Thanks for the handy calculator for surface roughness!

Something that the page doesn't mention, is that surface roughness not only increases attenuation, but also slows down wave propagation. This is also a contributing factor to the discrepancy between material DK and design DK (along with material anisotropy).

A good way to model both effects is with a gradient model, which gradually transitions from your bulk material to the surrounding free space using the RMS surface roughness Rq. It works well, but it assumes a fairly predictable statistical distribution of roughness, which frequent irregularities will invalidate.

The brute force, ugly way to get around this is to prototype a whole panel of microstrip filters, each tweaking the nominal geometry slightly (slight elongations or shrink factors applied to resonators, coupling gaps adjusted for over-etch or under-etch slightly) and to measure them all. The best performer's artwork would be filed with config management... and then we'd repeat the process each time the board house changed their process.

(Of course, something else that could mess around with your phase delay is ferromagnetic PCB track plating, like nickel used in ENIG and ENEPIG, which may also cause dispersion and odd resonances... but that's a whole different discussion.)

http://blog.rogerscorp.com/2011/11/03/what-is-design-dk/
https://www.rogerscorp.cn/documents/1669/acm/articles/Effect-of-conductor-profile-on-the-insertion-loss-phase-constant-and-dispersion-in-thin-high-frequency-transmission-lines.pdf
https://docplayer.org/69656651-Modeling-surface-roughness-in-cst-microwave-studio.html