There's an hard upper limit on the TPS Postgres can handle, and it only depends on network latency and contention. With 10ms RTT and 90% contention it's 111 TPS.
This is assuming an infinitely powerful and parallel Postgres cluster.
I really recommend watching that talk (also: supercool animations).

Yes, but you also do get full replication, high availability, storage fault tolerance etc, which might be something you want for your payments domain.

Regarding the pure performance aspect, Postgres might be fine for their workload, but there's an interesting fact I discovered from this talk [...]

Re: "this stuff should be proceduralised", check out TigerBeetle, a database that has double-entry accounting built in its data model (plus a bunch of other really cool features)
tigerbeetle.com

I also wrote an Elixir client for it
github.com/rbino/tigerb...

Since Elixir 1.17 you can also use `to_timeout/1`, which supports units up to weeks and allows you to pass multiple units, e.g. to_timeout(hour: 1, minute: 30)

hexdocs.pm/elixir/1.19....

I really enjoyed this talk about the topic at ElixirConf EU
youtu.be/AnbWtyB3pcQ

🙋

I have a couple of Keebio Iris, my general suggestion is picking something with not too many keys so you're forced to make your own layout with layers and move your hand less (e.g. I have all the parentheses on my home row in the lower layer).