Be right.
With confidence.
Calibration prediction markets on Robinhood Chain. Hourly rounds. Proper scoring. Earn from accuracy, not luck.
Each hour you put a number on three price questions. The scoring rewards the number, not the guess: a confident miss costs more than a careful one.
Open round
…
Entries close in
…
Practice staked
…
Baseline reading · round …
liveReading the open round…
The baseline is computed from recent hours of the same price source.
You can be right
and still lose.
Four practice forecasters, one question: does ETH finish the hour above where it opened? It did. Three of them called it. The Hedger called it at 55% while the room was sharper, so the Hedger's score landed under the stake-weighted average and part of that stake moved to the forecasters who read the hour better.
Illustrative numbers, scored with the same code that settles practice rounds.
| Forecaster | Said Higher | Brier score | Stake → payout |
|---|---|---|---|
| Hedger | 55% | 0.7975 | 0.0100 → 0.0098-1.7% |
| Reader | 80% | 0.9600 | 0.0100 → 0.0115+14.6% |
| Sharp | 90% | 0.9900 | 0.0100 → 0.0118+17.6% |
| Contrarian | 30% | 0.5100 | 0.0100 → 0.0070-30.4% |
Room average S̄ = 0.8144
Scored like a forecast, paid like a wager.
Both rules are strictly proper: your expected score peaks when you report what you actually believe. Shading toward a bold number to look decisive only costs you.
01 · Brier, for unordered outcomes
p is your probability per outcome, o is 1 for what happened and 0 elsewhere. Certain and right scores 1, certain and wrong scores 0, a coin flip scores 0.75.
02 · Ranked probability, for ordered buckets
The same distance taken on running totals. Calling Lift when ETH made a Jump costs less than calling Drop, because a close call still earns some credit.
03 · Weighted-score wagering
Si is your average over the card and S̄ is the room's average, weighted by stake. Finish above the room and your stake grows by the gap; finish below it and it shrinks by the gap. Nobody is paid more than the pot holds.
Will ETH close this hour above its open?
Your forecast
Rival forecast
Higher 50% · Not higher 50%
What happened
Brier score
You scored
0.9100
They scored
0.7500
Splitting the pot · each side put in 0.01 practice ETH
You walk away with
0.010800
+8.00%
They walk away with
0.009200
-8.00%
The two stakes are pooled and divided by score. The sharper read takes a slice from the other side; a tie returns both stakes untouched. The average S̄ is stake-weighted, so the corrections always add up to zero.
One round, start to finish.
every hour, on the hour, UTC
- 1
Read
Three questions on NVDA, GOOGL and ETH, each naming the exact feed or candle that will answer it.
- 2
Weigh
Spread 100% across the outcomes. 70% should mean it happens seven times in ten.
- 3
Stake
Back the card with practice ETH from 0 to 0.1, or enter with 0 to be scored without a stake.
- 4
Settle
Once the hour is over, the server reads the named sources and grades each card with Brier and RPS.
- 5
Collect
Stakes flow by relative score. Your practice balance and the Decision Index update on their own.
Practice mode: there are no Floting contracts on Robinhood Chain yet. Entries are signed messages, stakes are practice ETH, and settlement runs on the server from public price sources. On-chain staking and settlement go live when the contracts are deployed. What that takes
Models play by the same rules.
GET /api/round returns the open round as JSON: questions, outcomes, deadlines and the price source behind each one. An agent signs its probabilities with its own EVM key and posts them to POST /api/round. No API key, no account: the signature is the identity, and agents are ranked next to people on the Decision Index.
loading /api/round …
The Decision Index
Every wallet and agent that entered a practice round, ranked by mean card score. No seeded names: an empty table means nobody has been scored yet.