[I have no idea why this didn't post yesterday as it was meant to.]
The "55 percent rule" - that unobtrusive studies tend to show a 55% success rate of librarians answering reference queries fully and accurately - seems to have been written about most in the mid/late-1980s so my quick-and-dirty Google Scholar search isn't bringing up much handy full-text to link to. (Some, but closed access and gigantic files that freeze my computer for a minute so needn't be inflicted on anyone else. Who decided to scan black-and-white Library Journal articles in as colour???) Anyway, my quick-and-dirty impression of the literature thus surveyed is that the number was so shocking that it prompted vast flurries of a) studies to try and replicate/refute the results, and b) studies to say that users don't care about full and accurate answers anyway.
I have a different response, inspired by today's date, which is: If the reference service we're providing is so incomplete and inaccurate, why not save our time/salaries and just hand users a Magic 8-Ball instead?
[NB: This post is not guaranteed to be more than 55% indicative of the actual definition of the rule nor the state of the literature, but it is at least 55% flippant. What I actually think is that we should be developing clever chatbots to staff our virtual reference service. Or at least 55% of it.]