My example results differ from numbers in an older manual

Make the comparison reproducible

1. Use an unmodified copy of the complete supplied example and record its file hashes.
2. Record the InductEx® and engine versions, extraction mode, units and all non-default mesh/settings changes.
3. Compare the same quantity and orientation, not a design value with an extracted value or a matrix entry with a different convention.
4. Check that both outputs belong to completed runs. Repeat the comparison with controlled mesh refinement.

Account for algorithm changes

Support history records results changing after mesh and edge-slicing changes between older releases. An old manual table is a versioned reference, not a universal tolerance for every newer build.

Historical advice to disable an algorithm for comparison is a diagnostic for that release. It is not a recommendation to disable newer accuracy features for production.

Report a discrepancy

Include the exact input set, settings, versions, quantity, observed difference and convergence evidence. A fresh reference output for each published example remains part of this knowledge base's review work.

Compare with the design

Aim for low reported relative error in the model fit. The owner has not specified a new numerical cutoff for SVD or GMRES residuals.

For a meaningful nonzero design inductance, use 100 × abs(L_extracted − L_design) / abs(L_design) to compare the values. The general design target is below 10% deviation; below 5% is preferred.

Design agreement, model-fit residuals and solver convergence answer different questions. Check the physical model and refinement behaviour as well; these percentages are not universal solver-error or regression tolerances.