Continuous games of timing represent an ecologically relevant case of decision-making under uncertainty with many applications to tactical security scenarios. We present two experiments collecting data using the two-player, 20-second FlipIt game, in which a player has to react to an estimated, externally controlled quantity in light of a continuous payoff distribution. The quantity is represented once as time interval, and once as distance in time and space. Risk-taking propensity is an individual parameter obtained via a standardized survey instrument. We find that risk propensity affects participants' decisions in early and late rounds of the game and show that risk-seeking individuals fail to play well initially in the difficult, time-interval estimation task, but that they can be trained to make rational choices. Participants' biases are gaussian-distributed. They do not reflect the distribution of payoffs or expected utility. We present an ACT-R model using established error distributions for temporal and visual perception and instance-based learning.