Loosely speaking, being able to identify a causal effect means that you have a valid statistical strategy that infers or estimates the causal effect from the data, whether you collect them from observational studies or randomized experiments. When you use the CAUSALTRT or CAUSALMED procedure to estimate various types of causal effects or when you use the PSMATCH procedure to conduct propensity score modeling for matching, weighting, or stratification, you essentially assume the identification of the causal effect in question.
Practically, the identification assumption in these procedures is about whether you have measured and included a valid set of covariates in the regression model of the outcome variable and/or in the propensity score model for the treatment variable. In many situations, researchers might simply assume that a set of background (pretreatment) characteristics is a valid set of covariates in the model(s).
However, from the perspective of causal graph theory, the inclusion of all presumed background characteristics might not be an efficient or even valid modeling strategy to estimate causal effects. The primary purpose of PROC CAUSALGRAPH is to find valid modeling strategies or test the validity of a proposed strategy. When researchers can qualitatively specify the data generating process (including unmeasured variables and/or latent constructs) that explains how the treatment and outcome variables are related in the data, PROC CAUSALGRAPH can output valid measured covariate sets for regression adjustment or propensity score modeling.
Therefore, when you can reasonably assume the qualitative structure of the underlying data generating process, it is recommended that you use PROC CAUSALGRAPH to study the identification of causal effects before you use other causal analysis procedures. For illustrations of such a practice, see the examples in Thompson, Lamm, and Yung (2020) and Example 2 in Lamm, Thompson, and Yung (2019).