Hi Sam,

> I am refining a structure with Phenix.refine and encounter an issue
> which I appreciate your comments. The structure is in P1 and refined
> to 36A-3A. Rw/Rf = 23.9%/29.1%, RMS bonds 0.002, RMS angles 0.544,
> Ramachandran favoured 97.5%. The statistics is acceptable to me

This look fine to me too!

> but two numbers raised my concern. First is the clashscore of 7.7,
> which is below minimum of other structures shown by Polygon.

The lower, the better. So this is good! If it's lower than the others
(as shown by Polygon) then you are doing better on this one!

> Second is an average B of 37, with a minimum of 9.2. When I checked
> again in Coot (Average Temp fact. analysis), none of the residues or
> water look abnormal. I refined both individual and group B-factors and
> applied 'Optimize X-ray/adp weight'.

This looks fine too.

> So my questions are:
> (1) if a clashscore too high means a lot of bad contacts, how do you
> comprehend a clashscore too low? How can I modify my strategy to get
> around?

It can't be "too low". Zero means you don't have any single steric
clash, which is good.

> (2) is there an acceptable low value of B? (I sort of went through a
> similar thread in phenixbb appeared 9 years ago but seems the
> discussion did not come to a conclusion). Any advice I should reliax
> or tighten the adp restraints?

Check out recent thread on ccp4bb: "acceptable difference between
Average B-factor and Wilson B".
Bottom line, I think what you have is just fine, unless you meaningfully
and clearly define "low".

Pavel