This is one of the curses of digital measurement: Every one expects accuracy to be equal to the number of digits represented (typically to 1mV resolution): This is unreasonable.
Look at a typical LiFePO4 cell, the limiting voltage is given to be 3.65V +/- 50mV. This requires an accuracy of no more than 13.7%.
BMS Chips typically have a 12 bit a/d, which after averaging is increased to 14 bits.the reference is typically 5V. Working with the 12 bits, this gives a resolution of 1.22mV (NOT ACCURACY), Accuracy given in the data sheet is +/-0.45mV typ for a 3.6V measurement. So each cell measurement is going to be +/- 1.67mV, or 30 times more accurate than required.
Calibration of BMS chips therefore requires a significantly more accurate source than any typically available to a standard user. Also, no user means of BMS calibration is possible.
Some cheap BMS and shunt measurement circuits may not come close to this accuracy, as pointed out above.
For an 8 bit a/d, measuring 12v (FSD probably 20V) the resolution is 78mV, with a probable accuracy of ~2%, or +/- 240mV.
So the real answer is not to expect every voltage measurement in a system to agree to the last place of decimal.