I tried running their exact example queries, since I conveniently have a >100k row PostgreSQL table available of the same name. (> 100k rows) The first (offset)

I tried running their exact example queries, since I conveniently have a >100k row PostgreSQL table available of the same name. (> 100k rows) The first (offset) was indeed on the order of 5x slower, but I suspect what the author saw with a 40ms to a 0.1ms time difference was that it wasn't pre-warmed first run. For me, with offset, the FIRST run was 25ms, but subsequet runs are around 2-3ms. For `id > 10000`, it was under 0.05ms. It's also very odd that they don't do an `order by`. This seems important. Default ordering is semi-stable. It'll tend not to change, but it can at any point. Benchmarking without that may be misleading.