Standard cognitive assessments to detect dementia are administered infrequently and often long after symptoms are clear to even family members. With the advent of new drugs and therapies to delay the onset of dementia, it is important to both detect signs as early as possible and to provide monitoring of cognitive changes. This paper describes unobtrusive methods for monitoring user interactions with a computer that serve as a basis for algorithms to measure cognitive performance. We adapted a standard computer game currently enjoyed by elders at risk for dementia in order to monitor natural performance on a task that involved significant strategic planning throughout the game. This enabled us to collect cognitive performance data on individuals at frequent intervals. We monitored move-by-move appropriateness as distance to solution, and additionally modeled user thought processes using between-move data from the mouse device. We then used our resulting dynamic user model both to adapt the game difficulty and to detect meaningful individual cognitive trends.