In the Building Sticks Task (BST), subjects are presented (on a computer screen) with three building sticks of different lengths and a goal stick. The objective is to create a new stick the same length as the goal using the three building sticks, by either adding or subtracting lengths. There are two methods for solving BST problems: overshoot - starting with a stick larger than the goal and subtracting down, or undershoot - starting with a stick smaller than the goal and building up.
This model simulates the experiment in Lovett & Anderson (1995) in which different conditions had one problem-solving method be successful either 60% of the time or 80% of the time. Results show that people increased their preference for the more successful method as the experiment progressed (global sensitivity) and that they preferred to use the strategy that had worked on the preceding trial (local sensitivity). An ACT-R model that incorporates decay when learning production success rates can reproduce these results.
(Note that the model runs through a fixed sequence of 90 problems, chosen based on the condition. This is not completely representative of the experiment in that participants in the same condition all received different problem sequences, generated on the fly based on the condition's probability of success--80% or 60%.)