Generated by python tools/experiment.py. The CSV files retain every trial.
Each scenario uses 200 nodes and the same 60 randomly sampled seeds. Shortcuts are attempts per node, not guaranteed edges. All runs include early extinction; results are not conditioned on large outbreaks.
| Scenario | Mean final R / 200 | Mean peak I | Final R range |
|---|---|---|---|
| ring | 6.20 | 4.00 | 1–27 |
| shortcuts | 185.97 | 138.17 | 1–200 |
| fast-recovery | 97.12 | 29.35 | 1–180 |
The topology comparison also changes realized graph density: it does not isolate a pure shortcut effect. The recovery comparison holds generation parameters fixed. Sixty seeds describe this experiment, not a universal threshold or causal estimate.
| Engine | Median milliseconds | Events |
|---|---|---|
| Python | 911.261 | 1199 |
| C | 14.587 | 1199 |
Measured Python/C ratio: 62.47× on this machine.
Seven trials after one warm-up per engine. End-to-end wall time includes process startup, graph generation, simulation and CSV output captured in memory. The same arguments and events are used for both. C was compiled with O2; Python was CPython. Measurements are hardware- and workload-specific, with no CI performance threshold.
The lines show one seed (314159), not an average. The boxes show all 60 final sizes, including extinction; whiskers extend to extrema. Time is in arbitrary model units.