One white. Not one spectrum.

A spectral distribution describes radiation across wavelengths. RGB and XYZ are already a reduction to three numbers under specified rules. The inverse problem is therefore ambiguous: three coordinates do not reconstruct a unique original spectrum. A broad distribution and several narrow bands can produce identical coordinates for a chosen observer.

A digital image normally does not contain a spectrum for each pixel. RGB identifies colour in a stated space, not a lamp recipe or surface reflectance. Converting between RGB spaces translates coordinates; it cannot restore the discarded spectral dimension. Even a colourimetrically accurate display preview does not turn the screen into the original light source.

For an ideal diffuse, opaque, nonfluorescent surface, the reflected spectrum is proportional to illuminant E(λ) times reflectance R(λ). Three standard-observer functions weight that result before integration. Our discrete model replaces the integral with a 5 nm sum and normalizes a perfect diffuser to Y = 1 under each selected illuminant.

This compares spectral shape at a consistent photopic white level. It does not measure lux, watts or absolute object luminance. Geometry, directional reflection, specular highlights, transmission and fluorescence are outside this simplified model. Fluorescence can shift radiation between wavelengths, so a single E × R multiplication is insufficient.

The CIE 1931 2° functions define a standard colourimetric system. They are not the sensitivity curves of a particular camera’s three channels, nor simply three individual cone responses. The 2° designation concerns viewing conditions, not a lens, image resolution or the field of view of an entire shot.

A camera also reduces a spectral signal to a few channels, but with its own spectral sensitivities, optics, filters and processing. Two spectra matching for the standard observer need not match in raw camera responses. Profiles and matrices can reduce typical discrepancies; they do not reconstruct every possible spectrum from three values. A practical test needs the actual cameras and representative scene materials.

Metameric stimuli have different spectra yet the same tristimulus coordinates under specified conditions. For reflecting materials, both illuminant and observer must be named: a match under one illuminant does not promise a match under another. Changing the conditions changes the weighting of spectral regions.

The laboratory pair is constructed mathematically. Opposite perturbations are added to two reflectances, with zero contribution to all three discrete D65 integrals. The reflectances stay between zero and one, but are not measurements of a commercial paint or fabric. Their exact match belongs to this sampling grid and model; it does not prove a match for every person, camera or integration method.

White balance can align a neutral region, but it cannot retrospectively control spectral energy in the scene. If one lamp supplies little energy where two materials differ, the camera receives a different set of responses. Global channel gains may correct the neutral card while leaving errors in saturated objects.

In the TCS09 experiment, the same reflectance is integrated under D65 and the selected illuminant. Optional Bradford adaptation affects only the screen preview: it does not alter source XYZ or reconstruct the reflected spectrum. This separates a viewing-coordinate transform from a physical lighting change. Real scenes additionally introduce mixed illumination, specular reflection and camera limitations.

Correlated colour temperature summarizes a chromaticity near the Planckian locus. It does not list spectral peaks and gaps. Even identical white chromaticities allow many spectral distributions; a similar Kelvin label is therefore insufficient to promise identical material rendering.

D65 is a defined daylight illuminant, not any fixture labelled “6500 K.” Illuminant A likewise has a specified distribution. Our slider mixes two normalized spectra by proportion, not by Kelvin interpolation, and does not emulate a fixture’s CCT control. A real LED fixture may change channel mixture, spectrum and output as that control moves. Test its mode, dimming and modifiers in the actual working configuration.

In the traditional CIE method, general Ra averages the first eight test-colour results. It is not an average of every special index, nor a saturated-red index. R9 relates to TCS09 and is reported separately. “CRI 95” and “R9 95” therefore describe different parts of the assessment.

An index is the result of a specified method with a reference, observer and test-colour comparison. A high average does not mean zero error for every object and does not cover every production concern: stability, flicker, dimming, geometry or camera compatibility. Our laboratory shows TCS09 reflectance but does not implement the complete method. Neither its patch nor the metamer pair’s ΔE may be labelled CRI.

TLCI-2012 evaluates lighting for television practice through a camera-and-reproduction model. That differs from directly comparing spectra or evaluating through a standard visual observer. The model addresses problems relevant to television production, but it is not your specific sensor, RAW decoder and creative transform.

EBU recommendations discuss TLCI in the context of lighting equipment and its use. Comparing scores requires understanding the method and configuration. For a production using multiple cameras, supplement a model-based result with a shared chart, skin and representative-material shoot. The scale remains useful guidance, not a replacement for testing the complete production chain.

The Academy’s Spectral Similarity Index evaluates a spectral distribution against a chosen reference. The comparison itself is not tied to one particular camera. However, it still answers a question about spectral similarity, not universal aesthetic quality or guaranteed matching of every material.

A score without its reference is incomplete. A fixture may be close to one distribution and far from another; matching two fixtures to each other is another task again. Inconsistent reports cannot be compared as if their targets were identical. Nor can SSI be derived from CCT, white RGB or an eight-bit photograph of a spectrum. This chapter explains the metric’s purpose; the browser model does not calculate SSI.

Check what the instrument measured: spectral irradiance at a point, directional radiance or light reflected by a sample. These are not interchangeable merely because their plots share a wavelength axis. A comparison needs units, range, sampling interval, spectral resolution, geometry and normalization. An overly coarse grid can miss narrow structure in a source.

Record warm-up, power, mode, dimming, distance, angle, optics and modifiers. Without those details, the result is harder to reproduce on another day. In our CSV, 5 nm is the teaching table’s sampling interval, not a real spectrometer’s specified resolution. The 380–780 nm truncation and quadrature are explicit; values can differ from a complete standard calculation.

Begin with an aligned technical chain: a single decode, known input, matched exposure and controlled output. Record a neutral, a chart and real scene objects—especially saturated fabrics, paints, cosmetics and potentially fluorescent materials. Check global balance separately from object-dependent differences.

Then compare sources or cameras while changing one factor at a time. Preserve versions without baked creative transforms and note what remains after basic alignment. If the issue originates in the physical light, changing the source, mode or fixture combination may be more robust than correcting every object separately. A numerical metric narrows the search; the controlled scene tests the actual result.

The experiment establishes that different reflectances can match in one discrete colourimetric calculation and diverge when the illuminant changes. Separately, it shows the behaviour of published TCS09 data. Those data types are not conflated: observer functions, illuminants and the sample are tabulated; the metameric pair is constructed algorithmically.

Patches follow a fixed sRGB preview path, optionally with Bradford adaptation. Display clipping does not change XYZ or ΔE*ab. The lab contains no real camera sensitivities, TLCI camera model, complete CRI or SSI calculation, individual-vision simulation or monitor-match guarantee. To apply the lesson in production, preserve the conditions and repeat the check with your equipment.