These results compare simulated throughput with 16 internal reference values used to calibrate the model. They are not independent validation data. Original run logs and exact source rows have not yet been verified for each value. Calibration results as of 2026-09-08.
We configure the simulator with each reference envelope: cores, clock, RAM, disk, and network. This does not rerun a real service on the same hardware or reproduce the full original benchmark.
The harness offers 0.5×, 1×, and 4× the reference throughput, then compares completions at 4× load. A result within ±25% passes the internal calibration regression check.
Limits: We have no validated error bounds for p99 latency, traffic spikes, or production capacity. Fitting the calibration set does not establish accuracy in new environments. Check actual capacity with load tests on the target system.
Independent validation with original logs, full run settings, and a separate data set remains outstanding. For calibration questions, email us.
Click the confidence button in the simulator toolbar to see the same aggregate stats inline.
Open Simulator →