UUID

The article quantifies how human typing habits influence token counts and thus costs across OpenAI and Claude tokenizers, with concrete examples and practical implications.

128-bit trace IDs are mathematically justified to minimize collisions over immense lifetimes, with 128 bits offering a practical sweet spot between collision safety and storage costs, while 256 bits would unduly inflate bandwidth/storage without meaningful real-world benefit.

ULID provides time-ordered, lexicographically sortable 128-bit identifiers (48-bit timestamp + 80-bit randomness) that can replace UUIDs in Postgres with sortable indexes, but carries caveats for ultra-high-volume workloads.