FAQ

It is the model's probability that buyers other than the creator, the deployer and any declared tax-exempt wallets put 10% of the curve's graduation target in within two hours of the launch block. A 16% means: of launches that look like this one, about one in six attracted that much outside money in the first two hours. Next to it, "0.46× typical" is that probability over the base rate.

No. The score says whether outside money is likely to arrive, not what the price does. Plenty of launches attract buyers and still go to zero. Nothing on the desk is advice.

Open it and read How this score was made. The most common reasons, in order: the creator bought nothing of their own supply, the creator fee is above 3%, no socials on chain, fees routed to the deployer rather than a third party, paired with ETH rather than a stock. Each line shows the rate for launches in that group. The score is not a judgment of you; it is a count of what happened to launches that looked like yours.

A language model reads the name, ticker, description, quote asset and the tweet the token links to, and judges the idea: is the tweet a real beat, is the ticker the joke, does the pairing land, does it fit what runs on this chain, is it original. It is a judgment, not a measurement, and it is shown separately from the probability for that reason. Its bands are checked against outcomes on the record tab, so if taste turns out not to predict anything, the desk will say so.

Because the on-chain score only sees seven signals, and two launches with the same signals get the same number. The story score is there to separate them. The on-chain model is deliberately simple: a more elaborate one would be harder to check without being more honest.

They are recomputed from the logs every twenty seconds as windows close. AUC and the top-decile hit rate settle as the sample grows. The calibration table can show a gap between what the model said and what happened; that gap is the model's over-confidence in today's regime, published rather than hidden. Refitting calibration on this desk's own outcomes is a planned step, and it will not rescore old predictions.

Paste its address into the search box on the desk and press Enter. The agent reads it on chain and scores it on demand. That score is marked "not on the record" because it was not written before the outcome, so it never counts toward the track record.

No. It holds no wallet and signs nothing. It reads the chain and writes two files.

The chain, through two public RPC endpoints: the factory's launch events, each token's info and curve state, and the curve's buys and sells for the outcome. The linked tweet is resolved through X's public syndication endpoint. Nothing comes from a paid feed or an aggregator, and the desk names its source next to every number.

That the desk could not read that value from a source it trusts, so it prints nothing rather than a guess. Tax-exempt wallets read no signal when the launch transaction could not be decoded; the story reads no signal when the judge is off or has not reached that launch yet.

Yes. The prediction and outcome logs are plain JSON lines. The scoreboard recomputes every figure from them, and its tests enforce that every outcome is timestamped after the prediction it scores. Run npm test and read /api/scoreboard; the docs walk through it.

The chain reads, the model, the scoreboard and the flags are GRAID, an MIT-licensed agent by 0xfokki that has scored well over two hundred thousand launches on this chain. The tweet resolver, the story score, the API and the desk are ours.

It makes the desk fast and deep instead of delayed and shallow. Everyone sees the record; holders see the feed live rather than 30 minutes late, the story score and the judge's verdict on every launch, can score any token on demand, and get alerts when a launch clears a score they set. Proving you hold it is a wallet signature, not a transaction. The creator fee from the token's own launch pays the judge's bill; that is the whole mechanism, and it is written up in the docs.

No. The public view needs nothing and shows everything on the record, 30 minutes behind. A wallet is only for the live feed, the story scores, on-demand scoring and alerts. Until the token exists the gate is open and everyone sees the live view.