Realtime Control E2E Benchmark¶
Date: 2026-05-22
Command: PYTHONDONTWRITEBYTECODE=1 python scripts/benchmark_realtime_control_e2e.py --repeats 15 --steps 64
Claim Boundary¶
No-QPU software control-loop benchmark only. Results exclude provider queue, network transport, runtime session setup, dynamic-circuit hardware latency, readout latency, and QPU execution latency.
Environment¶
{
"platform": "Linux-6.17.0-23-generic-x86_64-with-glibc2.39",
"processor": "x86_64",
"python": "3.14.3",
"rust_feedback_policy_available": true
}
Rows¶
| n | repeats_successful | p95 tick ms | p99 tick ms | max tick ms | rust policy path |
|---|---|---|---|---|---|
| 2 | 15 | 4.267569 | 5.157374 | 18.401380 | true |
| 3 | 15 | 4.994387 | 6.156117 | 9.210675 | true |
| 4 | 12 | 6.919108 | 8.895908 | 13.749520 | true |
| 2 | 15 | 0.000128 | 0.000174 | 0.001220 | true |
| 3 | 15 | 0.000152 | 0.000184 | 0.000530 | true |
| 4 | 15 | 0.000190 | 0.000230 | 0.000977 | true |
Reproducibility¶
This artefact is deterministic at the configuration level (fixed seeds and fixed rows), but wall-time values depend on host load, CPU governor, and thermal state. Re-run the command on an isolated benchmark host before publication-grade speed claims.