when to use GCM?

[Waschenbacher]

I recently came across the GCM package and am quite impressed by its potential. I would like to adopt it for my project, but I have several conceptual questions before proceeding. I would greatly appreciate any insights you could share.

• How does GCM differ from other frameworks, such as the 4-stage process for (conditional) causal effect quantification via backdoor adjustments? My understanding is that GCM aims to recover the data-generating process (the causal mechanisms of all nodes), which allows for answering various causal influence questions, such as what-if scenarios, arrow strength, and intrinsic causal effects. In contrast, other frameworks seem limited to addressing (conditional) causal effect questions.

• When should we use GCM, and when not? It appears that the GCM framework is more powerful. However, I wonder if we should always opt for GCM or if it has limitations. To my knowledge, recovering the data-generation process can be more challenging and data-intensive than simply estimating causal effects. So I wonder what is the recommendation when it comes to GCM vs others.