Sounds like it has similar goals to bcache (not bcachefs) and LVM caching, except that it operates in userspace instead of at the kernel level. Can you explain what the benefits are of keeping this out of the kernel?

The slowness was not a tax on the real work; the slowness was the real work. It was how the work got good, and how the people producing the work got good

This line reminded me of a couple of articles, linked below, that I read on AI use in astrophysics. Developing junior researchers is a big part of the point of their work, so they really are going to have to limit their AI use to make sure development happens. But worry that industry won’t care; they’ve been hollowing out junior positions for years, because there’s no value in training a senior who is just going to jump ship to a company that doesn’t train juniors. That’s an existing problem, but AI seems likely to make it worse.

• The machines are fine. I’m worried about us. by Minas Karamanis
• Find the stable and pull out the bolt by Natalie B. Hogg

And that’s only, like, the most basic part of an industrial e-stop. On top of directly disconnecting the energy source, it also has to include redundant circuits (since it’s possible for a contactor to break and fail open, or weld and fail closed), and some even need to have signal pulses constantly going through them to actively confirm they haven’t been pressed. They absolutely should not just send a signal to a Linux computer, because general-purpose operating systems are too complex to be rated for safety.

This isn’t even my area of expertise yet I know this stuff. The fact that they failed at the very first requirement is really messed up.

Even if you press that big red emergency stop button on the mower itself, a hacker can send another command to unlock it, Makris says.

Holy cow. I work in factories, and I know enough about industrial safety to know that would not fly.