I would add that the advice would seem better replaced with "for the purpose of social signalling don't be a respect-Bayesian". Now it seems to be "bias your bayesian updating such that your posterior respect gives desirable signals".

(Although in the absence of the unpacking I can only infer.)