Thanks for doing the comparison!
There are a few INCAR params different from what we are using, noticeably EDIFF, ENCUT, and LREAL (see The EDIFF in Elastic calculation - #2 by mjwen). Given that, we probably need to look into the calculation details of mp-1016904 to see whether it is converged (and other potential problems) to figure out why there is a large discrepancy. I will check our calculations. We’ve found that LREAL=AUTO leads to convergence problems for small systems. If possible, it would be great you can recalculate it with LREAL=False and then we’ll see what happens.
We do not have new calculations for mp-2967 yet; will let you know once we got it.