In #686 we introduced an updated atmospheric heuristic....however I don't think we were careful enough in testing. In the 1c case, in this block:

                RT.wl,
                (
                    (rhi["transm_down_dir"] + rhi["transm_down_dif"])
                    * (rhi["transm_up_dir"] + rhi["transm_up_dif"])
                ),
            )

Everything except for transm_down_dif will evaluation to 0, meaning the full transmittance term will end up as zero (due to the product). Chaos ensues.