Seventy-seven Bitcoin addresses are sitting in the open holding 903 BTC, and not one of them can be spent, because nobody knows their private keys. dec0ded converts trading fees on Robinhood Chain into the raw compute that hunts those keys, beginning with the five where the arithmetic actually works in our favour. While trying to solve the puzzle we will also look for inactive wallets that are holding funds.
Three facts explain the entire project. None of them require you to trust anybody, including us.
In January 2015, a person known as saatoshi_rising funded 160 Bitcoin addresses in one transaction. The private key of puzzle #1 sat in a 1-bit range, #2 in a 2-bit range, and every step after that doubled. Puzzles up to #70 have all been cracked. In April 2023 the creator multiplied the remaining prizes by ten, which is why #140 now holds 14 BTC. Those coins have never moved.
An address is only a hash of a public key, so normally you learn nothing from it. But in May 2019 the creator spent 1,000 satoshis out of #140, #145, #150, #155 and #160, and a spend publishes the public key. With that key in hand the hunt stops being a guessing game and becomes a computation: about the square root of the range instead of the whole range.
There is no product here and nothing to believe in. Trading fees land in one public wallet, that wallet buys GPU hours, and those hours walk a keyspace. If a key falls out, the Bitcoin is real and the transaction is visible to everyone at the same moment. If it doesn't, the wallet still shows exactly where every dollar went.
Five steps. Every one of them is something an outsider can check, which is the only reason to lay it out like this.
Every trade on Robinhood Chain pays a fee. That fee is the project's only source of revenue.
They settle in a single address, 0xdec0ded…a221adf, with no multisig story and nowhere to hide.
The balance is spent on rented compute, priced in GPU-hours. Nothing is manufactured; hardware is bought at market rate.
Each worker gets a slice of the target's interval and walks it with Kangaroo, or hashes it blind on the targets that never leaked a key.
Two herds land on the same point, the private key drops out of the difference, and the coins move on Bitcoin in front of everybody.
Every box reports itself. There is no list to maintain: a new Vast instance shows up within fifteen seconds of the agent starting, and drops off if it goes quiet.
No rig has reported yet. Rows appear here within seconds of an agent starting on a box.
Every tile is a live Bitcoin address. Open one for its exact key range, the attack it allows, and a balance check straight from a public node. Board verified 2026-09-22.
Green tiles leaked a public key in 2019 and can be attacked with Kangaroo. Everything else has to be brute-forced, which is why the bounty on a tile tells you almost nothing about how hard it is.
The animation below is the actual algorithm, not a metaphor for it. Watch two herds walk until they step on the same point.
A blind puzzle leaves you one move: pick a key, hash it, compare it to the address, repeat. #71 hides its key somewhere in 270 candidates, so you expect to try about 590 billion billion of them before one matches.
A puzzle with an exposed public key gives you geometry instead. Kangaroo releases two herds onto the interval, and the length of every hop is decided by wherever that kangaroo currently stands. The moment one herd lands on a point the other already visited, both follow the identical path forever, and the gap between where they started is the private key. The work collapses to roughly the square root of the interval.
#135 fell exactly this way in July 2026, after the solver known as RetiredCoder spent about five months on 200 GPUs. #140 is the next rung on the same ladder, and it is where dec0ded starts.
hashes expected. 270 range, no public key, no shortcut of any kind.
hops expected. A range 269 times larger, for 1.6× the work.
Nothing about the puzzle is speculative. It has a transaction history, a body of solvers, and a frontier that keeps moving.
160 addresses funded in a single transaction, each private key placed in a range twice the size of the one before it.
The creator sweeps addresses #161–#256 back into the lower ones, raising every remaining prize.
The event this project depends on. 1,000 satoshis are spent out of #65, #70 and every fifth address up to #160. Each spend publishes that address's public key, which turned #140, #145, #150, #155 and #160 into computable targets.
Every unsolved prize is multiplied by ten. #140 becomes 14 BTC, #160 becomes 16 BTC.
#66 falls. The winning transaction is broadcast into the public mempool and most of the 6.6 BTC is replaced out from under the solver before it confirms, which is why every later solve was mined privately.
#67, #68 and #69 fall within ten weeks of each other, all three mined out of band.
#135 falls to RetiredCoder, the highest puzzle ever solved, using Kangaroo on its exposed public key.
"Finally I solved #135, it took about 5 months on 200 GPUs. It was… long. I quit, officially. Happy solving #140 without me."
#140 is the lowest unsolved target with an exposed public key. 77 puzzles remain, holding 903 BTC between them.
Dates, addresses and balances on this page were taken from privatekeys.pw ↗ and can be re-checked against any Bitcoin node.
The project's wallet was public before this page existed. It is the only accounting statement dec0ded will ever need.
Fees arrive here. Compute is paid for from here. Both directions are on-chain.
It does not hold a puzzle prize and it never will. A recovered key spends on Bitcoin,
not here, so if #140 is ever solved, the proof appears at
1QKBaU6WAeycb3DbKbLBkX7vJiaS8r42Xo
within seconds, on the most public ledger there is. You will not need this website to tell you.
dec0ded is an experiment. It does not promise a solve, a payout, or a schedule. The bounties are a public race that anybody can win at any moment, including people who have never heard of this project.
Target changes and anything spent out of this wallet get posted on @dec0dedtech ↗
Answered directly, including the ones without a comfortable answer.
No. The creator built the puzzle as a public measurement of how much cracking power the world has, said so on BitcoinTalk at the time, and has topped the prizes up twice since. The keys are meant to be found, and whoever finds one spends it. It has been an open race with published rules for eleven years.
Because #140 leaked its public key and #71 did not. In work terms they are almost the same target, roughly 1021 operations each, but #140 pays 14 BTC instead of 7.1, and its work can be split cleanly across rented machines. #71 stays on the board; it is simply worse value for the same electricity.
The tile turns, the hardware moves to the next target, and the board carries on. There are 77 of them holding 903 BTC, and the five computable ones alone hold 75 BTC. This is the reason the project is not named after a single puzzle.
A hundred GPUs pointed at #140 would expect to spend decades. Ten thousand would expect months. That gap is the entire argument for funding this with a fee stream instead of a fixed budget: the only lever that matters is how much compute is running, and fees compound.
It is also why nobody should treat a solve as scheduled. Expected work is a median, not a deadline. Every target on the board carries the exact figure, and luck moves a real run in both directions.
It is where the experiment lives and where the fees come from. The project wallet is public there, so the flow from trading activity to compute spending can be followed by anyone without a dashboard, an API key, or our permission.
Not yet, in the sense that there is no pool to point a machine at. What already exists is the reporting side of it: every rig the project runs publishes its own readings to the fleet panel above, so outside hardware can be added the day the work-splitting side is ready without asking anybody to take a number on trust. Changes get announced on @dec0dedtech ↗.