Rules of the hunt

Creator fees are assigned away at launch

Launching takes two signatures from the creator wallet. The first creates the coin. The second writes a pump.fun fee-sharing config that pays 100% of creator fees to one address, the vault for that mint. There is no split left for the creator. If the second transaction does not confirm, the coin is not listed here, because a coin whose fees still go to the creator is not a hunt.

The vault address is derived from a platform seed and shown on the coin page. Anyone can watch it. Spending from it requires the platform key, which is how a winner can be paid without asking the creator to sign a third time.

What the vault number means

A job claims creator fees every few minutes and leaves the SOL in the vault. The balance on the coin page is the spendable amount: the wallet’s SOL, minus a small float kept so the payout transaction has gas. The float is not prize money. If someone sends SOL directly to the vault address, that counts as well.

Clue progress uses growth since the round started, not the raw balance. The balance at the start is the seed. Only SOL that arrives after that counts toward the next clue. That is deliberate. A round that inherits a large vault should not publish every clue in the first minute.

The riddle is written once

When a round opens, a model writes a riddle from the coin’s name, ticker, and lore. It produces one answer, a few accepted variants, a scene-setting line, and six clues. Clue 1 is the hardest and is public immediately. Clues 2 through 6 stay on the server until the vault has grown another 0.2 SOL for each of them. They are easier on purpose. The prize and the difficulty move in opposite directions: more trading, bigger pot, plainer clues.

Before any guess is accepted, the page shows a SHA-256 hash of the answer. The salt that was mixed into that hash is not shown yet. The answer itself is not in the public API while the round is open. The model that wrote the clues is not something the coin page can query.

Who can submit, and how often

A guess is accepted only from a wallet that holds at least 0.1% of the coin’s supply at the moment of submission. The check reads the chain. A balance from ten minutes ago does not count. Selling below the line and guessing anyway fails.

Each wallet gets three guesses per hour on that round. The guess has to be signed by the wallet as a message, and the signature has to be fresh. A guess is a short phrase, 80 characters at most. The winner of a solved round cannot play the next one. They are blocked from submitting at all, so a correct phrase from that wallet is not treated as a win and is not announced as correct.

What “correct” means

Before comparison, the guess is lowercased, accents are removed, punctuation is turned into spaces, and repeated spaces are collapsed. A leading “the”, “a”, or “an” is ignored. “The golden bone” and “golden bone” are the same guess. The round also stores a short list of variants chosen when the riddle was written. A guess matches if it equals the answer or one of those variants after that cleanup.

Near misses that were not stored as variants are wrong. The server does not tell you that you were close. A wrong guess spends one of the three attempts and returns the number you have left.

The split, and when you see the answer

The first matching guess claims the round. Later correct guesses from other wallets lose, even if they were typed first, if they arrive after the claim. The answer and salt are published once the payout is recorded. Then a new round starts. It has a new phrase, a new salt, and a new hash. The seed left behind is the starting balance, not free clues.

If the last clue sits there

The 24 hour clock starts when clue 6 is released, not when the round starts. A quiet coin can sit on clue 1 for days. That is not a deadline. After clue 6 has been public for 24 hours with no correct guess, the answer and salt are published, nobody is paid, and the entire vault rolls into the next round. The next riddle starts immediately. Because the rolled SOL is seed, the new round still opens with one clue.

Checking a hash yourself

The commitment is SHA-256 of the salt, a colon, and the normalized answer, as UTF-8, with no extra spaces. Normalization is the same cleanup described above. After a round closes, the coin page and the public log both show the answer and the salt. Hash them. The result has to equal the hash that was on the page while the round was open.

Worked example, not a live round. Salt c0ffee12, answer golden bone. The string that gets hashed is c0ffee12:golden bone. The digest is:

854b9c84fd0e02fe7c7755e3790fc06f3a297fbe609ee7f0873d31a78a9b78bb

You can reproduce that in any SHA-256 tool. On a live round, use the salt and answer from the log, not this sample. “The golden bone” normalizes to the same answer, but the published answer is the canonical phrase, and that is the one inside the hash.

Guesses are not transactions

A guess goes to the server with a wallet signature over a message that includes the mint, the round number, the exact text, and a timestamp. It is not broadcast. Someone watching the chain cannot read the winning phrase and resubmit it before the payout confirms. The signature expires after two minutes, and a repeated signature is rejected.

While a round is open, API responses include the riddle, the released clues, the hash, the vault, and the clue progress. They do not include the answer, the salt, or the variant list. Those fields appear after the round is solved or expired. The public log then keeps the riddle, every clue, the hash, the salt, the answer, the winner, and the payout signature.