credit to the Global Phasing team who along with Phil Jeffrey spotted and supplied the fix for this nasty bug which was introduced in CCP4 7.0 update 039 (May 2017).
I am sorry to say this is one of the occasions when CCP4s internal testing has fallen well sort, and I apologise for the problems this will have caused. It looks like amplitudes with a I/sigma-<I>/sigma in the range 10-20 will have the wrong values. this will have made structure solution significantly more difficult. The Global Phasing team have been investigating the effects further.
Thanks to Clemens Vonrhein who e-mailed on the 8th, a fix went out in update 057 yesterday.
To clarify some questions I received in follow-up to my e-mail yesterday, this effects truncate, not ctruncate as used by default by ccp4i and ccp4i2. ctruncate is uneffected.