Post

Learning about TLA+ may have been a mistake. Now I want a much faster computer.

I’ve loved the practical problems of CAP since Sir Aphyr Kingsbury did terrible and wonderful things to databases back in The Day. Now I’m working through some hard problems with proving correctness in high volume distributed inventory systems, and I’m feeling thankful for what Twitter used to be.

peat

@peat.org

5 Sept

… the big ERPs are Actually Bad because humans aren’t good at counting things on shelves and the (very few) immutable bitemporal implementations are expensive AF and built on platforms that are extremely exposed to the AI bubble. Except for SAP. They’re just expensive AF.

peat

@peat.org

5 Sept

TLDR: if you think AI is a bubble, get out of ORCL and into SAP

peat

@peat.org

5 Sept

… or invest in something new. 😉

Rob Thomas

@rmichaelthomas.com

"Humans aren't good at counting things on shelves" stuck with me. That's not a bad entry — it's the shelf itself not being knowable, a different kind of wrong. Made me realize "is the count right" is really a few questions stacked together. Built a small thing that pulls them apart. 👀

5/9/26, 10:28 am0 reposts1 like

Rob Thomas

@rmichaelthomas.com

5 Sept

Followed the thread down a hole. Built a small thing that splits "is the count right" into separate checks — intact, attributable, legal move, and does it survive a shelf that won't hold still. Last one says "won't stabilize" not guess: github.com/rmichaelthom...

github.comGitHub - rmichaelthomas/inventory-attestContribute to rmichaelthomas/inventory-attest development by creating an account on GitHub.