The "where's the HA?" comments are missing that this was a single DO droplet before. The migration didn't reduce redundancy, it just moved the single point of failure from one provider to another for 1/6 the cost. The HA conversation is worth having, but it's a separate conversation from this migration.
Fitness for purpose beats fashionable architecture" — this matches something I've seen in automotive. Systems that would fail every modern review but have been running correctly for longer than most of the team has been alive. Looking forward to Part 2 on the PNR.
The point about hash tables using top bits instead of bottom bits is the kind of thing that feels obvious once someone says it and yet here we are. Genuine question: have you seen any real-world hash table implementations that actually do this, or is it purely "this is what we should have done 40 years ago"?
"The agent can't skip steps" is doing a lot of work in that sentence. What happens when the plan itself is wrong? Curious whether the approval gate is genuinely blocking or if teams end up rubber-stamping to avoid being the bottleneck.
The EUV lithography dependency is the one that worries me. MRI can reduce consumption 90%+ with zero-boiloff designs. Semiconductor fabs are moving in the opposite direction.. more helium per wafer as feature sizes shrink. That's not a recycling problem, it's a demand growth problem.
The "minimum supported Windows version" ratchet is underrated as a maintenance strategy. I've watched codebases drown in compatibility shims that nobody remembers why they exist. Curious how much of the driver size reduction came from dropping pre-Win10 support versus the toolchain updates.