Seth Axen

> Require that the `arviz_version` attribute is always included. Unless we intend for packages besides ArviZ to define their own implementations of the schema. In that case, there are other...

> Not sure if there is another issue on this or so far we have only discussed on slack, but I also think it would be good to move the...

Also the warm-up groups are not listed in the schema so technically they violate the schema. Is this just an oversight?

@OriolAbril the schema describes attributes as being an ordered dictionary. Should we relax the condition that it be ordered? Zarr.jl returns attributes as a plain unordered `Dict`. Should we also...

I'm currently implementing a Julia wrapper for ArviZ called [Arviz.jl](https://github.com/sdaxen/ArviZ.jl). It includes a `from_mcmcchains` function to convert an `MCMCChains.Chains` object into an `InferenceData`. That's the output type returned by a...

Julia has a `missing` value similar to a `nan` that explicitly annotates a value that is missing. Variables in [infinite mixture models in Turing](https://turing.ml/dev/tutorials/6-infinitemixturemodel/#infinite-mixture-model) sometimes take on this value. Do...

Also, is there a place where all of the recommended conventions for `sample_stats` names are documented?

> Connected to #1483, there is a latent need for being able to dispatch on distributions which have merely uncountable support and those which are absolute continuous. I'm not certain...

It's unfortunate we use `pdf` to refer to the PDF and the PMF. For a discrete distribution, it's returning probabilities so it should be unit less, but I think for...

Agreed, most of the overloads that use `.mat` could be written to be more efficient or just a little less efficient by never computing the full matrix.