This is clever and thoughtful, but it encodes a 32-bit timestamp in seven base32 characters. It would have been simpler and just one more character to use eight hex digits. Alternatively, maybe go to 4 second resolution and use 30 bit timestamps, so they fit in six base32 chars instead of seven.
Those approaches would both rollover at 2106, breaking sorting, and you'd potentially start getting collisions around 2566. With his approach he is starting with a 63-bit timestamp (Go uses signed int64s), then math.big.Bytes is dropping leading 0 bytes, meaning it will add more characters as needed.
That's a separate thing, the 15 byte uuid's. I mean the timestamp slugs. Those are 32 bit timestamps that run out at the same time other Unix timestamps do.
