Configure and interpret an Optimum™ yield study

Define what a passing sample means

Use the complete resistor-study ZIP in a new writable ASCII directory and run the supplied verify.toml first. The current Optimum™ contract uses fixed-step PHASE/BDF and a JoSIM entitlement. The CLI commands below are one checked route; the GUI study window and API use the same native study owner but keep their own interaction qualification.

The fixture passes when the 1 mA ramp produces the specified voltage response within the configured tolerance. Its resistance acceptance interval is 40–60 ohms around a 45-ohm nominal. This deliberately simple model provides a probability reference; it is not a Josephson timing or fabrication-process yield demonstration.

Set the spread and run a finite sample population

1. Keep yield.toml's selected parameter as RVAL and its num_samples as 400. Check that the circuit really uses RVAL.
2. Use yield.spread = 0.30 for the supplied fractional three-sigma model. For nominal 45 ohms, this means a 13.5-ohm three-sigma spread and sigma = 4.5 ohms, not sigma = 13.5 ohms.
3. Run the commands below from the extracted directory. Record Passes, Fails, their sum, the reported percentage, seed, sample count, model/specification and actual runtime identity.
4. Use the roll-off CSV as a set of finite-sample estimates. The supplied rolloff.toml uses 200 samples at each of spread 0.25, 0.50, 0.75 and 1.00.

optimum-cli -r 1234 verify.toml
if ($LASTEXITCODE -ne 0) { throw "Nominal verification failed" }
optimum-cli -r 1234 yield.toml
if ($LASTEXITCODE -ne 0) { throw "Yield study failed" }
optimum-cli -r 1234 rolloff.toml
if ($LASTEXITCODE -ne 0) { throw "Roll-off study failed" }
Get-Content .\yield_rolloff.csv

Keep fractional, absolute and zero spread distinct

The sampler uses normal draws and clips negative sampled values to zero. An explicitly configured spread is not a hard parameter bound. A realistic yield claim needs a justified process distribution, dependencies/correlations and an appropriate acceptance rule; this one-resistor example supplies none of that calibration.

Report the estimate and its uncertainty

With seed 1234, the checked current Windows source build reports 356 passes and 44 failures out of 400: an observed yield of 89%. The independently calculated Gaussian population probability for this ideal fixture is about 86.63%. These values answer different questions: one is a finite draw, the other is a model probability.

A Wilson 95% interval calculated from this observed 356/400 count is approximately 85.55–91.70%. This is a knowledge-base post-processing calculation, not a confidence bound returned by Optimum™, and it does not include uncertainty in a real process model or specification.

The checked roll-off CSV gives 92%, 76%, 56% and 51.5% at the four configured spreads. Those are sample estimates with only 200 trials per point. Do not require an exactly smooth or monotonic curve from every finite draw, or interpret one measured percentage as guaranteed production yield.

Reproduce draws and locate the output

The current source path honours the explicit CLI seed. The native receipt test compares all 400 parameter assignments and verdicts, not only aggregate counts, and they are identical in repeated runs and at one/two execution threads. Reproducibility still requires the same inputs and standard-library distribution implementation; it is not a promise of bitwise equality across platforms.

If a fixed seed gives different results, compare the actual component identities and effective seed rather than the version label alone. Earlier installed review binaries behaved differently. Preserve the complete receipt and matched settings; the agent guide shows how to inspect them.

Run roll-off from the project directory when using a relative output path. Margin reports are written beside the circuit, while the checked roll-off CSV is anchored to the calling directory. A different shell working directory does not move the TOML-relative circuit/specification inputs, but can change the roll-off destination.

Complete resistor.cir

* Public one-resistor Optimum teaching circuit; RVAL is actually consumed
.PARAM RVAL=45
I1 0 N PWL(0 0 10p 0 20p 1m 100p 1m)
R1 N 0 RVAL
.TRAN 1p 100p
.PRINT V(N)
.END

Complete resistor.sp

time N
0 0
1e-11 0
2e-11 0.05
5e-11 0.05
1e-10 0.05

Complete verify.toml

mode = "verify"

[verify]
method = "spec_file"
circuit = "resistor.cir"
file = "resistor.sp"
amplitude_threshold = 0.20

Complete yield.toml

mode = "yield"

[verify]
method = "spec_file"
circuit = "resistor.cir"
file = "resistor.sp"
amplitude_threshold = 0.20

[parameters]
search_string = "^RVAL$"

[yield]
num_samples = 400
spread = 0.30

Complete rolloff.toml

mode = "yield"

[verify]
method = "spec_file"
circuit = "resistor.cir"
file = "resistor.sp"
amplitude_threshold = 0.20

[parameters]
search_string = "^RVAL$"

[yield]
num_samples = 200
roll_off = true
step_size = 0.25
output = "yield_rolloff.csv"