Real-time oracle intelligence platform + paper prediction game

PilotSB — Sat, 28 Mar 2026 14:09:20 +0000

`DeltaScope is a real-time oracle intelligence platform + paper prediction game that monitors the gap between Pyth Network oracle prices and Hyperliquid DEX mark prices. It uses triple-source Pyth ingestion (dual Hermes WebSocket + REST polling) with Pyth Lazer ready for sub-50ms updates, an AI assistant with 6 Pyth-powered tools for querying 1,930+ price feeds, and a Predict & Win game where users bet on price direction using live Pyth oracle data — with confidence-aware settlement that refunds when oracle data is unreliable.

What It Does

Oracle prices and DEX mark prices should match — but they don't. The gap reveals liquidation risk, funding rate mechanics, and infrastructure health that most traders are blind to.

DeltaScope gives traders real-time visibility into:

Oracle-mark spread — Real-time discrepancy between Pyth oracle prices and Hyperliquid mark prices across 8 major assets
Pyth publish delay analytics — How stale is your oracle data right now? Median publish delay tracked in a 10-min rolling buffer
Infrastructure latency monitoring — Pyth Oracle Delay, Hyperliquid REST API latency, WebSocket delivery, and overall health score
Predict & Win — Paper prediction game where users predict price direction (UP/DOWN) using live Pyth prices, with alarm-based settlement, streak bonuses, and a global leaderboard
AI-powered analysis — Natural language queries across 1,930+ Pyth price feeds with 6 structured tools

This is the layer beneath the prices that nobody else shows.

Link

Live: https://deltascope.site

GitHub: https://github.com/0xPilotSB/deltascope

Pyth Features Used

✅ Price Feeds (off-chain) — Triple-source Pyth Hermes ingestion (dual WebSocket + REST polling) with freshness dedup by publishTime
✅ Price Feeds (Lazer-ready) — Pyth Pro/Lazer integration for sub-50ms updates via 3 redundant endpoints when PYTH_PRO_TOKEN is set
✅ Confidence Intervals — Displayed per-asset alongside oracle prices; used in prediction settlement (refund when confidence is unreliable)
✅ Historical Prices — AI chat tool queries Pyth Hermes /v2/updates/price/{timestamp} for backtesting
✅ TWAP — AI chat tool queries /v2/updates/twap/latest for time-weighted average prices (1-600s windows)
✅ Price Feed Search — AI tool searches 1,930+ Pyth feeds by symbol, name, or asset type via /v2/price_feeds
❌ Entropy (randomness) — Not used

Features

1. Dashboard (`/`)

Real-time price table for 8 major assets (BTC, ETH, SOL, HYPE, ARB, DOGE, AVAX, LINK) showing:

Pyth oracle price vs Hyperliquid mark price side-by-side
Oracle discrepancy badges (color-coded by severity)
24h change, annualized funding rates, open interest, 24h volume
Aggregate stats: total volume ($2.5B+), total OI ($4.2B+), avg funding rate
HIP-3 ecosystem overview (Hyperliquid's permissionless perp DEXs)

2. Ticker Analysis (`/analysis`)

Top 20 Hyperliquid leaderboard trader positions per asset
Long/short breakdown: counts, sizes, average entries, liquidation ranges
Expandable position details: leverage, unrealized PnL, margin used
Sort by position size, trader count, or PnL

3. Predict & Win (`/predict`) — NEW

Paper prediction game powered by live Pyth oracle prices:

Binary predictions: UP or DOWN on any of 8 assets
Two time windows: 1 minute (fast) and 5 minutes (standard)
Points economy: 1,000 starting balance, wager 10/25/50/100 per prediction
Alarm-based settlement: Predictions resolve automatically via Durable Object alarms against live Pyth prices
Streak bonuses: 3+ consecutive wins earn +10% bonus per extra win
Global leaderboard: Compete by points, win rate, and streaks
Safety mechanics: Refund on oracle unavailability or zero price movement
Zero real money — Validates the prediction market concept before on-chain deployment on HyperEVM

4. Latency Monitor (`/latency`)

Infrastructure intelligence inspired by HyperLatency:

Pyth Oracle Delay — Median publish delay across all tracked feeds
Hyperliquid REST API — Round-trip time to api.hyperliquid.xyz
WebSocket Delivery — Edge DO → browser delivery latency
Overall Health Score — Composite 0-100 score
Multi-line TradingView chart with 10-min rolling history (120 samples)
Source health table with P50/P95/MIN/MAX percentiles

5. AI Chat Assistant

Natural language queries powered by 6 structured Pyth tools:

