A timing number is only useful if you can say exactly what produced it and reproduce it. This topic covers the habits that make that possible: a run manifest that explains any difference between two runs, a semantic diff of constraint files, a tolerance policy for floating-point differences, and a final checklist.
Lesson | After it you can... |
|---|---|
Reach a verdict on a difference from manifests and slacks. | |
Classify a constraint change by risk. | |
Set a tolerance for run-to-run differences. | |
Gate signoff on evidence. |
Nobody can check it, and any later difference cannot be attributed to a cause. Reproducibility is what makes comparison meaningful.