Three channels. Not three “true colours”.

A display creates colour by adding light from three primaries. Zero in all channels means no contribution from those primaries, while increasing channels adds emitted light. This is fundamentally different from subtractive printing models in which dyes remove parts of the incoming spectrum.

Every RGB system defines the chromaticities of its red, green and blue primaries. Rec.709, Display P3 and Rec.2020 use different coordinates, so R=1, G=0, B=0 describes different reds in those spaces. The letter R names a channel in a system, not one physically fixed shade.

An RGB triplet needs a defined coordinate system and encoding. Context matters: CSS #FF0000 already means an sRGB red because the specification defines its space. Numbers in a video buffer or a poorly described file do not independently identify sRGB, camera RGB or another representation. Device and colour-management differences are a separate reason for mismatched display.

In linear-light RGB, doubling a channel value doubles that primary's contribution to physical light intensity. Many blending, blur, compositing and physically meaningful interpolation operations are correct in linear light. But storing images directly in linear form at limited bit depth is not always efficient.

A transfer function redistributes code values relative to light. Historically this was tied to CRT behaviour; today it also helps use code values efficiently and build a consistent camera → signal → display system. sRGB, BT.709-style camera encoding, gamma 2.4 display response, PQ, HLG and log curves are different things even though interfaces often group them under a Gamma menu.

Full-range 8-bit encoding has 256 code levels per channel, 0–255. Ten bits provide 1024 codes, 0–1023. Nominal 64–940 video levels describe 10-bit luma and corresponding video-range RGB; colour-difference components use a different nominal interval, commonly 64–960 with neutral at 512. Float can retain values outside 0–1. Bit depth, range and gamut describe different constraints.

If two RGB systems are defined through CIE coordinates and white points, linear RGB can be transformed through matrices, often using XYZ as an intermediate. If the white points differ, chromatic adaptation may be required. If the target gamut is smaller than the source, a matrix alone does not guarantee graceful out-of-gamut behaviour, which is where gamut mapping enters.

A negative RGB component after conversion usually means that the colour lies outside positive mixtures of the chosen primaries; it is not “negative light”. A value above 1 can exceed a reference level without indicating a corrupt file. Float preserves such values for further processing, but does not guarantee that every tool handles them correctly. Early clipping discards information that later transforms could otherwise use.

Alpha normally describes transparency or coverage. In premultiplied representation, RGB has already been multiplied by alpha; in straight/unpremultiplied representation it is stored independently. This matters greatly for compositing, but alpha does not expand the colour gamut and is not another primary.

In Resolve, the same pixel can successively exist as camera-space RGB, timeline/working RGB and output/display RGB. The question 'what are the RGB values here?' should therefore always be followed by 'which colour space and gamma is this node using at this point in the pipeline?'