CEXs lie with fair prices. Here's how to catch it in real time

RamzesVII — Wed, 25 Mar 2026 23:39:41 +0000

TL;DR: Pyth gets prices directly from institutional market makers (Citadel, Jane Street, Jump). When Binance/Gate/OKX diverges from that benchmark, even by 0.05%, it's measurable. Built a 3-screen dashboard to track it live across 30 assets. Link at bottom

Every CEX claims to show 'the real price'. But price is not a single number - it's a probability distribution.

Pyth Network doesn't scrape exchange prices. It receives data directly from first-party market makers - the firms that set prices, not react to them. That makes it a useful benchmark: when a CEX diverges from Pyth, it's not Pyth that's wrong.

The confidence interval is the interesting part

Every Pyth price comes with a CI - a range that reflects oracle uncertainty:

Tight CI (±0.01%) = high agreement, deep liquidity
Wide CI (±14%) = low publisher count, market closed, or disagreement

I use CI as a signal threshold:

Deviation inside CI = noise
Deviation outside CI = potential signal (arb, liquidation hunt, latency)

Concrete example: NVDA's CI is ±0.1% during NYSE hours. When the market closes it jumps to ±14%. The oracle literally tells you when to trust the price.

Pyth Radar - What I built

Pyth Radar is a real-time dashboard that measures how far 30 assets - crypto, forex, commodities & equities - deviate from the Pyth benchmark

Target audience: algo & HFT traders, arbitrageurs, DeFi risk managers, quants

Three screens:

1. Deviation Index - 0-100 stress score across 30 assets (market breadth, each asset equal weight). Radar chart + category cards toggle

2. Deviation Heatmap - grid of 30 assets with two independent signals. Fill = how far CEX is from Pyth. Border = CI stress vs historical baseline. Click any asset to drill down

3. Pyth Delta - per-asset view. Pyth benchmark vs live prices from Binance, Gate.io, OKX, Bybit. Composite price, delta % per exchange, divergence log, sparklines

Exchange coverage: 4 exchanges, 6 WebSocket connections.

What's next

Post-market and overnight feeds for equities - Pyth already has separate feeds. NVDA's post-market feed has 6 publishers and ±0.04% CI vs ±14% on the regular feed after close.
Telegram alerts when deviation crosses CI
More assets, exchanges & price feed types (eg. Overnight / Postmarket / etc. - for equities)

Links

Live: https://pyth-radar.vercel.app

GitHub: https://github.com/RamzesVII/pyth-radar

Pyth Network: https://www.pyth.network/

X Article: https://x.com/r_ladik/status/2036924687012552846

Built for Pyth Community Hackathon 2026: https://dev-forum.pyth.network/t/pyth-community-hackathon/548

DEV Community: RamzesVII

CEXs lie with fair prices. Here's how to catch it in real time