What are the main limitations on the accuracy of total energies calculated with fleur?

The context of this question is that I am looking at substitutional dopant structures within a TM oxide. I can arrange the doped atoms within the oxide crystal in a number of different ways. I'm using a different code to relax the geometries of these doped crystals. I'm then looking at the total energies of three different dopant configurations with three codes, including fleur. Fleur's energies are exactly in the opposite order to those calculated with the two other codes (one is a PAW code, the other a full potential code). PBE throughout. Energy differences between configurations are of the order of 0.1 eV.

With fleur I'm using -profile moderate, which ups the G cutoffs. Is this a situation where e.g. the MT radii might be expected to strongly influence relative energies?

Ah yes! The change in the geometries between the different configurations drove inpgen to use different MT spheres. That's a trick for young players when comparing energies!

Problem solved. Looking at the three geometries with consistent settings give the same total energy ordering as other codes. Thanks for the pointer.