Everything you need to recompute your own score.
Each row is a parameter, its value, the range it may ever take, and why it is that value. A parameter with no reason attached is a parameter nobody can argue with, so every row carries one.
one source
Where the money comes from.
There is no other source. No treasury, no emission, no reserve, no release curve, no buy-back. A round releases up to a fifth of the pool and the rest rolls over, so quiet trading shrinks the pool slowly; a round pays nothing only when the pool is empty. Your share is your credit times your stake, divided by that same quantity summed over everyone who scored — which means it depends on other people, and no honest page can predict it.
credit
The score, term by term.
parameters
The pack
| Parameter | Value | Range | Why |
|---|---|---|---|
| K | 128 items | Frozen for the season | One Jev request can answer the whole pack in parallel; the shipped rig sends it as two halves and uses the first complete reply for each. A chat model cannot emit 128 labels fast enough in one reply, so it fans the pack out much wider (the contenders were measured at 8 ways), pays for the shared policy card on every branch, and waits for its slowest branch. |
| Options per item | 5 to 6 | At most 6 | Guessing costs about 102 wrong answers a pack. The pass rate for a guesser is zero. |
| Pack interval | 10 minutes | 1 to 36 packs a round | The operator chooses how many packs to commit to a round, from the season's floor up to 36. When the round opens, the vault runs the smallest of three counts: that commitment, the count the round's own budget calls for, and the count the question budget can pay for. Anyone can recompute them from the round's events. |
| Packs per wallet | Every pack of the round | Fixed | 144 packs a day at full frequency. Every wallet answers the same pack. |
parameters
The clock
| Parameter | Value | Range | Why |
|---|---|---|---|
| The answer window, in blocks | 2 blocks after the pack is opened | 1 to 20; frozen for the season | A BNB Chain block is 0.45 s. Jev answers a pack in about 0.3 s, so a rig that has the beacon when the pack opens lands in the next block with one to spare. |
| The answer window, in seconds | Before the beacon's instant + 4 s | 1 to 30; frozen | A second clock nobody can move by holding the opening back. A block stamped at the limit or later is refused. |
| Who opens a pack | Anyone holding the beacon | Fixed | The contract verifies the beacon signature itself. Rigs open packs themselves, so an operator cannot delay a pack to shrink its window. The two blocks count from the opening, so a pack opened late stays answerable later, up to the beacon's instant + 4 s. |
| Where the clock starts | The beacon's own round | Identical for every wallet | The randomness beacon (drand quicknet) publishes a round every 3 seconds; nobody knows its signature more than one round early. |
| Region | Close to the model's API | Your choice | The rig's self-test decides. Measured: us-west-2 and Tokyo pass; a home connection in Asia did not. |
parameters
The score
| Parameter | Value | Range | Why |
|---|---|---|---|
| Pack credit | max(0, 2 × (E_max + 1) − (2 × wrong + blank)) | Fixed formula | A blank costs half a wrong answer. Only packs within the error budget score at all. |
| E_max, whole-pack error points | 2 | Frozen; never below 2 | Two wrong answers in 128 still score; three do not. |
| Season pass-rate multiplier | clamp((r − 0.60) / 0.25, 0, 1)², r = scored packs ÷ keyed packs | Formula fixed; floor up to 0.70, width at least 0.05; frozen for the season | Absences count. A void round, however it was voided, counts as passed for everyone, so a void never lowers any wallet's rate. It does take the round's whole award from everyone who scored in it. |
| Payout weight | credit × multiplier × your stake in the round | Linear, no per-wallet cap | Linear on stake makes splitting a stake across wallets exactly neutral. |
| Stake floor | 0.1% of supply, staked in the vault | Fixed at launch | Only tokens locked in the vault count. A stake is locked from your first answer in a round until that round closes, so one token can never count twice. |
parameters
The money
| Parameter | Value | Range | Why |
|---|---|---|---|
| Where rewards come from | The 2% buy/sell tax, in BNB | The only source | There is no treasury, no emission, no reserve and no release curve. If nobody trades, the vault collects nothing and the round pays nothing. |
| Question share | 5% of the tax income | At most 8%, fixed at launch | Pays for writing the questions. The factory commission is separate: 0 by default, at most 3%. |
| Released per round | 20% of the BNB pool | Fixed by the factory | Each six-hour round releases a fifth of the pool to the wallets that answered it, by credit and stake. Whatever a round does not pay goes back to the pool. Twelve rounds release about 93% of what the pool held when the season began; the rest rolls over. |
| Round / season | 6 hours / 12 rounds = 72 hours | Fixed at launch | 72 hours is a hard ceiling because a purpose-built classifier catches up in one to three days. |
| Objection window | 24 hours after the season's last round closes | 24 to 72 hours, fixed at launch | Until it ends, Flap's Guardian can void any round, which returns its budget to the pool; only the Guardian can call that. A round is also void when it closes if a pack of it was opened and its key never was; the operator can bring that about by withholding a key. |
| Claiming | claimRounds(from, to); never expires | Fixed | Each wallet's award for each round is readable on chain. No keeper pushes anything to you, and there is nothing to miss. |
the defence
What is deliberately not on this page.
A few parameters are the defence itself. Published, they would be a specification for beating the accuracy gate, so they are absent here on purpose rather than by oversight:
- The screening rule and the panel behind it
- These are the defence. Published, they are a specification for beating the accuracy gate.
Everything a miner needs in order to recompute their own score is above, and every round publishes enough to redo the arithmetic independently. Nothing private is needed to check that you were paid correctly.