01
Post a bounty
Paste any program id on Solana and put a USDC number on it. You sign a message, not a transaction, so nothing leaves your wallet. Bounties stack: every backer makes a target louder.
Bounties for closed Solana programs
Name a program you want opened and what it is worth to you. Breakers rebuild it in Rust, a referee replays real mainnet traffic against both binaries, and the source stays locked until somebody pays to pry it open.
$PRY
CA
dropping soon
COMMIT a3f9…c21b
LIVE
Three moves, one lever
Nobody pays for promises. Bounties cost nothing to sign, rebuilds are judged by a machine, and money only moves when someone wants the key.
02
Rebuild it in Rust
Breakers turn bytecode back into a crate, bounty or not. Your browser encrypts the crate before it uploads, and you set the asking price, 25 USDC at the least. The referee handles the rest.
03
Pry it open SOON
A passing verdict goes public right away. A buyer pays in USDC, receives the key, and their browser checks the archive against the commitment printed on the verdict before it saves a single byte.
Open the vault →
SolanaUSDCRustJupiterPhantomSolflareBackpackLiteSVMpump.fun
Where the money goes
0%
To the breaker
Paid out the moment an unlock settles, with an on-chain memo tying the payout to its verdict.
0days
First dibs
Whoever backs a target the hardest gets the first right to buy its rebuild. After that, anyone can.
0%
Bought & burned
Of every unlock, swapped into $PRY on Jupiter and sent to the incinerator.
0targets wanted
0USDC on the table
0proofs sealed
0verdicts issued
The referee trusts nothing you say
No uploaded binaries, ever. It compiles your crate itself with a pinned toolchain, offline, under a ten minute cap, then feeds the same mainnet instructions and account snapshots to the original and to your build.
REPLAY
0 / 1,000 matched
MAINNET
CORPUS
CORPUS
original.so
rebuild.so
Twin sandboxes, same compute budget, same account states. Real instructions first, fuzzed ones after.
A
Every case has to match: result, error code, return data, logs and every writable account byte.
B
A single difference fails the run, and the report points straight at it.
C
A pass needs at least 100 cases where the original program succeeds.
D
Compute-unit log lines are stripped. Everything else gets compared.
The vault
Every rebuild that passed, locked until someone opens it. Calibration runs are the referee testing itself on programs whose source is already public.