I have a question about when to actually stop refinement with Refmac5. What is a properly carried out refinement with good statistics?

All I could find was this advice: 

'zBOND and zANGL are the Z-scores of the deviations of bond lengths and angles from ideal values: at high resolution these should approach 1.0 while at low resolution, where structures are tightly restrained to ideal values, they should approach 0.0.'

What values of zBOND and zANGL are expected for a well-built and correctly refined protein structure at a resolution of 2.5 angstroms?

Should the geometric weights in Refmac5 be adjusted/varied so as to produce statistics of rmsBOND 2.0, rmsANGL 2.0 and rmsCHIRAL 0.2? Are these values appropriate for this resolution?

Some practical advice and your experience would very much be appreciated.

Ray Brown