Boxrec is just a guide, after all. No belts or ABC ranking points depend on them. Generally their rankings work I think, just not so much the further back you go. The ATG stuff is definitely wonky but then you have the extra problem of unrecorded fights and newspaper decisions and no contests and all that jazz. For the current and most recent ranks I think it's fine. Just not the older stuff.
The formulas that are used by boxrec for calculating the ratings have been tested on its database to have the highest predicting accuracy above other versions. There were other versions of the formula that were "tweaked" to get the all-time ratings look better near the top, but they had had worse predicting accuracy than the formula that is used now. By predicting accuracy I mean when two boxers A and B met, and prior to the fight A was ranked above B, and A won, that's good accuracy, but if there was a draw or a win for B, that's bad accuracy. So summing up all these "predictions" (for over a million bouts that have been entered into boxrec database), you get an overall predicting accuracy for this or that formula. After all, boxing rankings are used just for that - that a higher-ranked boxer will most probably defeat a lower-ranked one.