The same pixels. Not always the same colour.

First ask what the profile describes. An embedded image profile gives pixel values their colour meaning. A working space defines the representation used for editing. A display profile characterises a particular device mode so that a system can prepare its input signal. These are not three interchangeable files with attractive names.

A colour-management module, or CMM, performs transforms using profile data. A profile need not be merely three coefficients. This lab deliberately uses idealised matrix-based RGB models: enough to explore roles, but not to interpret an arbitrary .icc file.

The ICC.1 architecture connects device descriptions through a shared Profile Connection Space. Ordinary ICC.1 profiles use XYZ or Lab referenced to D50. This bridge avoids needing a unique standalone profile for every source-and-destination pair.

D50 in the PCS does not require setting the physical display to D50. A profile can describe a display with a different white while expressing adapted data against the common reference. Both RGB systems in this lab use D65: the path applies Bradford adaptation to D50 and back. Intermediate XYZ values cannot be compared directly with the original D65 values without accounting for the reference change.

Assignment changes the description of existing numbers. Conversion recalculates values from a known source representation into a destination. When the colour fits both spaces and the path is consistent, conversion can preserve colorimetry while RGB changes. Colours outside the destination gamut or permitted range need an additional decision.

Choose Keep, Assign or Convert in the lab. Compare file values and the declared space before looking at the result. This separates cause from impression. Conversion into the bounded 0…1 file model explicitly exposes clipping; there is no hidden professional gamut-mapping algorithm.

A missing embedded profile does not prove that the image was created in sRGB. It may indicate a lost description. An application still needs an interpretation rule: a default, a working space or a user choice. Distinguish that rule from evidence about the source.

In the web context, CSS Color 4 specifies sRGB handling for untagged colours and images unless otherwise specified; CSS rgb() and hexadecimal notation also have an sRGB context. This does not validate an arbitrary photo after its profile is removed, define every editor’s behaviour or supply a universal video interpretation.

The source colour can be interpreted correctly while the wrong drive values are calculated because a destination profile is stale or belongs to another display. The lab separates the actual display from the selected profile intentionally: a profile file does not alter the panel’s physical primaries.

After changing gamut mode, brightness or calibration, check whether the profile still matches the state. A factory profile may be a useful initial description, but its existence does not verify your individual device after adjustment. With several displays, check association to the actual screen and behaviour when moving a window.

A consistent chain performs the required transform once at the appropriate location. Bypassing colour management may send values to a device without reconciling spaces. A duplicate transform mistakenly interprets one transform’s output as original source values and processes it again. That is not an accuracy improvement.

Find the first divergence: source pixels, interpretation, display drive or emitted colour. A neutral sample can miss a primary mismatch when white and transfer functions agree, so include coloured samples. The lab’s Once, Bypass and Twice modes change the calculation rather than just the label.

ICC supports different reproduction aims: perceptual, saturation, media-relative colorimetric and ICC-absolute colorimetric. An intent name expresses an aim, not identical pixel output from every profile and engine. Actual results depend on profile data, available tables and processing choices.

Media-relative colorimetric relates the medium white to the reference white while aiming to preserve in-gamut colorimetry. ICC-absolute colorimetric retains the difference between media whites within the defined normalisation; it is useful, for example, when soft-proofing a particular paper. Perceptual allows coordinated changes to colours and tonal relationships to support the overall image appearance. Saturation prioritises vividness and may trade hue accuracy for it; that aim can suit a chart but is not a universal photographic choice. Neither exact out-of-gamut behaviour nor a particular conversion result can be inferred from these four names alone: check the profile pair, engine and test samples.

Do not interpret absolute as a promise of exact physical screen luminance in nits: ICC has specific normalisation and viewing assumptions. Black handling and out-of-gamut colours need separate attention. The lab has no intent selector because its matrix model does not implement those profile tables. That is more honest than labelling four identical algorithms differently.

A file may be structurally valid while a particular application path supports only some forms or applies its own assumptions. Header/tag checks, transform round-trip assessment and embedded-profile handling tests answer different questions. Record the profile version and the limitations of the software path.

ICC provides v4 handling tests but explicitly distinguishes them from display calibration. ColorSync can inspect and compare profiles; repairing structure does not validate the measurements used to create them. Preserve the original profile and diagnostic report instead of replacing everything with a supposedly universal download.

Applications on one OS cannot all be declared identically colour managed. Some use system services; others perform part of the processing themselves, and output mode can change the interaction. Microsoft documents ICC-application behaviour under Advanced Color separately. “Windows” without a mode and application version is therefore too broad a description.

On macOS, ColorSync provides profile tools, but their presence does not describe every video output. On Linux, likewise record the application, graphics session, compositor and display mode rather than generalising from one viewer. This is a reproducibility checklist, not a claim that every system supports every feature.

A tagged still image and a video stream may communicate colour meaning differently. Video primaries, transfer characteristic, matrix and range need separate checks; a space name in an interface does not prove that all signalling is consistent. Do not assign your monitor profile to source video merely because two players differ.

When bringing graphics into video, establish the still’s source space, the project’s expected input and where conversion to the working representation occurs. At export, check values and description. A viewing LUT or device profile should not accidentally become part of a master intended for an unknown viewer’s display.

A useful test has known input values, a profile and an expected transform. Use a neutral ramp, moderately saturated colours and a separate out-of-gamut sample. Preserve the source and a converted copy. Confirm first that both applications read the same file rather than different versions or previews.

A screenshot is another file captured at a particular stage and viewed through another path; it does not measure emitted monitor light. A support report should include the source, path description, versions and control-sample observations. Two photographs of screens alone leave too many unknowns.

Agree on the representation stored in the delivery file, retained profile information and verification method. Keep a source master and separate delivery copies instead of repeatedly converting a single file for different screens. After handoff, compare a known control region through a defined path.

This experiment calculates sRGB and Display P3 matrices, their shared transfer function and Bradford adaptation through D50. It does not read ICC files, query your display profile or calibrate hardware. Swatches are bounded sRGB previews. Max.|ΔXYZ| is numerical difference with Yn = 1, not ΔE or a perceptual acceptability score.