Your CM appears to be using HyperLynx or some other tool for a quick study. The fabricators I use give much more detail that when plugged back in to Allegro match much closer. Here some items to consider:
1) dielectric constant depends on glass structure, resin:glass ratio. Neither of you are showing variances in Er when changing thicknesses of dielectric. Not going to be accurate.
2) copper weight? What thickness are you using? What frequency is the impedance measured at? Using what instrument? At what transmission line length? Using a Polar impedance analyzer? VNA? It will vary the results.
3) Pressed thickness does not appear to be accounted for. When I see four mils for example it seems to be a starting thickness. Not a finished thickness.
four) Be sure to get the *exact* glass structure if you want to move this stackup around to other vendors. If you don't call out the exact construction and copper your results will vary more than the 3 ohms you're worried about.
There are more gremlins to attack in this stackup but these are some biggies. Get with the CM's fabricator and get *their* stackup and you can probably dial this in closer.