Prototype. The flow you see is simulated; nothing here is live and nothing moves money.

CLOCKED — every morning at a random minute a secret shopper orders a $1 transfer and a $1 swap on Arc, times it to final, keeps the receipt, and on Sunday prints the column.

00061218
window 06:00-10:00 UTC · today's dot marks the current minute936 msmedian send-to-finalOpen a tabsee the house shopper's wallet cap, orders & schedule

simulated — this week's mediansimulated

Ordered 06:41:07 UTC. Menu says sub-second. Plate came out in 1243 ms. Fee 0.000421 USDC to 18 decimals.

A $1 transfer and a $1 swap at a random minute, send to final on a stopwatch, fee to the last of 18 decimals, read against the launch claims every Sunday. Open a tab to see the house shopper's wallet cap and schedule now, then watch each run land on the spike.

moved · Upcoming

Arc

2026-09-16 · 1d 12h ago

Worker dry-runs on Arc testnet; page lists the launch claims (sub-second finality, ~$0.001 fee target) that will be tested.

Official source: https://www.circle.com/pressroom/circle-announces-founding-validator-cohort-and-major-integrations-for-arc-ahead-of-september-16-mainnet-launch · verified 2026-09-16T18:48:42+00:00

  • is on track for a public mainnet launch on September 16, 2026” — official
  • Arc Mainnet will go live on September 16, 2026.” — official

Sources

the shopper walks in at a minute nobody can predict

The minute is drawn, not picked

You set a window. The shopper draws a minute inside it each night. Nobody gets to order in a quiet hour.

window: 06:0010:00 UTC · next minute drawn: 2026-09-18T06:00:00.000Z

today's order ticket clipped to the rail

Today's order

Ordered, final, and the stopwatch reading — or, most days for now: no order yet today.

Read receipt →
NO ORDER YET TODAY
Next minute drawn: 2026-09-18T06:00:00.000Z

the menu board: one coffee, one slice, one dollar each

On the menu: one coffee, one slice

The coffee is a $1 USDC transfer. The slice is a $1 swap on univ4-arc. Same order every day so the only thing that changes is the kitchen.

Coffee — $1 transfer

$1.00

Slice — $1 swap (univ4-arc)

$1.00

stopwatch readings pinned under the menu board promise

Menu says. Stopwatch says.

Every dot is one run, send to final, measured from the RPC. The line is what launch said (C3).

simulated

the till roll printing the check

The check, to the last digit

Fees are paid in USDC and printed exactly as the chain reports them. Nothing rounded, nothing averaged away — launch targets about $0.001 per transfer (C10).

0.000424 USDC last runsimulated

RunFee (USDC, 18dp)
R0300.000424000000000000simulated
R0290.000507000000000000simulated
R0270.000576000000000000simulated
R0260.000490000000000000simulated
R0240.000554000000000000simulated

receipts pushed down onto the steel spindle by the register

Receipts speared on a steel spindle by the register

The spike

Every receipt, newest on top, 50 at a time. Filter slow, pricey, sent back.

SlowPriceySent back
RunSentLatencyFee
R0302026-09-16 09:58437 mssimulated0.000424simulated
R0292026-09-15 08:53779 mssimulated0.000507simulated
R0282026-09-14 08:03sent back
R0272026-09-13 07:561682 mssimulated0.000576simulated
R0262026-09-12 08:56936 mssimulated0.000490simulated
R0252026-09-11 07:29sent back
R0242026-09-10 08:531814 mssimulated0.000554simulated
R0232026-09-09 07:30627 mssimulated0.000472simulated
R0222026-09-08 09:46999 mssimulated0.000607simulated
R0212026-09-07 07:201654 mssimulated0.000597simulated

the tray of orders sent back to the kitchen

Sent back

Failed runs stay on the record with the reason the kitchen gave. A benchmark that drops them is a menu, not a review.

2 sent back this week: R028, R025

simulated
  • maxFeePerGas below 20 Gwei — dropped by the mempool (C13)
  • RPC timeout on send — no receipt to record
  • Swap slippage exceeded on univ4-arc

the kitchen swing door: what the dining room can and cannot see

What we can see from the dining room

Clocked measures what a customer sees: the time from our worker's send call to the RPC reporting the transaction final, and the fee on the receipt in USDC to 18 decimals. Each run keeps the exact RPC call it was measured from. Runs happen at a random minute inside the shopper's window so results are not tuned to quiet hours.

We do not measure validator internals, we do not average away failed runs, and a week with no runs shows no numbers. The worker signs only from the shopper's own agent wallet under the spend limit the shopper set; the shopper can revoke it at any time. Repeated block timestamps on Arc are flagged, not trusted (C14) — we cannot see inside the kitchen, only what comes back out through the swing door.

  • Latency is measured send-call to RPC-confirms-final, never a self-reported number.
  • Fees are read from the receipt itself, printed to 18 decimals, never rounded.
  • A week with zero runs shows zero numbers — no average papers over a quiet week.
  • The house shopper only ever signs from its own capped wallet, never a visitor's.

Read the full method and every sourced claim →

the columnist's typewriter in the corner booth

Sunday column

The week's slowest, priciest and failed runs, typed up next to the claims they test. Published by the person who funded the wallet.

Read the column →
Post on X
An empty stool at the counter with a blank order ticket on the rail

an empty stool at the counter with a blank order ticket

Pull up a stool

Bring your own agent wallet, set your own cap, revoke it when you like. Clocked never holds your money.

Open a tab