Combining heavy-tailed keystroke dynamics with eye tracking reveals temporal flexibility in human problem solving
Abstract
Human problem solving unfolds through complex temporal dynamics that reflect the coordination of perception, cognition, and action. The fine structure of this process is still not very well understood. We use high resolution keystroke logging combined with synchronized eye tracking to examine the micro-temporal structure of human problem solving during written solution construction. Looking at 131,718 inter-keystroke intervals, we find heavy-tailed timing patterns spanning multiple orders of magnitude within the observed range. A three-component Gaussian mixture summarizes overlapping short, intermediate, and long pause ranges. Triangulation with gaze data reveals a systematic shift of attention toward external representations with increasing pause duration. Associations between tail exponents and performance differ across tasks. A larger across-task range of tail exponents (αrange) is positively associated with overall performance after adjusting for prior knowledge and measured process covariates (β= 0.351, pHC3 = 0.0165), accounting for an additional 11.3 percentage points of explained variance in this sample. These findings are consistent with the hypothesis that successful reasoning involves adaptive reconfiguration of temporal organization across task demands. The heavy-tailed timing dynamics represent a signature of task dependent cognitive organisation and highlight the potential of keystroke data as an unobtrusive and scalable window into the process of problem solving.
Related
- Tracking Coupled Granular Temperature and Entropy Dynamics in Granular Materials via Dielectric Spectroscopy
- Quantum state smoothing for tracking optical quantum dynamics and recovering purity and squeezing
- Temporal complexity hierarchies in solvable quantum many-body dynamics
- Fundamental limits to identifying node and tie memory in temporal networks: marginal artefacts and spreading dynamics
- Temporal entanglement transition in chaotic quantum many-body dynamics
BibTeX
@misc{dinc2026combining,
title = {{Combining heavy-tailed keystroke dynamics with eye tracking reveals temporal flexibility in human problem solving}},
author = {Yavuz Dinc and Fabian Kieser and Peter Wulff and Johannes Zierenberg and Niklas Stausberg and Karina E. Avila and Jochen Kuhn and Stefan Küchemann},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.40376},
eprint = {2610.40376},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.40376},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Combining heavy-tailed keystroke dynamics with eye tracking reveals temporal flexibility in human problem solving
AU - Yavuz Dinc
AU - Fabian Kieser
AU - Peter Wulff
AU - Johannes Zierenberg
AU - Niklas Stausberg
AU - Karina E. Avila
AU - Jochen Kuhn
AU - Stefan Küchemann
PY - 2026
DA - 2026/10/29
PB - arXiv
DO - 10.48550/arxiv.2610.40376
UR - https://arxiv.org/abs/2610.40376
AB - Human problem solving unfolds through complex temporal dynamics that reflect the coordination of perception, cognition, and action. The fine structure of this process is still not very well understood. We use high resolution keystroke logging combined with synchronized eye tracking to examine the micro-temporal structure of human problem solving during written solution construction. Looking at 131,718 inter-keystroke intervals, we find heavy-tailed timing patterns spanning multiple orders of magnitude within the observed range. A three-component Gaussian mixture summarizes overlapping short, intermediate, and long pause ranges. Triangulation with gaze data reveals a systematic shift of attention toward external representations with increasing pause duration. Associations between tail exponents and performance differ across tasks. A larger across-task range of tail exponents (α_range) is positively associated with overall performance after adjusting for prior knowledge and measured process covariates (β= 0.351, p_HC3 = 0.0165), accounting for an additional 11.3 percentage points of explained variance in this sample. These findings are consistent with the hypothesis that successful reasoning involves adaptive reconfiguration of temporal organization across task demands. The heavy-tailed timing dynamics represent a signature of task dependent cognitive organisation and highlight the potential of keystroke data as an unobtrusive and scalable window into the process of problem solving.
ER -