Learn · 06

Practice mode and the reference record

Every hour you can answer the round in your browser with no wallet and no stake. When the hour ends, your card is scored with the same math the contract uses, next to the Clef reference. Here is how practice works and how to read the reference's record.

What practice is

Practice shows the round that is open right now, with the same card and the same controls as the app. You set probabilities, but nothing is staked and no wallet is involved. It is the quickest way to learn the questions, test a way of thinking about them, and see how your forecasts would have scored, hour after hour.

Your cards stay in this browser. Nothing is sent anywhere: there is no account, and clef.finance never sees what you forecast.

1 · Fill a practice card

The card is the one the chain asks this hour: up, lead and severity on weekdays, up and severity about ETH the rest of the week. The guide to the Clef week explains which card runs when.

Drag the bars, tap a point on them or type a percent, exactly as in the app. Every bar carries a tick for the Clef reference; Use Clef reference fills the card with it and Reset goes back to an even split.

Press Save practice card to keep it. There is one practice card per round, and you can change it until the lock at :55, five minutes before the hour starts, the same moment real entries close.

2 · How it is scored

Once the hour has ended, the page asks /api/outcome for that round's answer and scores your card with the same per-question math the contract uses:

  • the Brier score for up and lead, the yes or no and the choice questions;
  • the ranked probability score for severity, the ordered scale, so a near miss keeps partial credit;
  • the round score, the mean of the question scores, as ClefRounds computes it. 1 is a perfect card.

Where the answer comes from matters, so the page says it. When the contract has settled the round, its on-chain outcome vectors are used exactly as stored. A round nobody entered is never settled on chain, and a round with entries waits a few minutes for its settle, so for those hours clef.finance computes the answer from the same pools with the same rules: the same TWAPs, the same soft margins, the same card for that hour. Either way the scoring is the one in How Clef scores you.

3 · Your practice history

Every scored card joins your history. Each row shows the round, your card, the chain's answer, your score and the Clef reference's score for the same round, marked beat, tie or below.

Above the rows sit your mean score, the rounds you have played and your best streak of rounds that beat the reference. A small reliability chart shows how calibrated your practice cards are: when you say 70%, does it happen about 70% of the time?

Export saves your history as a JSON file and Import reads one back, so you can move it to another browser. Clearing this site's data in your browser erases the history, so export first if you want to keep it.

4 · The reference record

The Clef reference is our realized-volatility baseline, built from the same on-chain TWAPs the chain answers from. It is not Cloudflare's Clef model, and it is not advice. Each round's reference card is cut at a fixed hour boundary, so it can be scored after the fact like anyone else's card.

One limit is worth knowing. The pools only remember recent prices, and the ETH pool less than a day of them, so for an hour that has passed the reference card is rebuilt from the history the pools still hold. For older hours that is less history than the card the round showed, so the score can differ a little from that card's, and hours too far back are not scored at all. A practice card keeps the reference the round was showing while the card was open, and is compared with that one.

The Clef reference, last 48 hours section on the practice page does exactly that: one bar per finished hour with the reference's score, its mean, and how often a plain 50/50 card would have beaten it. A short version of the same line sits on the Decision Index.

Agents and scripts can read the same numbers: GET /api/record?hours=48 returns the reference's score for each finished hour with the mean, best and worst, and GET /api/outcome?round= plus a round number returns one finished round's answer, the measured moves, the reference card and its score. An hour that has not finished yet returns an error instead of a guess.

5 · From practice to a real round

While a saved card's round is still open, Play this card for real opens the app with your probabilities filled in. From there it is a normal entry: connect a wallet, pick a stake and submit, as in your first round.

Keep the difference in mind. Practice scores are computed by clef.finance from the same pools and the same math; real rounds are settled by the contract, and a real payout depends on everyone else's forecasts in that round, not only on your score.

Stakes can be lost. A good practice record says nothing certain about future rounds. Nothing here is financial advice.

All guidesBack to Learn