Here's a reference: https://physics.nist.gov/cgi-bin/cuu/Value?e
>A conservative estimate of the height difference between clock and experiment contributes a gravitational red shift uncertainty of 7.7 × 10−17.
So, it appears the 4e-15 is not an actual measured value, but a synthetic value based on estimation and measurement.
It's like when you take a caliper and measure the width of an object, you typically neglect the error due to thermal expansion of the caliper since it's hotter/colder than when it was marked with gradations. If you're really anal about uncertainty (like these guys), you'd estimate the effect of this expansion and say "a conservative estimate says this contributes an uncertainty of 0.0003 mm".
They're all listed in table I and the 4.2 * 10^-15 just comes from adding them in quadrature.
An important piece of background info for non-atomic-physicists is that the Hydrogen 1S-2S transition frequency is probably the most accurately measured physical property in the world, thanks to the pioneering work by T. Hansch. Back in 2011 they reported an uncertainty of 4e-12, https://arxiv.org/abs/1107.3101.