I noticed a somewhat consistent discrepancy between measurements and characteristic inputs so I began by recreating the shape using those inputs, and overlaying it on top of the original design.
Subtracting the characterization-spec-generated curve from the design-intent curve showed a systematic error significant enough to be measured.
The findings led to a recharacterization of design values that was backed up with my verification script. Subsequent results were closer to intended values by 1-2 orders of magnitude.
gLike
Characterization of a Complex Optical Surface

Designed algorithm in MATLAB that compared characterization assumptions to actual design.

Analyzed multiple designs in various configurations.

Proved that it was impossible to measure a product to within required accuracy because present characterization parameters did not match the design.

Showed algorithm to be robust. Developed new characterization method to make sure measurement results behaved as expected.

Gil Zimmerman
Biomedical Engineer Rochester, NY