an isolated fixture, inspect the comparison list itself and verify whether the source is still included.
A proposed repair might use an appropriate native control and normal activation behavior rather than an immediate press handler. The developer must still check that the old handler is removed; adding another handler can accidentally make one deliberate action run twice.
3. Run three distinct cases
- Deliberate completion: press and release on the intended control. Confirm exactly one intended result.
- Changed intention: press, move away while still pressed, then release. Record whether the action is canceled under the chosen interaction model.
- Recovery: where the design offers undo or reversal, complete the action and verify the actual previous state is restored.
Keep these cases separate in the test record. A working undo button does not prove an aborted press did nothing. Likewise, a row reappearing does not prove its original source association, order and selection state were restored.
4. Use an isolated fixture before real data
Create a small synthetic evidence list with recognizable source identifiers and saved initial state. Test against that fixture rather than deleting real research records or repeatedly invoking a paid operation. A cancellation test should not create the harm it is meant to prevent.
Record the tested input method and environment. Mouse evidence is not a substitute for a physical touch interaction. A simulated test can check handler logic, while an actual device test can reveal whether the intended gesture remains practical in the rendered layout.
Also exercise the control's keyboard path separately. Changing pointer event handling should not accidentally remove ordinary keyboard activation or create duplicate submissions. Do not count that additional check as proof of every other accessibility requirement.
5. Hand off a precise cancellation record
- Control label, page and relevant task.
- Input method and initial fixture state.
- Press, movement and release sequence.
- Observed visual feedback and actual state change.
- Chosen cancellation, reversal or undo behavior.
- Retest result, including duplicate-action checks.
If an essential-use exception is proposed, document why completing on press is necessary for that specific function rather than treating speed or styling preference as sufficient justification.
A sound completion note reports the cases actually tested and any missing device coverage. The goal is a predictable opportunity to avoid or recover from the wrong action, with evidence that the underlying research state matches what the interface shows.