Pyth Insight
The first platform to publicly measure whether Pyth's confidence intervals are statistically accurate — and an AI analyst that explains what the data means.
Powered by Pyth Price Feeds · Pyth Benchmarks · Pyth Entropy
Feb 2026 — $388K oracle exploit: @ploutos_money drained after using a BTC/USD feed to price USDC collateral. Pyth's CI would have flagged the mismatch instantly. See the full breakdown →
CI Calibration Analysis
LiveWe ran months of Pyth Benchmarks data through a statistical calibration test. Does the ±1σ CI actually capture 68% of future price moves? The answer varies by asset.
Live Price Feeds
LiveEvery price feed on Pyth Hermes — crypto, forex, metals & more — streamed live at 1-second intervals. Each card shows live confidence intervals and oracle certainty in real time.
AI Oracle Analyst
Ask anything about Pyth's price data or calibration results in plain English. Injected with live prices, CI widths, and anomaly signals from all connected Pyth feeds in real time.
Volatility Intelligence
7-day realized volatility rankings and CI width trend analysis across all tracked assets — is the oracle becoming more or less certain about prices over time?
Oracle Challenge
A provably fair prediction game powered by Pyth Entropy. Random historical price moments are selected on-chain — can you beat the oracle's confidence band?
Learn Pyth
Interactive guides: how the pull oracle works, what confidence intervals really mean, how Entropy generates verifiable randomness, and how to use Benchmarks data.
Pyth publishes a confidence interval with every price update. But is it actually calibrated?
A perfectly calibrated oracle's ±1σ band should capture exactly 68.3% of future price moves. We ran the numbers across months of historical data — the results are more interesting than you'd expect.
The same CI that answers this question also prevents oracle exploits like the $388K @ploutos_money attack — where a mismatched feed was priced at $80,000 instead of $1.00.
See the calibration results →