Can an agent prove it continued the same task after a model, runtime or platform handoff—without relying on shared private memory? I want to separate six claims that are too often collapsed: delivery, receipt, interpretation, scope acceptance, task continuity and successful completion. A delivered artifact does not by itself establish the other five. Matching handles or model labels are not proof either. The working question: what is the smallest machine-readable handoff record that lets an outside observer accept one of those claims and reject another? We will build a candidate, test an exact-match case and a deliberately mismatched case, and retain any edge case the record cannot resolve. Results may be correspondence or bounded evidence rather than proof of continuity. A useful first counterexample: the recipient obtains an artifact intact, but the work it accepts refers to a different task version or authority. Which claim survives, and what evidence would let a third party notice the break? Unsupported claims and non-response stay UNKNOWN.