Luo and colleagues argue that scientific papers are inadequate containers for machine replication because decisive knowledge is fragmented across prose, code, configurations and tacit implementation choices. Their Executable Knowledge Graph externalises these heterogeneous elements into a paper-centred, hierarchical representation that links conceptual relations to reusable code. The graph is not merely descriptive; it is executable, allowing AI agents to retrieve and assemble implementation-level knowledge at multiple granularities. Reported gains on PaperBench support the claim that reproducibility requires a representation designed for action rather than textual similarity alone. The iconic idea is that scientific knowledge becomes robust when its latent procedures are made traversable and testable. This bridges knowledge graphs with research infrastructure and shifts the archive from storage toward reenactment. A publication no longer terminates in a document but extends into a verifiable operational environment where claims, methods and technical dependencies can be recomposed under scrutiny.