The C option assumes that the reference histogram contains the
theoretically expected values with no (or negligible) errors.
Examples might be a flat distribution hand-inserted as the expectation
for a phi distribution, or
a long data run to be compared with shorter data runs.