Questions.

plainly

Ten of them.

Is this TypeSafe, or endorsed by them?

No. It is not affiliated with, endorsed by or authorised by TypeSafe AI. Jev is their product. This project names it the way one names any tool it depends on, and uses none of their branding, colours or domains.

Do I need my own API account?

Yes. The rig calls the API with your key, from your machine, in your account. There is no shared key and no proxy of ours in the middle.

Can I use a different model?

You can send the pack anywhere you like. Both gates are public and measured. The two contenders that were timed need 1.4 and 1.8 s for a pack; the faster one is within tens of milliseconds of the window, not safely outside it. On the whole screened pool the contenders were wrong on 3.6% to 7.4% of items, and with two errors allowed in 128 that leaves them passing about 15% of packs at best, where Jev's sample shows no errors. A pack that misses either gate scores zero, not less.

Who decides my score?

The contract. It checks that your answers match what you committed inside the window, then scores them against a key that was committed before the round. Nobody posts a score for you.

Where does the money come from?

One place: the 2% tax on buys and sells, less Flap's own protocol fee (on testnet the vault received 90% of the tax, 1.8% of the buy), held in the vault as BNB. There is no treasury and no emission. A round releases up to a fifth of what the pool holds and the rest rolls over, so quiet trading shrinks the pool slowly; a round pays nothing only when the pool is empty.

What will I earn?

Unknown, and this site will not guess. Your share of a round is your credit times your stake, divided by the same quantity summed over everyone who scored. Both the pool and the field are unknown in advance, so any number here would be an invention.

Does staking more earn more?

Yes, proportionally. Share comes from capital; the multiplier on it comes from the tool you run. A better tool does not hand a small holder a large share, and this project does not pretend it does.

Is running many wallets worth it?

No. Payout is linear in stake, so splitting one stake across ten wallets pays what one wallet pays and never more (each share is rounded down), and costs ten times the gas. A stake is locked for the round, so the same tokens cannot be counted twice.

What if one of your labels is wrong?

It happens, and it has: the first real run caught two label bugs. The chain cannot tell a wrong label from a right one, so Flap's Guardian can void a round until the season's objection window ends; the round's budget returns to the pool, and its packs count as passed for everyone. Only the Guardian can call that void. The operator can still void a round another way, set out in the next answer.

Could the people writing the questions mine against me?

They hold the answers before anyone else, and no cryptography removes that: a wallet they control could score in full, and the chain cannot tell it from yours. What the chain does stop: the operator's three registered addresses can never answer, and every key is committed before the round, so no label changes after anyone has answered. What it does not stop: the operator can void a whole round. A round in which an opened pack never has its key opened is void when it closes. Each pack carries a sealed copy of its key that any rig can open, but the chain cannot check that copy is right, so an operator that seals a wrong one and withholds the key can void the round after seeing who answered. A void lowers nobody's pass rate, but nobody is paid for that round, so the operator can pick a round in which one wallet holds most of the weight. Rigs refuse a pack with no usable sealed key, and the replay tool names a wrong one afterwards. The season's ban list is the operator's choice too: a listed wallet proven on chain before the season's first round cannot answer that season. Said plainly, the residual risk is real.