Matthew Kay

One more minor TODO for this issue: - [x] Once the vignette describing the different draws formats exists, add a reference to it in the `rvar` vignette here: https://github.com/stan-dev/posterior/blob/0184c56f0f6f7a7f9183f418983c1d028fef2942/vignettes/rvar.Rmd#L99

> Colloquially stating the point above in R memory do whatever you want, whether it's one object, or 5 or 10, but would be cool and have most of the...

Looking at this: https://arviz-devs.github.io/arviz/schema/schema.html The format that comes to mind for me is a list where each element is draws object representing one "group" in the terminology of InferenceData. Questions...

@jburos I checked out the `tbl_cube` thing and it seems pretty useful! Another approach in that spirit could be to use `rvar` columns in a tibble. From a quick test...

> You see that the rvars code is not nice as I have to use many times drop. This is due to the fact that I nest matrices and vectors...

One option is to ingest `posterior_...()` function output via `rvar()`, as the internal format of `rvar()` is (by design) the same as the output of those functions. `rvar()` lets you...

> For rvars it might be the easiest(?) to maintain an attribute as all transformations are fully custom there anywhere. Yeah, the annoying stuff to make this work in rvars...

I like the idea of code linting and a PR template. I'll push back a bit on spending time cleaning up commit history: I'm not sure I see the point...

I think getting this to work with summarise_draws as suggested will require allowing the measures argument to accept functions? (see #58). I like that approach. Thinking about general use, this...

Some thoughts here: - from what I remember of JAGS (it's been a few years) it just uses `integer`s for discrete parameters. That seems fine enough for most use cases....