One chain. No accidental transitions.

Begin with three descriptions rather than a preset: what the decoded source numbers mean, the representation in which grading operates, and the intended output. Name the owner of each transition: project settings, a node, a LUT or another application. Colour management then becomes a chain that can be checked rather than trust in an automatic mode.

The name Rec.709 alone does not describe a file completely. Primaries, transfer function, code range, component representation and processing history all matter. Viewing conditions also matter to the creative judgement. Management reconciles these definitions; it does not decide contrast, mood or shot matching for you.

Ordinary DaVinci YRGB and DaVinci YRGB Color Managed differ mainly in how transforms are organised. A manual workflow places transforms explicitly; a managed workflow performs some of them outside the visible clip nodes. No CST in the graph does not mean no transform, and unmanaged does not mean media decoding and display handling cease to exist.

Do not expect arbitrary RCM and CST presets to produce identical pixels. Tone rendering, gamut limits, adaptation and tool behaviour may differ. The laboratory matches only because both organisations deliberately use identical functions. Its switch changes ownership, not mathematical definitions.

Input describes data after decoding, not necessarily the label on the camera body. A RAW decoder may already have chosen an output gamut and curve; a proxy may have a baked LUT; graphics may have been exported with a different profile. Establish the actual state from recording settings, documentation and file history first.

Assigning Input Color Space does not rewrite the original media file. A managed project uses that assignment to calculate a transform, so the viewed image can change immediately. Automatic identification is useful when metadata is trustworthy. Missing or suspicious tags require investigation, not a search through menus for an attractive result.

The working representation defines coordinates and numerical encoding presented to tools. A wide gamut and floating point help avoid early restrictions, but do not guarantee every effect accepts negative components or arbitrary values above one. Check the requirements of the particular operation.

DWG and Intermediate are separate components: primaries and white point, and a relative-light encoding function. Intermediate values are not directly nits or linear exposure. An operation requiring linear light needs a matching conversion to the appropriate linear space and back, without accidentally adding display rendering.

Output preparation relates a creative result to its intended reproduction. It may include tone and gamut rendering, adaptation and encoding. ACES explicitly separates rendering from Display Encoding. This is a useful distinction, not a promise that all software and modes have identical stages.

The rule of one output transform means avoiding an accidental second rendering of the scene for the same target. It does not prohibit a deliberate display-to-display conversion, calibration path or separate SDR version. Each further operation needs a defined input, output and purpose.

Automatic operation reduces the decisions that must be made manually. It helps with consistent sources, but cannot replace checking at least one example of every material type. Pay particular attention to proxies, graphics, older masters and intermediates created by transcoding.

Before grading, record the selected system, identified inputs and output goal. Do not treat menu choices copied from another software version as a universal standard. If automation hides a required behaviour, move to explicit configuration deliberately, comparing before and after without simultaneously changing the creative grade.

Explicit project control is useful when input, working and output definitions must be reproducible. Record those selections separately from the tone/gamut-rendering options actually available in your mode. For HDR also record the agreed white and monitoring conditions.

Do not change working space after grading without testing: identical control values in another mathematical environment may produce a different result. Duplicate the project or create a controlled version first, then compare test shots and effects. A longer list of settings is not, by itself, better colour management.

A CST explicitly connects known representations. In a manual chain it may serve input or output; inside a managed chain it may prepare data for a tool expecting a different input. Consequently, never use a CST in RCM is too broad. The risk is a repeated or incorrectly described transition, not the node itself.

A conversion pair around an effect must return data to the expected state. If the forward transition included clipping, tone rendering or other irreversible processing, an inverse matrix cannot restore the lost information. Inspect tone/gamut-mapping settings as well as space names.

Finished graphics generally have an intended appearance, while camera material describes a different image state. That does not mean graphics must always bypass colour management. Interpret their profile correctly and place them coherently either in the shared scene or in the output portion of the project.

Check straight versus premultiplied alpha at transparent edges. Alpha represents coverage or opacity, not another sRGB-encoded colour channel. Nonlinear processing of premultiplied RGB without respecting that relationship may alter edges. Disabling all colour management hides the cause rather than making the composite reproducible.

A shared grade separates reusable creative work from decisions belonging to a particular master. Each output needs its own target, rendering, encoding, range, format and viewing conditions. Output-specific trims are valid, but their SDR or HDR ownership should be clear.

Do not derive a new HDR master from clipped SDR merely because the file can be tagged PQ. Return to the sources and common grade before losses when preparing another rendition. Compare renders when changing transform systems or versions: the same target name does not guarantee the same picture.

Begin with a project copy and a small representative set: a neutral patch, saturated object, highlights, skin and graphics. Choose who owns the technical transitions. Identify actual inputs, define the working environment and configure one understandable output. Confirm this technical foundation can be explained without creative correction.

Then add balancing, shot matching and the look. Establish the expected representation of each third-party effect. Export a short sequence and re-import it through a separate verification path as already rendered material. This is a diagnostic order rather than a universal list of setting values.

Audit every transform, including less visible ones: RAW decoding, project input, group and clip nodes, LUTs, output and monitoring. For each operation record the state before and after, its version, and whether it has been baked into an intermediate. Check purpose as well as count.

A visual symptom does not prove its cause. Excess contrast may come from repeated rendering, a range mismatch, the wrong source interpretation or a deliberate grade. Isolate one variable. Measurements are useful only when their location in the signal path is known.