Shadow Deployment (2D): Mirrored Traffic, Zero User Risk

Shadow deployment (dark launch) forks a share of live production traffic to a candidate model running in parallel, so its predictions can be logged and compared against the live model's answers without ever being shown to a real user. This 2D simulator draws the two lanes as a flat pipeline diagram: a steady stream of production requests flows client → model → user, while a sampled subset is mirrored down to a shadow model whose result is only ever scored — never served. Tune the request rate, shadow sampling percentage, the candidate's simulated disagreement rate and its compute overhead; a rolling agreement-rate chart and a latency/traffic-split bar panel update alongside the lane diagram in real time.