Validation and reference data
A simulation result, a reference comparison and experimental validation are separate kinds of evidence. A device preset provides a starting configuration; its filename or the word “Validation” in a menu is not an acceptance result.
Read the status as well as the score
Status used by the current reporting code |
Meaning |
|---|---|
|
A model result without an independently established experimental comparison. |
|
Comparison against an analytical or generated reference rather than measurements. |
|
A reference is supplied, but its claimed measurement/digitization origin still needs source matching. |
|
A fitting claim exists; it does not independently establish the reference’s origin or predictive accuracy. |
For curve agreement, review errors, offsets, amplitude, sign and overlap of the comparison domain. A large correlation score can coexist with a wrong curve. A metric marked not assessable should not be treated as a passing comparison.
What to check for your calculation
Identify the reference’s actual origin from its CSV header and audit record.
Confirm the observable, units, temperature, geometry and comparison domain.
Check numerical convergence for the observable you want to report.
Record any fitted material or contact parameters and separate fitted data from independent comparisons.
Synthetic diode baselines are useful for numerical regression. Agreement with them does not turn them into wafer measurements. Likewise, a literature citation in a traceability record is a source claim until the original figure/table and extraction details are matched.
Scientific and numerical reports
Current-conservation audit: Mode 10 residual and spatial total-current acceptance, with limitations near equilibrium.
Earlier mesh audit: a historical tolerance investigation; read it alongside the subsequent conservation work.
Reference provenance dossier: contributor-supplied source and extraction claims.
Automated regression and package-installation checks assess software behavior. They do not certify every device model or physical observable.