A structure we refined back in 2017, using phenix.real_space_refine on a cryoEM map @ 3.7A (version 1.11.1–2575) produces a clashscore of ~8 but the exact same refinement run with the latest phenix (version 1.17.1) generates a clashscore of ~22.  Our refinements with the latest version on other EM structures tend to generate pretty high clashscores too.

I think i've tracked this down to the nonbonded_weight parameter, which defaults to 1000 in the older version, and to 100 in the latest version.

At which version did the default change, and is there a reason for it?  I couldn't find it in the changelog.

Alan