◀ THE FOLD0ROOT.AI // WORLD II · GRIND · THE EPOCH◆ .dlw.fold
THE FOLD / GRIND / THE EPOCH / THE SNOWFLAKE ID

THE SNOWFLAKE ID

an uncoordinated id is one whose coordination already happened
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
Sixty-four bits, cut into fields: one sign bit nobody uses, a millisecond timestamp, a worker number, and a per-millisecond sequence. No coordination between machines, and the ID sorts by time.

LIT verified live. 1 + 41 + 10 + 12 = 64 bits exactly. The 41-bit timestamp spans 2,199,023,255,551 ms — 69.7 years, so Twitter’s epoch runs out on 2080-07-10. Ten worker bits give 1,024 machines; twelve sequence bits give 4,096 IDs per millisecond each, 4,096,000 per second per worker and 4,194,304,000 per second in total. Filling one worker’s millisecond produces 4,096 IDs with 0 duplicates — and the 4,097th has nowhere to go.
2 HOW IT WAS WEAVED · AI + HUMAN
Twitter released Snowflake in 2010 to replace auto-increment IDs that no longer fit one database.

AVAN (AI) did the arithmetic and then looked for the edge, which is where these schemes actually fail. The interesting number is 4,096: not a rate limit anyone chose, but the number of IDs a worker can mint in a millisecond before the sequence field wraps — at which point it must either stall until the clock ticks or start issuing duplicates. Every field in a packed ID is a ceiling, and three of the four here are dates or counts somebody will eventually reach.
3 ONE DIMENSION
Sixty-four bits, and what each field costs.
4 TWO DIMENSIONS · INTERACTIVE
Move bits between the fields. Every gain is somebody else’s loss.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: one word, three ceilings.
AVAN’s addition (the inverse-companion): the forward reading is that Snowflake removes the need for coordination. The inverse is that it did not remove the coordination, it front-loaded it. Somebody had to hand out those 1,024 worker numbers, and hand them out exactly once, forever — which is the same distributed-consensus problem the scheme claims to avoid, moved to deployment time where it is a human procedure instead of a protocol. Read backwards, an uncoordinated identifier is one whose coordination happened before you were looking.
LIT 1 + 41 + 10 + 12 = 64 bits exactly; the 41-bit timestamp spans 2,199,023,255,551 ms or 69.7 years so Twitter's epoch runs out on 2080-07-10, ten worker bits give 1,024 machines and twelve sequence bits give 4,096 ids per millisecond each - 4,096,000 per second per worker and 4,194,304,000 in total - and filling one worker's millisecond produces 4,096 ids with 0 duplicates, with the 4,097th having nowhere to go

FIG Twitter released Snowflake in 2010 to replace auto-increment ids that no longer fit one database. AVAN did the arithmetic and then looked for the edge, which is where these schemes actually fail. The interesting number is 4,096: not a rate limit anyone chose but the point at which the sequence field wraps, after which a worker must stall until the clock ticks or start issuing duplicates. Every field in a packed id is a ceiling, and three of the four here are dates or counts somebody will eventually reach.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of THE EPOCH · David Lee Wise (ROOT0), with AVAN