I have an extremely low resolution data set to 6.5 A.  The data were collected on a Pilatus detector, indexed and further processed in XDS, space group P321.  Pointless determines spacegroup to P3221 or P3121 and L-test shows no twinning. Phenix X-triage indicates twinning, but the twin fraction varies enormously depending on the method used: Britton analysis 0.337, H-test 0.260, ML 0.022. The UCLA twin server also thinks that data are twinned. How do I evaluate which information is right?