The written tree generator adds three children per node until the requested depth. Depth two therefore has one root, three first-level nodes, and nine second-level nodes, totaling thirteen. Depth three adds twenty seven leaves, totaling forty. At seven pixels per generated node the bars span ninety one and two hundred eighty. This is source arithmetic, not observed agent spawning. The offline page cannot load Cytoscape and replaces its whole tool with a failure message before initialization. It demonstrates neither an interactive graph nor a live delegation cycle. For the forty-node example, the initial telemetry formula excludes the root and models thirty nine calls. It assumes independent per-call reliability of point nine six five. Without retry, raising that to the thirty ninth power gives about twenty four point nine two percent compound reliability. With retry it squares the point zero three five failure probability per call, then compounds, giving about ninety five point three three percent. At three pixels per modeled percentage point the bars span about seventy four point seven six and two hundred eighty six. Those independence and retry assumptions are authored, not measured guarantees, and cannot prove retries are necessary for every workflow. The written initial token estimator uses eight hundred times one point eight to the level for each delegate, plus one thousand for the root. For the three-level, three-child tree that sums to approximately one hundred fifty four thousand six hundred nineteen tokens. The run loop instead charges rounded one thousand times one point five to the level, giving one hundred sixteen thousand eight hundred seventy five for a complete traversal. At point zero zero two pixels per modeled token the bars span about three hundred nine point two four and two hundred thirty three point seven five. The formulas differ before any failure. Offline runtime failure prevents either display from being verified; neither is real provider usage.