A 4R pipeline costs too conveniently close to 500G leaving two such pipelines in 900s. It's not quite clear how the rest of cost can make things faster, but a couple sad arms do help initialization.

Another latency I can see is that separating salt and cardinal and breaking cardinal into 2-atom-molecules can be done without handoff. However this only makes sense on the slower pipeline and my slower pipeline is too kinked to allow that swing.