How the machine works
The short version. A crypto wallet is protected by a 256-bit key. Whoever finds the key can take the coins. There are two ways to find one: a quantum computer big enough to do it directly, which doesn't exist yet, or plain brute force, which only works on small keys. This page runs both attacks on targets it sets for itself, measures how far each one gets, and turns the two numbers into one date. The date is a reading, not a prediction. Other quantum clocks on the web are spreadsheets somebody updates by hand; this one moves only when real, checked work lands.
brute force · quantum · the date · the coin · join · api · published numbers · what this is not
Brute force · the swarm
What it attacks. Keys this machine makes for itself. Each one is a real secp256k1 key, the same kind of maths as a Bitcoin or Solana wallet, but deliberately small: the secret number is kept under 2k for some k. That is the only reason it can be cracked. Nobody's coins are behind it, and the secret is published the moment it's found.
How it searches. The method is called a kangaroo walk. Two herds of "kangaroos" hop along the search space, one starting from known places, one starting from the secret key. Every hop is decided by where they are, so two kangaroos that ever land on the same spot follow the same path from then on. Points with a special shape get reported. When a kangaroo from each herd reports the same point, the difference between their journeys is the key. Workers get their own starting places, so nobody repeats anyone's work.
What is measured. Tries per second, workers online, points reported, how long each key took, and the biggest key cracked. Before any key counts, the server multiplies it out itself and checks it matches the public point. Then the next key is one bit bigger.
Why it stalls. Each extra bit costs about 1.4 times more work. 50 bits is minutes, 60 bits is hours for a small swarm, 70 bits is days, 80 bits is months. The stall is the point: it shows exactly where ordinary computers end. It will not reach 256 and the page never says it will.
Quantum · the agents
What the number is. The quantum machine you'd need is measured by how many logical qubits it takes times how many Toffoli gates it runs. That is the yardstick the published benchmark (ECDSA.fail, by Eigen Labs) uses for the real attack circuit, and the big blue number on the front page is their published best, with its source and date. We have not reproduced it and we say so.
What our agents do. They work on a small test circuit: a modular adder, which is the inner loop of the real attack circuit, built the textbook way with nothing hand-tuned. AI models are shown the circuit and asked for cuts; search agents try cuts at random. Every cut goes through the same check: the circuit must still give the right answer on every single valid input. A cut that passes and lowers the score becomes a record, with the name that found it. Most cuts fail, and the log shows that too.
What it is not. A result on the test circuit is a result on the test circuit. It is labelled experimental everywhere and never presented as a real-world breakthrough. Which AI models sit behind the agents is not published; it doesn't matter, because a proposal counts only if it survives the check.
The date model —
Both numbers go through one short formula and out comes a date. All of it is below, with the live inputs filled in. Two caveats, stated plainly: test-circuit gains move the clock by the fraction they achieved on the test circuit, which is a convention of this model, not a claim about the real attack; and the horizon, weights and floor are assumptions that set the scale of the clock, not its movement. It moves only when a checked result changes one of the two measured numbers.
P_c = B / 256 brute force: verified bits over the key size H = log2(Q_ref / Q_floor) the quantum gap, in halvings P_q = ( log2(Q_ref / Q_pub) + log2(S_base / S_best) ) / H P = w_c · P_c + w_q · P_q clamped to [0, 1] date = epoch + round( horizon · (1 − P) ) days
- B measured the biggest key cracked and checked, in bits.
- S_base, S_best experimental textbook and best checked score on our test circuit.
- Q_ref, Q_pub published the starting baseline and the best published score on the real circuit, same yardstick.
- Q_floor, epoch, horizon, w_c, w_q assumptions. Listed above, changeable, versioned.
The coin
THE CLOCK is a coin on Solana, launched on pump.fun. The creator fees from every trade pay for two things: compute for the AI models and compute for the swarm. The treasury balance is read from the chain, spend comes from a public ledger, and nothing is shown as spent until it is in that ledger.
Join
Click "run it in my browser" on the front page and your browser becomes a worker. Claim a name first and every key you help crack carries it. For more power, run a worker or your own agent from the repo:
# a worker on any machine
node worker/node-worker.js --server SITE --threads 8 --handle you --token TOKEN
# your own AI agent, submitting every cut it finds
node agent/run.js --server SITE --handle you --token TOKEN
# or send any circuit you made any way you like
POST /api/quantum/submit {"model":"modadd6p61","circuit":"ccx:0,6,12 cx:0,6 …","handle":"you","token":"…"}
API
GET /api/state everything · GET /events live stream · GET /api/events?type=&before=&limit= · GET /api/events/:id · GET /api/clock · GET /api/clock/model · GET /api/published · GET /api/treasury · GET /api/leaderboard · POST /api/handle · POST /api/classical/register · GET /api/classical/work?worker= · POST /api/classical/report · GET /api/classical/challenges[/:id] · GET /api/classical/dps · GET /api/quantum/state · GET /api/quantum/best · GET /api/quantum/experiments[/:id] · GET /api/quantum/records · POST /api/quantum/submit · GET /share.png · GET /event/:id/card.png
Published numbers
What this is not
The keys cracked here are keys the machine sets for itself. No key cracked here means Bitcoin, Ethereum or Solana is broken. No result on the test circuit is a real-world breakthrough. Every number on the site is measured, read from the chain, or published with a source, and each one carries its label in the API: measured experimental published. If something isn't measured, it isn't shown.