The raw data seems to be a pure measure of light intensity per wavelength, so there is considerable variation depending on ambient lighting, sample distance, sample irregularity, temperature, SCiO internals etc.
It is possible in the SDK to download these actual data, but no stats are generally available on deviation of either raw or processed scans. The models do give a reliability index.
I see that CP have just added a new processing method, but I have not tried it out yet.
best regards, Roger.