I found an error in calc_parspace, in the 'extra sampling' code block. This section is run when the profiling step indicates a poor coverage of the sample: it should resample in the problem areas. Due to the error, it just duplicated the sample sets that were selected as starting points, so this step is useless. This error thus only influences the results for data sets where the initial sampling rounds lead to poor results, and when profiling is included: the additional step does not improve things. I don't expect this error to cause serious problems in optimisation or error propagation. The sample is just not as nice as it could/should have been, and the CIs will thus be less accurate. Only in cases where the sampling is really bad can there be larger discrepancies, and in those cases, it should be clear from the parameter-space plot that the sample should not be used for inference! So always check these plots carefully.
I have already repaired the error; the repaired version will be present in BYOM version 6.3. I expect to release that version in the week of 22 August 2022. Sorry for this. UPDATE: VERSION 6.3 RELEASED 22 Aug. 2022.
The good news is that the algorithm works better than I feared for five free parameters. The extra sampling after profiling really works very well (when it is implemented correctly).