Where the target XYZ came from

A color-correction matrix is fitted against target values that state what each patch should produce. Those targets are not measured by the camera; they come from the chart’s spectral reflectances rendered under the capture illuminant. The reference therefore limits how the fitted matrix can be interpreted, which makes its provenance part of the result rather than a footnote.

What was used

The target XYZ came from a spectral reference associated with the same ColorChecker SG product as the photographed chart, rendered under an explicitly measured sphere spectrum. The illuminant is taken from the measurement sidecars retained with the session — not inferred from capture metadata or from the date.

That gives a physically specified 140-patch target: reflectance times illuminant, integrated against the CIE observer.

What was not established

Identity with the photographed chart unit was not established. The reference describes the chart product; no per-unit spectral measurement of the specific physical chart that was photographed was retained. Charts of the same product vary between units and change with age and handling.

The reference was verified against manufacturer nominal values across all 140 patches and differs from them by mean ΔE76 1.34.

That figure needs care. It is a ΔE76 measurement, under a different weighting convention from the CIEDE2000 used for the camera result, so it cannot be subtracted from that result or treated as an error budget for it. What it establishes is narrower and still useful: two plausible descriptions of the chart product disagree, so the reported camera result cannot be assigned to the camera alone. These data do not establish whether or how much of the reported difference came from reference mismatch.

What the comparison supports

With a compatible reference, the study can measure how well a linear model maps this capture’s RGB to a declared target, and whether its training score remains representative across patch partitions. Both are properties of the complete measured and modeled path, not of the camera in isolation.

What it cannot support is a per-unit chart calibration, or the conclusion that the residual is attributable to the camera alone.

What a per-unit measurement would add

Measuring the photographed chart directly on a well-characterized spectrophotometer, with calibration and measurement uncertainty retained, would replace the product-level substitution with a per-unit reference. It would make reference mismatch estimable rather than silently coupled to the residual. Remaining difference would still include model, capture, and measurement contributions.