Zoom with the mouse wheel or the buttons, drag to move, double-click to see the whole city again. Both panels move together.
Reference demand · scale 1.000 · seed 408
16126 / 16126 arrived by 09:00.
GPT-OSS-scaled demand · scale 0.823 · seed 408
12685 / 13250 arrived by 09:00, 553 still on the network, 12 never inserted.
moving, above 30 km/hslow, 2 to 30 km/hstopped, below 2 km/hbus
All demand variants
| Demand variant | scale | Cohort cars per seed | Time loss, completed trips (s) | Elapsed, all scheduled cars (s) | Unfinished, all seeds | Completion, minimum |
|---|---|---|---|---|---|---|
| Reference demand | 1.000 | 16126-16126 | 113.9 | 398.3 | 0 | 100.0% |
| Constant-only fallback | 0.632 | 10174-10231 | 102.4 | 386.2 | 0 | 100.0% |
| Gemma-scaled demand | 0.687 | 11069-11116 | 103.9 | 387.9 | 0 | 100.0% |
| Llama-scaled demand | 1.164 | 18769-18826 | 123.2 | 419.0 | 157 | 99.2% |
| GPT-OSS-scaled demand | 0.823 | 13250-13308 | 107.1 | 450.6 | 565 | 95.7% |
Means over the seeds listed per variant. Time loss counts completed cohort trips only; elapsed time counts every scheduled car to its arrival or to the 09:00 horizon, so it does not hide unfinished trips.
Where these runs come from
BeST Berlin SUMO Traffic scenario, Mitte route subset (Schrab et al., CC BY 4.0); SUMO 1.22.0
15% of vehicles report their position every 30 s (SUMO floating-car output, geographic coordinates). time-to-teleport -1, ignore-junction-blocker 20, step 1 s, 06:30 to 09:00.
These are sensitivity runs on technical scenarios, not calibrated city forecasts. Every number on these pages comes from the recorded SUMO output; the site adds nothing.
- Reference demand:
temp/simulation/runs/viz-20260912/reference/fcd.xml· SHA-256 c54a7c0a7e113a7b… - GPT-OSS-scaled demand:
temp/simulation/runs/viz-20260912/gptoss/fcd.xml· SHA-256 e8f609b1a1c9031b…