Immediately after each session, I debriefed with our team to organise rich summaries of digital, paper notes, audio/video files, and surface preliminary codes. The recordings were transcribed via Otter.ai and manually edited; all data were collated, formatted for consistency, and imported to NVivo for systematic viewing and processing. Metrics were gathered in one Excel file.

First running an analysis of the metrics (time on task, errors, SUS), I then conducted thematic analysis over a week, coding user statements as well as observer descriptions. I started with our predefined codes of larger categories ('subject calibration context' and 'RGP') which facilitated theme identification while leaving room to explore data and challenge preconceived notions.

Next, I reviewed the extracts of coded data under devised themes to reform or refine them in terms of validity, coherence, clarity and relevancy—that is, whether each had been effectively derived data, appropriately grouped together, adequately descriptive of our understanding, and possessive of the explanatory power to answer our RQs.

Finally, I wanted to reduce my personal biases in interpreting the data. I involved colleagues across Product and UX to critique the themes with fresh eyes over one week, providing a main guideline: