While the model is by no means a complete model of all the aspects of behavior in the SPL task, the model is quite complex, reflecting the complexity of the task. There are 48 chunk types and 116 productions~. Much of the model is concerned with interactions with the environment. Most importantly, the model distinguishes between internal and external memory, and productions do not directly access external memory.
The top level goal of the task is a discovery goal. It specifies that the goal of the task is discover whether a given hypothesis is correct. To implement the full goal of the task---discover which of two hypotheses is correct---we push two discovery goals onto the stack, one for each hypothesis. The discovery goal leads to repeated pushing of experiment goals to perform individual experiments and evaluate-evidence goals to examine these experiments have produced results which satisfy the discovery goal. The terminating conditions that evaluate-evidence seeks is that some consequence of the hypothesis under test has been directly proven true or false.
Each hypothesis for a variable consists of an effect direction (greater-than, less-than, or equal-to), and a belief level (true and maybe). Each experiment can either 1) create a new hypothesis with initial truth level of maybe (if no prior hypothesis for the given variable existed), 2) move the belief level from maybe to true (if the same effect direction was observed), 3) move the belief level from true to maybe (if a different effect direction was observed), or 4) change the effect direction holding the belief level at maybe (if the belief level was maybe and a different effect direction was observed).
Each experiment goal leads to the design, running, and encoding of one experiment. There are three different subtypes of experiment goal which can be thought of as different experimental paradigms: hypothesis-testing, explore-system, and test-factor. The hypothesis-testing and explore-system subtypes correspond to the two general approaches that subjects took towards the task: designing experiments relevant to the hypotheses versus simply investigating the effects of the six factors. The three types of goals lead to similar experiment design behaviors although there are some critical differences. The subtype used most often by the expert models is the hypothesis-testing goal.
A hypothesis-testing goal produces the overall structure of designing an experiment to test a hypothesis. A hypothesis-testing goal pushes two primary goals: to vary the hypothesis factors and to fix the remaining variables. The goal to vary hypothesis factors selects whether to vary one or two factors relating to the hypothesis and which factors to vary. The goal to fix the remaining variables then iterates over the variables visually and selects the remaining unselected variables.
After the experiment is designed and run, the resulting table must be encoded. The encode-table goal initializes and directs the search over the table, setting ENCODE-DIMENSION subgoals for each of the table dimensions. The slots of encode-table are method, start-cell, and current-dimension, and searched-dim. All of these slots start out empty and are filled in that order.
In searching the table, the model encodes effects that it finds and tests their generalities, and looks for problems in the design of the experiment if its expectations are violated. If such experiment problems are found, this will influence the design of the subsequent experiment.