ACR with Hidden Reference (ACR-HR), defined in ITU-T P.910 section 8.6.2, is the bridge between the single- and double-stimulus worlds and the most common method in practice. You run an ordinary single-stimulus ACR test, but quietly slip the pristine source clips into the set as unlabelled items; the viewer rates them without knowing they are references and never realises they are doing a comparison. In analysis you subtract: for each viewer and processed video sequence the differential viewer score is DV = V(PVS) minus V(REF) plus 5, where the plus 5 shifts the result onto a 1-to-5 scale on which 5 means as good as the reference. Averaging the DVs gives the clip's DMOS, where higher means better. This recovers a degradation score while keeping ACR's throughput, and because it measures each viewer against their own reading of the source, it cancels personal harshness and content preference. The catch: P.910 warns it may give wider confidence intervals than ACR, and it only works if the hidden reference is genuinely excellent.
