The problem AmberMarket solves

AmberMarket – Every signal is a market

The Problem It Solves

Every signal is a market. Every junction is an asset.

AmberMarket transforms real-world traffic data into on-chain prediction markets grounded in something measurable, verifiable, and live: how many vehicles cross a junction during a single green-light window.

No abstract outcomes. No human referees. Just a CV oracle watching actual CCTV streams and settling bets automatically.

What you can do with AmberMarket:

• Stake on live traffic — Place $AMBER bets on real-time junction vehicle counts, settled by a YOLO-powered computer vision oracle reading live camera feeds. Transparent, automated, tamper-resistant.
• Own a junction — Mint an AmberJunctionNFT to claim ownership of a real-world intersection and earn royalties every time others bet on your junction.
• Get a Web3 identity — Claim a free .ambermarket.eth ENS subname and appear across leaderboards and live markets with a human-readable name.
• Trust the machine, not the middleman — Smart contracts hold all funds; the oracle provides the count; winners are paid by tolerance (e.g. +-15% or exact). Zero manual intervention.

The stack in one line: Real-time CV data + on-chain prediction markets + automated oracle settlement + ENS identity, all wired together in a single hackathon build.

Challenges we ran into

Challenges We Ran Into

1. Bridging Python CV and a Node backend
The vehicle-counting pipeline runs on Python (YOLO + OpenCV), while the rest of the stack is Node.js + Socket.io. Rather than splitting into two long-running services, we spawn the CV pipeline as a subprocess and consume JSON lines from stdout, keeping the server single-process while still streaming live counts to clients. Getting clean shutdown, timeouts, and graceful fallbacks (a "fake mode" for when YOLO is not installed) took a few iterations to get right.

2. ENS subnames via Namespace
We wanted free .ambermarket.eth identities with no gas cost for users. The Namespace offchain manager requires domain-scoped API keys, something we discovered the hard way after hitting repeated 401s with a default key. Switching to domainApiKeys: { [ENS_PARENT_NAME]: NAMESPACE_API_KEY } fixed it. We also added rate-limiting (10-minute claim cooldown) and proper UI differentiation between "subname not found" and "API unavailable" for a cleaner demo experience.

3. Taming real CCTV streams
Caltrans D3 HLS streams are not framed for lane-level counting. They capture wide scenes with sidewalks, adjacent lanes, and irrelevant zones. We introduced a per-junction viewport config (e.g. 0.30,0.40,0.70,0.90) so the CV model only counts within a defined region of the frame. This meaningfully reduced false positives and made counts accurate enough to actually settle bets on.

4. Phase loop and frontend sync
The server drives a strict market cycle: RED_OPEN > GREEN_COUNTING > SETTLING, broadcast over Socket.io. Keeping the client's countdown timers, bet-open/closed states, and settlement modals in sync with backend phase (and on-chain state when the oracle is active) required a single authoritative source of truth and clearly named events, so the UI never showed "Place Bet" after the market had already locked.