Yeah I did, the level is calibrated as best as I could get it. The bubble is only a smidge off.
You are correct in what you said earlier, if the lathe is not level, it's fine. You're looking for relative changes.
If your level is not calibrated in any way, it doesn't matter "much". If you can get the bubble close to centered, (and NEVER rotate the level...), you'll be guaranteed to be in the "accurate range" of the bubble. (Although the one you're using might be beyond that issue).
I think I know what I'm doing wrong here though. I'm shimming the headstock side and it seems that you should avoid doing that at all costs unless you're trying to get rid of nod in the headstock. You should always shim the tailstock side first.
I'll agree with that, but it's not because it's not equally effective at the stage you're at. There are several mechanical reasons why it's "better" not to shim the headstock end. If you're in a position to start over from the tailstock end, you should do it, but there's still more to sort out here.
I think you have a combination of "V" ways and flat ways. That's gonna make it tricky to get a dead nutz level reading anyhow. The flat ways are going to show wear, where the very tip top of the V ways are not going to show wear. That'll be down on the sides.
My suggestion is to A, use your level (even out of level) to establish a flat plane across the ways as close as you can. Level is cool, but if you didn't want to calibrate the level, it doesn't matter. What you NEED is to find a flat plane. Level is only a convenience. If you've got two V ways (I think you've got one V for the saddle, and another for the tailstock? If I'm off, you can still use the principal here.... Put the level (on a parallel if needed) across the tip top of the two V ways, where it's unworn. Do as you're doing, and make the reading the same up and down the lathe bed. That's a baseline. A jumping off point. Then you know where the lathe bed is to start. That is important, as you DO want to use the "twist" measurement to get it to a minum, but if there's any wear, you're actually, in the end, going to probably "dial out" some wear....
Once that's done, put the level on the carriage. At that point, you can use the level (calibrated or not) to track -deviations- from the original "plane". That would be (typically) closest to the headstock, but it won't be consistent across the whole bed. That will need to be accounted for on a judgement basis, to get the most repeatable results over the longest possible distance.
After that's as good as it can (reasonably) be, your final adjustment (shims and otherwise) will be based not on "level", but on actual cuts. And this can be a process, but setting your stuff up "close to" being in plane will make that process of identifying errors go a lot better.
For your lathe, which has a foot on each end- You are going to "dial in" enough twist to make it cut right, that the level you have is more than you'll ever need. Folks calling out a half thou resolution, they're not wrong. If you were SETTING the lathe with the level, you'd want at least that. Your lathe is not heavy enough to "sag", its not long enough to have six, eight, ten (or more) sets of feet along it's length, that ALL need to be set to the same height and same level. Accurate levels are almost required to set these up, because "absolute level" is a datum, a reference, a concret thing that you can pick up and carry from one place to the next, and the next, and the next... You've got a foot at each end. You don't need to carry "level" to any place except across the ways. "Bubble not moving" is plenty level enough to get that baseline adjustment done, the secondary "accounting for wear" adjustment done (IF that's an issue- I have no idea), and move on to taking cuts, doing math, tearing your hair out and sorting out a crooked headstock from a shifted tailstock, while not blaming either of those possibilities for what might just be deflection in the test bar... That's where you want to put your headaches. The leveling is really just as simple as you have described. "planar" in your case is every bit as good as level. I only would add the extra step of "baselining" the machine from unworn areas to know an actual "start point" so that you know where you're at to begin with, and help with keeping track of overall changes.