See Matias' memo made for EBEX Memo from 2006 (.pdf), based on “Benchmark Parameters for CMB Polarization Experiments”

Imagine a final pointing error per pixel at the level of the map (individual sampling pointing error can be larger provided they are small enough that E and B are linear over the error range, and hence the error on E and B averages down as the pointing error averages down for repeated hits)

B-modes get created from E-modes from this pointing error, in a way completely analogous to lensing

With a first order Taylor development on Q and U as a function of pointing error (Q(theta + dtheta) = Q(theta) + dtheta dQ/dtheta), you can analytically calculate the spurious B-modes coming from a pointing error power spectrum for a given E-mode ps

To make the calculation, we assume the pointing error power spectrum is white with a cutoff scale \ell_s (i.e. no pointing errors on scales smaller than \ell_s, pointing ps dies down at \ell > \ell_s)

Note that previously for larger T/S, the tighest constraint on pointing came from not affecting lensing B-modes at \ell=1000.