When I started learning crystallography around 2007, there were very few public datasets and I envied those who have access to datasets. Senior people looked like wizards, knowing all space groups, deciphering Harker sections, writing cryptic keyworded inputs without GUI.

I asked them how they learned all this and their answer was "by experience". They had worked with a dataset in this space group, so they knew where the symmetry axis is. They had processed twinned datasets before, so they knew how to tell.

They had struggled with a program, so they knew which parameter to change. You cannot really learn these things by just reading text books or papers (although this is also very important). You need real experiences!

But without having access to datasets, how could I learn?

My lab was working on very hard membrane proteins, so there were not so many datasets around. Moreover, my colleagues said I should focus on sample prep; I should learn data processing only after I got crystals. So I was very frustrated (even depressed).

Experts accumulate more and more experiences because many people send them difficult datasets. Those close to experts can also learn a lot via internal lab seminars or "private communication".
But people outside the circle could never have such experiences...

Such memory of starvation is one reason why I am so keen on open science. I want everybody to have access to datasets to play with. I also disseminate my experiences and know-hows via mailing lists and tweets.

Same for source codes. Some people knew internals of closed-source programs like SHELX and XDS via "personal communication" with the developers but I felt it is very unfair.

For X-ray macromolecular crystallography, we have lots of raw data available online!

- https://proteindiffraction.org/ 
- https://data.sbgrid.org/ 

Small wedge datasets
- https://github.com/keitaroyam/yamtbx/wiki

Serial crystallography (XFEL and synchrotrons)
- https://cxidb.org/browse.html 

Unfortunately, MicroED datasets are very rare and there are few small molecular datasets (both X-ray and ED). Although most of small molecular crystallography is well-established and straightforward, there are interesting pathologies like order-disorder.