Display Polarity and Cognitive Performance: A Within-Subjects Experiment Comparing Dark and Light Mode Web Interfaces

We report a within-subjects experiment (N = 80; Mₐᵍᵉ = 22.8 yr, SD = 5.8) in which university students completed eight cognitively diverse tasks under dark and light mode web interfaces in counterbalanced order. Task performance was captured with millisecond precision via CogBench, a purpose-built browser platform; perceived workload was assessed with the NASA Task Load Index (α = .708); and visual comfort was rated on four independent 7-point indicators. Eleven dependent variables were tested with paired-samples t-tests, Wilcoxon signed-rank validation, and Benjamini–Hochberg FDR and Bonferroni corrections. Light mode produced faster response times (Mᴅ = 6,273 ms vs. Mʟ = 5,108 ms; d = +0.522, p < .001) and completion times (d = +0.449, p < .001), both surviving all correction thresholds; dark mode yielded lower eye strain (3.18 vs. 3.78; d = −0.293, p = .011), corroborated non-parametrically (p = .016). Accuracy, composite workload, and remaining comfort subscales showed no significant differences; post-hoc power was below 12 % for these measures, rendering the null results inconclusive. Exploratory analyses revealed a preference–performance alignment rate of 38.8 % and a task-complexity dose-response in which the light-mode advantage grew from negligible on simple tasks to 13 percentage points on 2-back working memory. Findings support user-controlled theme selection as the most defensible default policy.