Inspect both branches
A reference and candidate may pass through different colour handling before a comparison. Record how each is decoded, interpreted and transformed. If one branch changes range or transfer treatment silently, the observed difference cannot be assigned only to the encoder.
Keep the approved source interpretation and desired viewing domain explicit. A matching codec name or frame dimension does not establish that the pictures reach the comparison in an equivalent representation.
Use a small owned sample with neutral and bright material to inspect the preparation path. Preserve any intentional colour difference as part of the delivery task rather than hiding it behind a compression-only label.
Name the scope of the result
If the question concerns the entire delivered SDR derivative, include the planned conversion and say so. If it concerns encoding alone, hold the relevant interpretation and preparation constant. These are legitimate but different experiments.
The VMAF documentation makes model assumptions relevant to interpreting results; it does not make every pair automatically fair. Review the actual metric requirements and record any common preprocessing.
VideoColorSpace by Mozilla Contributors is available under CC-BY-SA 2.5 or later. The branch-audit method here is original guidance. Report what was held constant and what changed, then inspect meaningful passages. A well-described limited result is more useful than a universal ranking based on an unexplained colour mismatch.
Questions answered
Do matching dimensions establish a fair colour comparison?
No. Interpretation and processing in both branches matter.
Can a complete delivery-path comparison include tone mapping?
Yes, if that is the stated task and the result is not presented as compression alone.
Sources & further reading
Official references for the subjects discussed. Service terms and device software can change.