Tool	What It Does	Pyth API Used
`searchPriceFeeds`	Search 1,930+ feeds by symbol/name	`/v2/price_feeds`
`getLatestPrices`	Real-time prices with confidence	`/v2/updates/price/latest`
`getHistoricalPrice`	Price at any historical timestamp	`/v2/updates/price/{timestamp}`
`getTwap`	Time-weighted average (1-600s)	`/v2/updates/twap/latest`
`getHyperliquidData`	Cross-reference with HL perps	Hyperliquid REST API
`analyzePriceFeed`	Full analysis package	Multiple endpoints combined

6. Developer API (`/developers`)

REST + WebSocket endpoints documented for integration:

GET /api/prices — Current merged Pyth + Hyperliquid state
GET /api/latency — Latency history + source health
GET /api/hip3 — HIP-3 ecosystem data
WS /ws/prices — Real-time stream (16ms broadcast coalescing)

Screenshots / Media

Dashboard — Real-time Oracle vs DEX Prices

Predict & Win — Paper Prediction Game

Binary UP/DOWN predictions on live Pyth prices with points economy, countdown timers, streak bonuses, and global leaderboard

Latency Monitor — Infrastructure Intelligence

Live Pyth oracle delay (1826ms Hermes → client), Hyperliquid API latency (85ms), WebSocket delivery, and overall health score with TradingView charts

AI Chat — 6 Pyth-Powered Tools

Natural language queries: "Compare BTC and ETH funding rates" → AI fetches real-time Pyth prices, queries Hyperliquid funding/OI, computes discrepancies, and presents structured analysis

Architecture

Predict & Win Settlement Flow

`plaintext
User clicks UP/DOWN → PredictionGame DO

Validate wager (sufficient points)
Fetch entry price from PriceAggregator DO (live Pyth price)
Deduct points atomically
Store prediction in SQLite
Schedule DO alarm (duration + 1s)

Alarm fires → Settlement

Fetch exit price from PriceAggregator DO
Compare direction (up/down vs actual movement)
Calculate payout (wager + streak bonus)
Update points, wins/losses, streak in SQLite
Reschedule alarm if more pending predictions `

Tech Stack

Framework/Language: React Router 7 (TypeScript), Vite, Tailwind CSS 4, shadcn/ui

Blockchain/Oracle: Pyth Network (Hermes WebSocket + REST + Lazer-ready), Hyperliquid L1 (WebSocket + REST)

Agent Framework: Vercel AI SDK + Cloudflare Workers AI (6 structured Pyth tools with codemode)

Deployment: Cloudflare Workers + 4 Durable Objects (edge-deployed, 24/7 uptime via DO alarms, Smart Placement)

Full Backend Stack

Component	Technology	Purpose
Runtime	Cloudflare Workers	Edge-deployed serverless runtime with Smart Placement
PriceAggregator DO	Durable Object (stateful)	4 upstream connections (2× Pyth WS + 1× HL WS + 1× REST), 16ms broadcast coalescing, 24/7 alarm keep-alive
PredictionGame DO	Durable Object (SQLite)	Paper prediction market: users, predictions, leaderboard, alarm-based settlement
Chat DO	Durable Object (AI SDK)	Streaming LLM with 6 structured Pyth tools
ChatSessionsDO	Durable Object (SQLite)	Session index per anonymous user
Oracle Ingestion	Pyth Hermes (dual WS + REST)	Triple-source freshness dedup by publishTime
Oracle Ingestion (Pro)	Pyth Lazer (optional)	3 redundant WS endpoints, real_time channel, sub-50ms
DEX Data	Hyperliquid API	allMids WebSocket + metaAndAssetCtxs REST (3s poll)

Full Frontend Stack

Component	Technology
Framework	React Router 7 (SSR) with typed loaders/actions
State	Zustand (WebSocket connection + tick aggregation)
Charts	TradingView Lightweight Charts (candlestick + multi-line latency)
UI	shadcn/ui + Tailwind CSS 4 (dark theme, responsive)
Typography	Space Grotesk (self-hosted via @fontsource)
AI Chat	@cloudflare/ai-chat + agents SDK (lazy-mounted popup)

Key Design Decisions

Single global PriceAggregator — One DO instance holds all upstream connections. Zero coordination overhead, guaranteed consistency.
16ms broadcast coalescing — queueMicrotask() for immediate dispatch or setTimeout() for remaining window. Prevents WebSocket storm from multiple upstream sources.
Incremental snapshot caching — Only dirtyAssets get JSON recomputed per broadcast. Unchanged assets reuse cached objects.
24/7 keep-alive via DO Alarms — 25-second alarm cycle keeps all upstream connections alive even with zero clients. Eliminates cold-start delays.
Alarm-based prediction settlement — PredictionGame DO schedules alarms per prediction duration. Settlement fetches live prices from PriceAggregator via DO-to-DO RPC. Atomic SQLite updates prevent race conditions.
Triple Pyth source — Dual Hermes WS (main + beta) + REST polling. Freshest publishTime wins per asset. ~10-15% lower oracle delay than single WS.`

DEV Community: PilotSB