Hi Ian,
Your critical anomalous stats are:
1. CC-1/2 ... half-dataset correlation for a SAD dataset. Should be above about 0.3 in resolution range of interest. You can calculate this with phenix.anomalous_signal
2. Anomalous correlation between wavelengths. You get this if you run phenix.autosol with MAD data and look at the scaling log file. It will llist each pair-wise comparison of wavelengths and a resolution breakdown of CC values. Again, looking for 0.3 between peak and remote to indicate that data are strong to a particular resolution.
3. If you have independent data sets (different anomalous atom for example), and each one has a "solution", you can often get a very good idea of whether both are correct by calculating the map-map correlation (phenix.get_cc_mtz_mtz). If one is wrong, you get a low value (like 0.05); this tells you nothing. If the datasets are really the same you may get a very high number like 0.7 and that tells you nothing as well. But if both are right, and they are independent, you might get a value like 0.2 - 0.5. This is very encouraging if you get it. You can use this trick for all kinds of structure solution tries and you can automate it to compare everything in sight. The key is independence...
AutoSol won't really let you choose the number of sites...even if you tell it 5, it will look for more.
I'd say...always try hard with existing data and if you have the opportunity to get SeMet data, pursue that in parallel until you get it or you solve the structure with existing data.
All the best,
Tom T