Aura — The Internet Aura Market

Trade the trend line of the internet's mood. Aura turns real online attention into a live, AI-scored on-chain index — then lets anyone trade prediction markets on where each name's aura goes next.

▶ Live app · 🎤 Pitch video · 🎬 Technical demo · ⛓ Contracts on Monad testnet · 🧩 Sponsor deep dives · 𝕏 @auramarkets_

Every market is one handle — a person, project, coin or brand — and one time window. The handle's index, its aura, is scored by an AI from the handle's own scoped sources (news RSS, X, Reddit, YouTube, GitHub, per-user OAuth connections) and written on-chain. Each window asks one question: does this handle's aura close higher or lower?

---

Contents

• Try it in two minutes
• Why it's different
• How it works
• Built with
• On-chain
• Known limitations & what's next
• Repository layout
• Where the data comes from
• The aura → market decision path
• Run it locally · Environment variables · Verification & CI
• Chainlink CRE workflows · Redeploy runbook · The adapter contract · Docs & conventions

---

Try it in two minutes

Nothing to install — it runs on Monad testnet with test money.

1. Open app.auramarkets.world. The home page is the Atlas: every handle is a villager in its category's district. Bigger villagers have more aura; 🔥 marks the biggest movers. Drag to pan, scroll or pinch to zoom. How it works walks the whole loop in six steps.
2. Sign in with your email. Privy creates an embedded wallet for you — no extension, no seed phrase. (Your first transaction needs a little testnet MON for gas: faucet.monad.xyz.)
3. Click a villager to open its market: the aura chart (hover a point to read the AI's one-line why it moved), the live window, YES/NO prices, the order book and the handle's source feed.
4. Tap "Get 1,000 test USDC" in the trade ticket, then buy YES or NO. Each trade is a real, user-signed transaction on Monad testnet.
5. Watch the window close. It freezes at the end, settles on the aura it closed at, and winning shares claim 1 USDC each.

Why it's different

• The asset is momentum, not an event. Prediction markets usually price "will X happen by date Y" — slow to resolve, easy to dispute. Aura prices the continuous trend of attention, in windows from 15 minutes to a day.
• The model advises; code decides. The AI returns a direction, a magnitude in a ±5–20% band and a one-line rationale. Code scales that by an evidence weight (credibility × volume × reach) and clamps it, so a confident take on one weak article moves the index far less than a broad, high-reach story.
• Every move is explained and replayable. The memo is written on-chain with the move. Each tick is hashed over the exact items the model saw, its raw reply and the applied percent: re-fetch the evidence, re-run the prompt, compare the hash.
• No observation, no settlement. A window with no aura observation is left to VOID and refunds traders, instead of resolving on a stale number.

How it works

  scoped sources ──▶ aura engine ──▶ on-chain aura ──▶ market windows
  (RSS · APIs ·     (AI scores      (Handle.updateAura   (MarketMaker LMSR,
   OAuth feeds)      text + reach)   + AuraHistory)       YES/NO per slot)
        ▲                                   │
        └────────── indexer (Envio) ────────┘  every entity the UI + oracle read

1. Ingest. Per-handle adapters pull real items with real reach (author, likes, replies, views). Each source is a JSON config stored on-chain in Handle.sources[]. Adapters fail closed without credentials and never invent a metric.
2. Score. The aura engine builds the signal (normalized volume + engagement + item texts), asks the AI for direction + magnitude + memo, scales it by evidence, writes the new aura on-chain, and pushes it over WebSocket.
3. Trade. The window engine opens a market window per handle per slot. Prices come from an LMSR market maker plus an order book, backed by the auraLP pool; early believers can stake $AURA on a handle and earn a cut of every window that settles.
4. Settle. At windowEnd the window resolves from the aura series and the next slot opens in the same tick, so a handle always has a live window.

The full step-by-step, with what is and isn't deterministic, is the decision path below.

This is the beginning of the README. Read the rest at the original. Read the full README ↗