The Collaborative International Dictionary of English v.0.48:

fudge factor \fudge factor\, n.
   a term or factor inserted into a calculation to compensate
   for anticipated errors, or to arbitrarily make the result
   conform to some desired conclusion.
   [PJC]

WordNet (r) 3.0 (2006):

fudge factor
    n 1: a quantity that is added or subtracted in order to increase
         the accuracy of a scientific measure [syn: correction,
         fudge factor]

The Jargon File (version 4.4.7, 29 Dec 2003):

fudge factor
 n.

    [common] A value or parameter that is varied in an ad hoc way to produce
    the desired result. The terms tolerance and slop are also used, though
    these usually indicate a one-sided leeway, such as a buffer that is made
    larger than necessary because one isn't sure exactly how large it needs to
    be, and it is better to waste a little space than to lose completely for
    not having enough. A fudge factor, on the other hand, can often be tweaked
    in more than one direction. A good example is the fuzz typically allowed in
    floating-point calculations: two numbers being compared for equality must
    be allowed to differ by a small amount; if that amount is too small, a
    computation may never terminate, while if it is too large, results will be
    needlessly inaccurate. Fudge factors are frequently adjusted incorrectly by
    programmers who don't fully understand their import. See also coefficient
    of X.


The Free On-line Dictionary of Computing (30 December 2018):

fudge factor

   A value or parameter that is varied in an ad hoc way to
   produce the desired result.  The terms "tolerance" and slop
   are also used, though these usually indicate a one-sided
   leeway, such as a buffer that is made larger than necessary
   because one isn't sure exactly how large it needs to be, and
   it is better to waste a little space than to lose completely
   for not having enough.  A fudge factor, on the other hand, can
   often be tweaked in more than one direction.  A good example
   is the "fuzz" typically allowed in floating-point
   calculations: two numbers being compared for equality must be
   allowed to differ by a small amount; if that amount is too
   small, a computation may never terminate, while if it is too
   large, results will be needlessly inaccurate.  Fudge factors
   are frequently adjusted incorrectly by programmers who don't
   fully understand their import.