De-identified aggregate eye-tracking data for visual attention and cognitive workload analysis in a screen-based augmented-reality cabinet customisation interface

Stimuli were displayed on a desktop monitor at an approximate viewing distance of 60 cm in a controlled usability laboratory with stable illuminance of 300 lux.The deposited files contain processed, aggregate summary values rather than raw participant-level records. Raw eye-tracking recordings, gaze-coordinate streams, screen-interaction videos, time-stamped behavioural logs, file paths, record identifiers, and participant-level rows are not included because these materials contain participant-level behavioural information and were collected under consent conditions that do not permit public sharing.

The deposited data therefore support public inspection of the manuscript's reported tables and figure-level values, but they are not intended to reconstruct individual gaze trajectories or reproduce record-level preprocessing.The file "dynamic task summary.csv" contains 20 rows and 10 columns. Each row reports one aggregate dynamic-phase metric for one task and measure. The columns identify the phase, task, interaction stage, measure, unit, sample size, mean, standard deviation, and approximate 95% confidence interval bounds.

Measures include absolute mean pupil diameter (MPD, mm), baseline-corrected MPD (mm), blink rate (blinks/s), and average saccade distance (ASD, reported in both pixels and degrees of visual angle). The four dynamic task stages are T1 spatial mapping, T2 AI-assisted parametric adjustment, T3 layout error-proofing, and T4 material selection. Confidence intervals were calculated from the reported means and standard deviations as mean +/- 1.96

• SD / sqrt(n).The file "static representative AOI summary.csv" contains 4 rows and 8 columns. Each row reports aggregate static-phase visual-attention metrics for one representative interface screen and target area of interest. The columns identify the phase, task, screen, target AOI, time to first fixation (TFF, seconds), fixation-duration proportion (FDP, percent), fixation-count proportion (FCP, percent), and sample size. The four representative screens correspond to the login action button, AI-recommended dimension area, red collision-warning border, and selected wood-veneer card.The file "group summary T2-T4.csv" contains 8 rows and 10 columns. It reports aggregate group-level summaries for two exploratory analyses described in the manuscript: T2 mean pupil diameter by user-profile group and T4 average saccade distance by user-profile group. The four user-profile groups are function-driven, spatial-renovation, technology-oriented, and specialised-needs users, with n = 5 per group. The columns provide task, metric, unit, group, sample size, mean, standard deviation, degree-based ASD values where applicable, and the manuscript-reported statistical note. The file "data dictionary.csv" contains 25 rows and 2 columns defining the variables used across the deposited CSV files.The temporal scope of the data is limited to the single laboratory data-collection period for this study, and the geographical scope is Nanjing, China, where the experiment was conducted at Nanjing Forestry University. The data have no geospatial resolution because they describe laboratory eye-tracking and interface-interaction summaries rather than observations distributed across geographic locations. The temporal resolution of the raw eye-tracking device was 250 Hz, but the deposited data are aggregate summaries at the task, screen, measure, and group levels.Missing values in the deposited files are intentional and limited to fields that are not applicable. For example, "mean_deg" and "sd_deg" are blank for T2 MPD rows in "group summary T2-T4.csv" because pupil diameter is measured in millimetres and has no degree-of-visual-angle equivalent. No participant-level missingness indicators are provided because participant-level data are not part of this public dataset. Known limitations are that the files contain summary statistics derived from the manuscript rather than raw data, and confidence intervals are approximate values calculated from reported means, standard deviations, and sample sizes. The dataset is suitable for checking the reported aggregate values and understanding the data structure behind the manuscript, but it should not be used for reanalysis requiring individual observations, gaze samples, video review, or alternative preprocessing assumptions.All tabular files are stored in comma-separated values (CSV) format and can be opened with common spreadsheet or statistical software, including Microsoft Excel, R, Python, SPSS, JASP, Jamovi, or similar tools. The accompanying Word document provides a human-readable summary of the deposited data package.