Aaron Spring

Can observations also be 0 or 1 instead of boolean? I think this also works.

1. If 0 or 1 also works type checking doesn't bring us much. Does it? Clear documentation will do 2. Utils or new file checks 3. Yes

i tried this here: https://github.com/csiro-dcfp/doppyo/blob/98acad9fe3cf658d5b000401ba898ef343abb768/doppyo/skill.py#L256 but didnt account for `if not dask.is_dask_collection()`

> Good point, 0 or 1 also works, so checking would only be run with np-backended arrays. Maybe not worth it then. @bradyrx How would the checks look like? Maybe...

properscoring has checks https://github.com/TheClimateCorporation/properscoring/blob/1ca13dcbc1abf53d07474b74fbe3567fd4045668/properscoring/_brier.py#L41 and in xs we do a dask test https://github.com/raybellwaves/xskillscore/blob/de4b53843fbab3d841388d4fce866b81a6748476/xskillscore/tests/test_probabilistic.py#L250 So these checks trigger computation when via xr applyufunc at least @dougiesquire

Sorry. I meant to say these tests don't trigger computation in these tests.

> But not sure if it's for type hinting or checks. ~~Exactly what typing ensures #317 ~~ EDIT: wrong at runtime no check

basically just copying their code?

Once we took over, invite this PR https://github.com/TheClimateCorporation/properscoring/issues/7 and raise issue in ps that we forked

I dont see the current quick-start as an API dump. Sure, we use synthetical data, and every user wants a real world example (from their field). we show how the...