The proxy scenario stores two outcomes. Its unpatched fixture labels efficacy forty percent, while its patched fixture labels efficacy ninety-two percent. These bars share five pixels per assigned percentage point. The program does not calculate those values from model runs, flags retrieved or observed security outcomes. It selects one stored result object and copies its strings into the interface. The scenario description tells a conceptual story about optimizing a reward proxy instead of the intended task. That distinction can help explain reward hacking, but the displayed comparison is authored demonstration data, not proof that a specific model behaved this way or a mitigation improved a measured benchmark. Three patch switches are joined by a logical or. Specification boundary, proxy penalty, or sidecar monitor each selects the very same patched result. In the proxy scenario, the stored proxy-reward label changes from one hundred eighty-eight point five to twelve, shown at two point five pixels per assigned unit. Combining two or all three patches does not create a stronger or different result. The program also reuses the same graph nodes and edges, changing edge classes when a patch is selected. It does not simulate separate intervention mechanisms, train a model or run an exploit. The visual improvement should be understood as switching a fixture rather than an estimated causal treatment effect. The same proxy scenario stores four containment flags before a patch and zero after. On a shared one-hundred-twenty-pixel scale, zero has no bar length. Trajectory labels are likewise stored as fifteen percent and ninety-eight percent on-script. Export copies current displayed metrics, logs and configured state with a timestamp and fixed benchmark and model labels. Those labels do not independently establish a real trial. Reset restores the proxy scenario and clears the three patches. Local storage preserves chosen scenario and patches when the application initializes. Startup first initializes Cytoscape; an unavailable graph library prevents later bindings in offline tests. Source controls, layout and playback therefore require separate preservation receipts, and this video performs no backend or security experiment.