Hi Louis, Greg,
There are two main reasons why I still recommend some post-refinement of models from ISOLDE before deposition:
1. ISOLDE currently doesn't attempt any B-factor refinement at all, so without some post-refinement the B-factors in your output model will be whatever they were in the model(s) you used as your starting point.
2. The AMBER force field used by ISOLDE is great in many ways, but it doesn't quite replicate the fine details of equilibrium bond lengths and angles seen in ultra-high resolution structures - due to various systematic differences your model will always be flagged for high RMSDs in these metrics.
All that being said, yes: I've often observed the same behaviour as you. That's why I included some options (under the "isolde write" command) to write out input scripts designed to limit the refinement as much as possible to only the above (tightening bond lengths/angles and refining B-factors, with all more "aggressive" options disabled). See for example
https://tristanic.github.io/isolde/static/isolde/doc/commands/isolde.html#isolde-write-phenixrsrinput for the cryo-EM case (to get the equivalent documentation within ChimeraX, do "usage isolde write" then click the applicable link in the log).
Best,
Tristan