Input-referred (unrendered) would a bit more accurate
That's because raw doesn't accurately record the colorimetry of the scene.
I learned about the terms scene-referred and output-referred in a
post by Eric Chan. I also often see scene-referred
Yes, that's a commonly used term, doesn't necessarily mean there is no better one, and that the better one is not in use. Even the abbreviation RIMM, meaning "Reference Input Medium Metric", contains "input".
"Images in an unrendered image state are directly related to the colorimetry of a real or hypothetical original scene." - Image States and Standard Color Encodings, by Spaulding, Woolfe, Giorgianni, Kodak, 2000.
Since we can't restore the scene colorimetry from raw data, the relation is not a bijection. Since some of the scene colorimetry may be clipped, it's difficult to define the relation as scene-referred. In the presence of noise "scene-referred" becomes moot.
Terminology defined for film ("scene" = "negative", not the actual scene) is sometimes not very adequate for digital.
"Unrendered" is a simpler and a more understandable term, though, strictly speaking, it may also be incorrect, depending on the definition of "render" (do sensors render the scene? - some would say yes, as one of the meanings of "render" is "translate", and that a sensor certainly does).
--
http://www.libraw.org/