Total indicated runout: A test

Thoughts other than a CMM is the wrong tool for TIR?

Ensure your cylinder on the bar is the same size as your "datum" feature on the part, and I mean lenght. The length of measured axis is going to matter a lot.

Ensure the cylinder you measure on the bar as your "feature" is the same length as the feature on the part.

Ensure that the "feature" cylinder is the same distance away from the "datum" feature as the spread is on the actual part. This matters a TON.

Iterate the "datum" feature, meaning, measure, align it, measure it again, align it again. The more out of cylindrical you think it is going to be, the more you iterate.

Do NOT rotate the head between measurements of the datum and the feature. Extreme precision CMM's do NOT come with articulating heads.
You have axis errors in the machine on X, Y and Z. Adding in two more for A and B is not going to be your friend.

Align the part as CLOSE to phsyically in a machine axis as possible to try and negate the error of one of the machine axes.

With all that, I have a part, .50" long "datum" bore. 4 inches away is the "feature" bore, also .50" long.

The error in the machine gives about .00085" range of error in measurement. Meaning if I run the same part 30 times, the results have a range of .00085 to each other.
An $85k roundness gage with articulating level table repeats at something like 9 millionths on the exact same part.
That is the correct tool to check TIR.
The CMM is simply not.

Now, if your datum is 4 inches long and the feature is 0 inches away (like a counterbore) and only 1 inch long, you'll get MUCH better results than I got on that ONE configuration.
Good enough to state you have certainty for .0002 tolerance? Probably not.

You don't have a hyperaccurate CMM with a double isolated slab under it, I know because you have a 2.5° articulating head. I was going to test on a friend's shops machine that is hyper-accurate, but they had a crash and didn't have a new probe in time.
The customer changed the print, so I don't have to inspect that anymore, so I never went over to give it a go and see if it solved the problem.

I don't think it would, not for the tolerance you are playing with. But I could be wrong as I've never been given a machine that expensive to play with. :"(

Good luck.

I have no choice I have to articulate. Features are often on different sides of the part and are unaccessible without articulation.

If you are going to articulate, absolutely, positively autocalibrate tip after each rotation. Repeatability can be significant, especially on a worn wrist.

Also: what everyone else was saying, but with exception that CMMs aren't THAT terrible on runout when using a rotary table (but size goes out the window). This doesn't sound like a situation that lets you use one, though.

hmmmmm autocalibrate after EACH rotation? for both angles?

can you recommend me a rotary table?

Yes, after every change to either A or B angle. As far as brands, I'd go through Hexagon, so you know it's supported by PC-DMIS and your controller.
If it's the ideal situation where you can scan with rotary table in motion, drag so the side of the ball is being used, rather than the pole, and "pull" away from the probe. If nothing else, you can eliminate all need for angle changes, usually.

The rotary table is simulating what the roundness gage is doing, so that is a fantastic recommendation, though if he's stuck articulating the head, it might not work.

Still a fantastic suggestion.

Would a STAR probe at 0,0 with two tips work better than articulating?

That would solve one issue

ran that test also.
diameter sizes were consistent with articulation.
runout actually got slightly worse readings but repeated with each other.

no discernable improvement from articulation

When brilliant six figure paid "engineers" tell me to check runout & tol is that tight I say use bench center & indicator!