Off the top of my head, it seems self-evident that we can't assess the effectiveness of the filters without actual data. You may hear hundreds of complaints of 'false positives', but bots rarely complain, so those hundreds of false positives may be the collateral damage from blocking either a single bot or hundreds of millions of bots.

Although I'm not normally massively against using anecdotal data for hypothesis building, in this case, it seems as though without actual data, it'd be silly to comment further.