The case
E₁ → H₁ → H₂ → C, built into several manuscripts. Years later a replication undermines E₁. A group that kept its conclusions and not their dependencies “cannot immediately identify which conclusions need reconsideration.” A record that kept the chain flags H₁, H₂ and C, and “does not automatically retract them. They may have acquired independent support, or the original inferences may survive for different reasons.”
What it can show
A7: “it distinguishes dependency tracing from belief propagation. A dependency can tell us what needs reconsideration without determining what should ultimately be believed.” A12 put the same thing as review, not retraction: “Changes to epistemic grounds should propagate obligations to review, not automatically propagate changes of belief.”
What it cannot show
That a dependency graph alone settles what to revise (CASE-089), or that keeping it is cheaper than the alternatives (CLAIM-303).