Complex Problem Solving (CPS) tasks require students to engage into in-depth interaction with the environment.

We investigated how behavioral indicators captured in computer-generated log files related to CPS performance.

The relation between time on task and CPS performance followed an inverted u-shape.

Noninterfering observation was positively, high intervention frequency negatively related to CPS performance.

Relations decreased in size but remained stable after controlling for a multistep exploration strategy (VOTAT).