getting-started

To set up causal abstraction experiments for mechanistic interpretability, follow stepwise onboarding guidance covering causal tasks, analyses, Hydra configs, runner workflows, and artifact output locations.

Causalab is a framework for mechanistic interpretability experiments that maps codebase structure to workflow stages, distinguishing tasks, analyses, configs, and the runner where artifacts are saved.

Create and run causal tasks using Claude skills by invoking setup-task, plan-experiment, and run-experiment as an end-to-end loop to drive mechanistic interpretability analyses.

Understanding counterfactuals and interchange intervention is helpful, as the framework's mental model covers causal models, counterfactuals, and interchange intervention for mechanistic interpretability experiments.

Causalab writes artifacts and experiment outputs to specific locations determined by the codebase structure, guided by the runner workflow and Hydra configs during analyses.

Yes, you can use Hydra configs to manage mechanistic interpretability experiment setups, as the onboarding workflow explicitly covers configs and the runner for task execution.