Make noisy and spread simulations reproducible

Choose and record the seed

A fixed seed repeats spread- and noise-related random choices for a given execution configuration. Preserve the netlist and includes, exact executable/core identity, formulation, integrator, time steps, noise sample interval, output requests and precision as well.

Use a nonzero unsigned 64-bit integer. Zero requests automatic/time-based seeding; do not use zero as a fixed baseline. A fixed seed alone is not a promise of identical results across tool versions, platforms, changed circuits or numerical settings.

Set it in the GUI

Other displayed values remain defaults unless their Override checkbox is selected. For this supplied example, the netlist requests Voltage analysis; the pictured Phase default does not override it.

1. Open the saved source netlist and its Run settings. In Reproducibility, enable Set seed and enter a nonzero value such as 12345.
2. Select Override seed if this value must outrank a .OPTION SEED entry in the netlist or a control block. Without that override, a deck seed can win.
3. Save the source, preview the saved plan and inspect its effective seed before running. Record the selected case configuration with the result.

Save the complete example as seeded_noise.cir

* Fixed-seed example: explicit sampled noise and resistor spread
V1 N 0 NOISE(1n 0 0.5p)
R1 N 0 50 spread=0.05
.OPTION ANALYSIS=VOLTAGE INTEGRATION=BDF SEED=77
.TRAN 0.5p 20p 0 0.5p
.PRINT V(N) I(R1)
.END

Repeat it from the command line

The explicit CLI seed overrides the deck seed of 77. Use separate output names so the previous run remains available. Compare parsed numerical rows, not banners or timestamps.

josim-pro-cli --seed 12345 -P 17 -o first.csv seeded_noise.cir
josim-pro-cli --seed 12345 -P 17 -o second.csv seeded_noise.cir
josim-pro-cli --seed 12346 -P 17 -o changed.csv seeded_noise.cir

Check the result

The checked example has 41 finite rows over 0–20 ps, with V(N) in volts and I(R1) in amperes. Repeated seed 12345 runs produced exactly equal parsed samples; seed 12346 changed them. Omitting --seed matched an explicit CLI seed of 77 in this environment.

These are reproducibility observations for this runtime. They do not validate the physical noise spectrum, statistical distribution or cross-platform bitwise behaviour.

If repetition fails

1. Check the effective seed in the actual run record, including CLI/GUI override precedence and block-local options.
2. Compare complete input bytes, runtime identity, time coordinates, solver settings and output precision before comparing trace values.
3. For an invalid seed diagnostic, supply a nonnegative unsigned integer; use a nonzero value for a deterministic baseline.