Matt Thompson

Is there no shorter path to passing this through the Python layer?

Thanks for the report! I can reproduce this and I agree it's a bug. I can't offer a solution or quick workaround at the moment (barring an OpenEye license). This...

A number of things have changed in the past almost-three years: * The OpenFF Toolkit is capable of passing residue information onto downstream formats, provided the input data contains residues...

Is this completed after #1372? I don't know the scope here.

I'd advocate for not including this in our object models (`Molecule`, `Topology`) today - which does not prevent its implementation in subclasses of those models.

Our current object models unfortunately don't leave much room for this to be implemented. * `Molecule` attempts to be close to a cheminformatics representation of a molecule, which does not...

> actual charge differences [might be] negligible From your tinkering you have a sense of what order of magnitude these differences tend to be?

FTR we added some basic energy tests in https://github.com/openforcefield/openff-toolkit/commit/14c963fae0bc2e12d328070eb726a7848f87cf25 as part of the saga resolved in #808. This happened before #722; it's possible but not guaranteed that these would have...

It looks like that should have been caught between 0.8.3. and 0.8.4? Or, put more precisely, it looks like there were differences in the energies generated by each version of...

I don't have a great feel for the lower bound of how much run-to-run/version-to-version variance is generally acceptable (mean errors of 0.1 are probably too much but tightening things to...