Tests and docs

[tapios]

Last major step of RRTMGP refactor:

• Some added tests for paths not previously covered
• Extensive docs, including how-to guides, tutorials (e.g., reproducing Manabe's RCE results within the docbuilds), and exposition of concepts (RTE, optics etc)

Closes #515, #585

[tapios]

Yes, it’s not deterministically reproducible on GPUs. To make it reproducible, we’d have to pass, e.g., a column index as seed (what I had a week ago or so). (Here it says it’s statistically reproducible, which is true but perhaps unhelpful.) On Jul 11, 2026, at 20:18, Zhaoyi Shen ***@***.***> wrote:  @szy21 commented on this pull request.

________________________________ In docs/src/howto/gpu.md<#607 (comment)>:

+stochastically (McICA) using the global RNG. On the CPU,

+`update_fluxes!(solver, seed)` with `reset_rng_seed = true` makes results +reproducible. On the GPU, the per-thread RNG state produces statistically My claude says it's only reproducible on single-threaded CPU. Is that the case? — Reply to this email directly, view it on GitHub<#607?email_source=notifications&email_token=ADT3LJHJIWS4ZHQVJGBCBDT5EL7OFA5CNFSNUABKM5UWIORPF5TWS5BNNB2WEL2QOVWGYUTFOF2WK43UKJSXM2LFO4XTINRXHEZTCNRSGI22M4TFMFZW63VGMF2XI2DPOKSWK5TFNZ2KYZTPN52GK4S7MNWGSY3L#pullrequestreview-4679316225>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ADT3LJEDJJZOT5ZV5XZWZA35EL7OFAVCNFSNUABFKJSXA33TNF2G64TZHMZDAOBTGIZTSNZQHNEXG43VMU5TIOBUGM2TCMZXGA22C5QC>. You are receiving this because you authored the thread.Message ID: ***@***.***>