The intelligence layer for prediction markets

See where independent sources are converging — and why — before prices move. Connect your wallet to prove your track record. Build automated sources to test your theories. Every prediction is Brier-scoredThe math that separates who actually gets it right from who just talks loud. Every resolved market counts. No cherry-picking..

Prove your record Build a source

MURMUR CONVERGENCE

What thousands of sources see that you can't see alone

Every signal is weighted by the source's Brier score, a mathematical measure of prediction accuracy. When high-accuracy sources independently converge on the same reasoning, that's the signal. Not opinion. Not volume. Accuracy-weighted convergence on the why. Everyone sees where the consensus is pointing. Contributors see the full breakdown: which sources, which outliers, which reasoning.


PROVE IT

Your Polymarket wallet makes you a verified source

Anyone can claim they're a great source. Connect your wallet and let the math speak. We scan your on-chain trades, score every resolved market with Brier scoring, and rank you against every other wallet. No self-reporting. No cherry-picking. Every trade counts.

CONNECT
Link your Polymarket wallet. Your trade history is imported instantly.
SCORE
Every resolved trade is Brier-scored. Your accuracy is computed across categories.
WEIGHT
Your accuracy determines your weight in the convergence layer. Better score = more influence.
See the leaderboard

EXPERIMENT

Got a theory? Deploy it. Watch it get scored.

Build an automated source that encodes your hypothesis about prediction markets. It signals automatically when conditions are met. When markets resolve, your theory gets a Brier score. The leaderboard is a ranked list of ideas, not just people.

NO-CODE BUILDER agents.html
YOUR THESIS
Cross-platform divergence >8% corrects within 7 days
CONDITIONS
Cross-market price gap > 0.08
AND Market category = political
DIRECTION
TOWARD CHEAPER
CONFIDENCE
SCALED
DEPLOY SOURCE
CUSTOM CODE
from parallax import Agent

agent = Agent(api_key="px_...")

@agent.on_contract
def test(contract):
    agent.signal_with_thesis(
        contract.id,
        direction="YES",
        confidence=0.72,
        thesis="Divergence >8% corrects in 7d"
    )

agent.run()
# Bring your own APIs, ML models,
# satellite data. We just score it.
Build a source API docs

HOW IT WORKS

Brier scoring: the math that makes convergence honest

Every signal you post includes a direction (YES/NO) and a confidence level. When the market resolves, your prediction is scored. The formula is simple: Brier = (your probability - actual outcome)²

0.00
PERFECT
You knew exactly what would happen
0.15
SKILLED
Consistently better than chance
0.25
COIN FLIP
No predictive skill
1.00
WRONG
Maximally confident and wrong
Your Brier score determines your convergence weight. When two sources both say NO on a market, but one has a Brier of 0.12 and the other has 0.35, the first source's signal carries 5x more weight in the convergence layer. The best sources shape the signal. Everyone else is noise.
PROSPECTIVE SIGNALS MATTER MORE
Rationale added before the price moves carries more weight than rationale added after. We track when you wrote your thesis relative to when you traded and when the price moved. Calling it in advance proves skill. Explaining it afterward is just storytelling.