11 Causal Learning with Infinitesimal Creativity

CLIC: From Closure Defects to Admitted Causal Structure

Many causal-discovery procedures select a graph or equivalence class from a declared model family. CLIC asks a stronger question: what should a discovery system do when intervention geometry supplies evidence that the declared causal language itself is inadequate? The answer developed here is to treat nonclosure as a typed diagnostic, actively seek counter-witness interventions, and admit a structural extension only when uncertainty and unsupported alternatives have been made explicit.

CLIC abbreviates Causal Learning with Infinitesimal Creativity. It is the DIAL-SKFM member of the algorithmic family specified in Chapter 10. Its experiments form the causal counterpart to RELIC: RELIC studies structural accommodation in sequential decisions, whereas CLIC studies structural accommodation in intervention-driven causal inference.

In one sentence, CLIC asks whether a reproducible failure of intervention fields to fit the current causal declaration can be localized, challenged by a discriminating intervention, and resolved by a finite causal proposal that survives evidence withheld from its construction. Each clause is a separate gate; nonclosure alone answers none of them.