Hello,
I cannot either get to the same values with the JLCPCB calculator even how much I wanted to. I tried to ask help or material datasheet but got just cited to their all datasheets page. The first answer was pretty much that they select material RANDOM. from all FR4 they happen to have. That must not be the case as the values seem to differ a lot between materials. If this is the case then it does not even pay so much back to make a test boards to figure out the perfect width for a trace as it can next be something else.
But there is something. I started to suspect when I noticed a big difference in the performance between different batches.
Did you select the Impedance Control option and choose a specific stackup? If you order without choosing a stackup then it wouldn't be surprising to have differences between boards. On the other hand, if you're choosing a specific stackup and getting very different results each time then that's concerning.
At this price point I don't expect extremely good process control, but I would at least expect to receive what I ordered.
As for impedance calculations: Their stackup and impedance documentation is confusing. They list the "nominal thickness" of the prepreg, presumably as measured prior to lamination. They include this cryptic note on their impedance page at
https://jlcpcb.com/impedance0.1mm (3.94 mil) is nominal thickness of 3313 prepreg. Use 3.5 mil as the thickness of 3313 prepreg when the controlled impedance tracks are on top/bottom, use 5.5 mil(2116 prepreg) or 8.5mil(7628 prepreg) when tracks are inside.
Their own impedance calculator shows 3.5 mil for the substrate height in the diagram. The diagram also shows the soldermask height at 0.8 mil with a 3.8 Er, which is a significant contributor at these thicknesses. If your calculator doesn't account for soldermask then you'll get different values.
If I use the 3.5mil thickness and add the soldermask, I get numbers somewhat close to their calculator. If I use the 3.94mil nominal thickness and no soldermask then the calculators disagree significantly.
I'm not convinced their stackup diagrams are accurate, though. Would be interesting to get some actual tests performed on these boards.