So, to be consistent with TN 32, the bias, precession, and nutation are defined as going from terrestrial to celestial; i.e.,

y_GCRF = [B][PR][NU][GST][PM]*y_ITRF

where, y_ITRF is a position vector whose components are written in the ITRF frame, and

y_GCRS is a position vector whose components are written in the GCRF frame.

Note for reference purposes, the equation below is how the astronomical almanac defines those quantities:

y_ITRF = [PM][GST][NU][PR][B]*y_GCRF

So, in effect the EO matrix calculator definitions of [PM],[GST],[NU],[PR], and [B] are transposed from what the astronomical almanac publishes.

Please note the difference.