Behavioural UX research methodology, designed at Imperial

Panel composition

Verified UK adults across four neurodivergent profiles: ADHD, dyslexia, autism, and dyspraxia. Each tester completes identity verification, a mental-capacity quiz that confirms they understand the engagement, and a device-and-experience profile we use to match them to your audience. Testers are paid an effective rate of £24 per hour. We do not use synthetic personas, geographically arbitraged labour, or undisclosed crowdsourcing.

• ▸Identity verified via third-party provider before any work begins
• ▸Mental-capacity quiz is pass-or-fail, not advisory
• ▸ND status optionally self-verified during onboarding
• ▸Compensation transparent up front, paid through Stripe Connect

Recording stack

Three streams captured simultaneously per session, in the tester's own environment via a Chrome extension: behavioural data (clicks, scrolls, keystrokes, navigation), self-report (voice commentary), and physiological signal (facial reactions). The streams are correlated by an analysis layer to produce ranked findings. Testers approve their own session notes before anything reaches the client.

• ▸Behavioural events recorded with millisecond precision
• ▸Voice commentary opt-in, captured via Web Speech API in-browser
• ▸Facial reactions processed locally where possible, with explicit consent
• ▸Tester confirmation gate: nothing is shipped without the tester's approval

Severity rubric

Findings are ranked by behavioural impact, not just by tester opinion. The rubric is fixed and the same rubric is used in every engagement, so severity is comparable across studies and over time.

• ▸Critical: blocks the journey for ≥30% of the cohort, or causes abandonment in 2+ profiles
• ▸High: causes substantial dwell-time inflation or task-failure for 15–29% of the cohort
• ▸Medium: friction observed in 5–14% of the cohort, recoverable with effort
• ▸Low: <5%, single-profile, or surfaced via voice but not behavioural data

WCAG mapping

Every finding that touches an accessibility-relevant pattern is mapped to a WCAG 2.2 success criterion. We map only the criteria that cognitive accessibility research can legitimately observe. The criteria that depend on assistive-technology testing (screen readers, magnifiers, switch devices) are listed as out-of-scope in every report and require a separate provider.

• ▸In scope: 1.3.1, 1.4.3, 1.4.10, 2.4.6, 3.1.3, 3.1.5, 3.2.4, 3.3.1, 3.3.2, 3.3.3, 3.3.4, 4.1.3 (and any subset surfaced by tester behaviour)
• ▸Explicitly out of scope: 1.1.1 alt-text accuracy, 1.3.5/6, 2.1.x keyboard-only paths, 2.5.x pointer alternatives, 4.1.2 Name/Role/Value (AT-dependent)
• ▸Each finding includes the criterion, the evidence type, and the cohort split
• ▸Reports list the out-of-scope criteria so nothing is implied by omission

A/B rounds protocol

An engagement does not stop at the first report. Each journey is retested across rounds while your engineers ship fixes. The engagement signs off when one of three conditions is met: task-success across all profiles plateaus, the target you set on the kickoff call is met, or the contracted round budget is reached. Whichever comes first is documented in the close-out.

• ▸Round 1 establishes baseline; rounds 2–N retest specific fixes
• ▸Per-round delta calculated against baseline cohort, not just round-on-round
• ▸Plateau defined as: no profile gains more than 4 percentage points across two consecutive rounds
• ▸Sign-off documented in the remediation log with the round and the delta that closed each finding