In 8.6.2 Scaling and transformation process, equation 8-271 is used when transform_skip_rotation_enabled_flag is equal to 1 and nTbS is equal to 4:

r[ x ][ y ] = ( TransCoeffLevel[ nTbS − x −1 ][ nTbS − y − 1 ] << tsShift )

This corresponds to the code at line 1861 of TComTrQuant.cpp:

rpcResidual[(k * uiStride) + j] = Pel(pcCoeff[coefficientIndex]);

but in the code, no shift is applied.

I believe equation 8-271 should change to:

r[ x ][ y ] = TransCoeffLevel[ nTbS − x −1 ][ nTbS − y − 1 ]