The saved resilient mesh contains five nodes and five directed edges. A planner writes shared context and delegates to a generator, whose output connects to a verifier. The verifier has edges to memory and a fallback synthesizer. The linear chain has three agent nodes and two edges; the shared star has four nodes and three edges, and the distributed task graph four nodes and four edges. Both bars share eighty pixels per node. Recovery percentages, latency strings, trapped counts and resilience grades are preset constants. Changing topology loads those values without measuring correctness, actual agent execution or recovery rate. A graph can represent a verification route, but drawing that edge does not implement an AST verifier or a fallback decision. Run flow collects the current edge array and highlights one edge every five hundred milliseconds in array order. Five mesh edges take two and a half seconds to reach the fifth pulse; two chain edges take one second. Both bars use two hundred pixels per second of pulse progression. A further interval clears the highlight and logs completion. These timings belong to the animation, not preset consensus latency. The function does not traverse dependencies, execute tools or move genuine memory data. Repeated clicks create separate intervals without a shared cancellation handle. A topology switch during animation can leave callbacks referencing removed edges. Successful cycle text therefore means a local animation reached its end, not that a workflow succeeded. Fault injection randomly selects a node whose type is agent, marks it failed, then marks it recovering after eight hundred milliseconds and active twelve hundred milliseconds later. Both bars share two hundred fifty pixels per second. Despite the comment about checking a verifier, the code never requires a verifier node or traverses the fallback edge. The same timers can restore a linear-chain agent. No output is validated, no actual consensus is computed and no memory corruption is detected. Reset loads the resilient mesh without updating the preset dropdown, so export can label a newly reset graph with the previously selected preset. JSON reads displayed metrics and current graph elements; it cannot certify model recovery. The offline Cytoscape dependency is blocked, so native preservation checks may retain inert controls and initialization errors. No external model or backend request is made.