Coverage increased (+0.03%) to 99.119% when pulling e9abd04 on rlue:api/register_method into 25a22f7 on github:master.

Coverage increased (+0.03%) to 99.119% when pulling 66823c9 on rlue:api/register_method into 25a22f7 on github:master.

Ninja edit: Fixed CI failures and rewrote commit messages.

Note that two out of four CI failures were caused by the auto-configuration "magic" I introduced in the second commit of this PR. This meant that in order to get the tests to pass, I had to add a line to the test setup to revert that "magic".

If the maintainers find this change of behavior unacceptable, I'd be happy to roll this PR back by one commit.

I would like to see this as well, especially with the readme updates with the named arg corrections.

Anything I can do to help get this merged?

I like the idea of a class method to set the default experiment rather than overriding new. What do you think about set_default(klass) instead of register(klass)?register feels like adding something to a registry, implying multiple items, whereas set_default is more explicit.

I was initially concerned about automatically registering upon module inclusion since it could lead to require-time ordering issues if there are multiple base experiments. Since Scientist.run is only ever expecting a single implementation, though, that should be fine.

Good point; I'm on board!

Thanks! I've released this as 1.5.0