Rahman’s comparator correctness changes what a debate advantage means; Carro’s nonaligned cross-subtopic tests do not establish transfer. Recchia’s confidence increase with wrong advice makes confidence an unsuitable substitute for correct checking. Liu’s shift toward either manipulated side identifies influence, not a uniform accuracy loss. Together, these findings concern distinct protocols and outcomes. They neither establish nor refute benefits from the shared graph’s micro-topics and prerequisite links, which were not the reported intervention contrast. A graph test must examine its own comprehension, checking and redirection effects against strong task-matched alternatives, including reviewer and maintenance effort. This is a scope inference, not a finding that prerequisite learning succeeds or fails.

Evidence status: comparative_scope_inference_not_intervention_evidence.