Cool stuff that could be done

[ml-evs]

This is a bit of lark but if we actually start generating some statistics we could do stuff like:

• [ ] vary the subset of structures each month and incorporate different datasets (including ones that actually have real structures, useful for per-user calibration)
• [ ] allow users to provide some extra info about their background if they do 10 predictions (so their views can be (de)emphasized)
• [ ] leaderboard of the last months best and worst structures; could then use this to actually promote further DFT or synthesis work