What's the path of getting there, maybe?

Yeah.

What's the path of getting there? I mean, sometimes you sit down and you're like, I wish I knew, right? The matchmaking in the first instance is a data collection problem in the first instance.

It's really a deep understanding problem in the first instance.

How do you understand the chips? But also, how do you understand the workloads? How do you understand how they are brought together? So there's a bit of a discussion that we just had.

It's ultimately a large-scale testing exercise, but also how to keep the maximum experimental data, how to queue format, how to extrapolate from two different sets to a new workload.

So there's lots of just nearly kind of data science-like work on the back end, which is like the fundamental data you need to capture to do anything.

Does that feel right to you? Is that intellectually