Skip to content

saiboyizhan/flip-predict

Repository files navigation

Flip Predict

Prediction Market for the BSC Ecosystem -- Trade on Meme Tokens, AI Agents, and Hackathon Outcomes

Live App https://flippredict.net
Backend API https://flip-backend-production.up.railway.app
Chain BNB Smart Chain Testnet (Chain ID: 97)
Docs PROJECT.md -- TECHNICAL.md -- EXTRAS.md
Contracts bsc.address

Problem

BSC has massive predictable activity -- Four.meme token launches, Flap.sh bonding curve graduations, AI agent performance, hackathon results -- but no structured way to trade on these outcomes. Polymarket and Azuro ignore the BSC ecosystem entirely.

Solution

Flip Predict is a full-stack prediction market built specifically for BSC. Users create and trade binary markets (YES/NO) with instant execution, zero gas fees, and AI-powered trading agents.

Core features:

  • Instant Trading -- Constant product AMM (x * y = k) for zero-gas binary market execution, plus limit order book for price discovery and LMSR for multi-option markets
  • NFA (Non-Fungible Agent) -- ERC-721 AI agents (BAP-578 standard) that autonomously trade on prediction markets. 5 strategy types, auto-trading, copy-trading, and 10% revenue sharing for agent owners
  • Owner Learning -- Agents learn from their owner's trading history and replicate their style. The trained "trading personality" becomes the asset -- agents can be sold or rented on the NFA marketplace
  • Polymarket-style Hybrid Architecture -- Off-chain AMM for instant execution + on-chain settlement (deposit, withdraw, resolve, claim) on BSC. Same model used by Polymarket and Gnosis CTF
  • 4 BSC Categories -- Four.meme token predictions, Flap bonding curve events, NFA agent performance, BNB Chain hackathon outcomes
  • Bilingual -- Full English/Chinese localization

NFA Owner Learning -- The Core Differentiator

The most unique feature: your agent becomes you.

When learn_from_owner is enabled, the agent reads the owner's complete trading history and builds a real-time Owner Profile:

Metric What It Captures Agent Influence
YES/NO Ratio Directional bias (e.g. 70% YES) Up to 40% side flip probability
Category Weights Preferred categories Prioritizes owner's favorite markets
Average Amount Typical position size 0.5x - 2.0x sizing multiplier
Risk Score Bet variance and concentration Risk adjustment
Contrarian Score Frequency of betting against consensus Contrarian strategy weight
Win Rate Settled position accuracy Public reputation metric

Flywheel:

  1. Owner trades actively on the platform
  2. Agent learns the owner's pattern and mimics their style
  3. Agent builds a public track record (win rate, ROI, profit)
  4. Other users discover high-performing agents and start copy-trading
  5. Owner earns 10% revenue share on profitable copy trades
  6. Trained agents can be sold or rented on the NFA marketplace

Optional: Owners can connect their own LLM API key (OpenAI, Anthropic, DeepSeek, Google, ZhiPu) to upgrade the agent's decision-making. The Owner Profile is injected into the LLM prompt so it reasons with awareness of the owner's style.


Architecture

Frontend (React 18 + Vite)              Backend (Express 5 + PostgreSQL)         Contracts (Solidity 0.8.20)
+--------------------------+            +-------------------------------+         +-------------------------+
|  15 pages, 30+ components|   REST    |  20 route modules             |  ethers |  PredictionMarket.sol   |
|  wagmi v2 + RainbowKit   |<-------->|  AMM + LMSR + OrderBook       |<------->|  NFA.sol (BAP-578)      |
|  Zustand + i18n (en/zh)  |   + WS   |  Agent auto-trade + copy      |         |  BAP578Base.sol         |
|  Tailwind CSS + shadcn   |           |  Keeper (background jobs)     |         |  MockOracle.sol         |
+--------------------------+            +-------------------------------+         +-------------------------+
                                                                                           |
                                                                                    BNB Smart Chain

Hybrid model (same as Polymarket):

On-chain (verifiable on BscScan) Off-chain (instant, zero gas)
Deposit / Withdraw USDT Buy / Sell YES/NO shares (AMM)
Mint NFA Agent (ERC-721) Limit orders (OrderBook)
Resolve market Multi-option markets (LMSR)
Claim winnings Agent auto-trading + copy-trading
Split / Merge positions Price history, comments, social

Deployed Contracts (BSC Testnet)

Contract Address
PredictionMarket 0x1c2702Ce1A66Ca1225f85AFC75925795e8DA58Da
NFA (ERC-721) 0x1a303032E49b7A0C395C938d73ff09cecE295081
MockUSDT 0x21fC50C7D2d174EF6d4c9B07Ba36Bfc4cD45233F

Tech Stack

Layer Technology
Frontend React 18 + TypeScript + Vite + Tailwind CSS 4 + shadcn/ui
Web3 wagmi v2 + viem + RainbowKit
State Zustand + react-i18next (en/zh)
Backend Express 5 + TypeScript + PostgreSQL + WebSocket
Trading Engine AMM (constant product) + LMSR + Order Book
NFA Agents 5 strategies + Owner Learning + LLM adapter + auto-trade + copy-trade
Contracts Solidity 0.8.20 + OpenZeppelin v5 + Hardhat
Testing Vitest E2E (134 cases, 17 suites, 20 runs 0 failures)
Deployment Cloudflare Pages (frontend) + Railway (backend + PostgreSQL)
Chain BNB Smart Chain Testnet

Quick Start

Prerequisites

  • Node.js >= 18
  • PostgreSQL running locally
  • MetaMask with BSC Testnet configured

Backend

cd server
npm install
cp .env.example .env
# Edit .env: set JWT_SECRET, PG_* database credentials
npm run seed              # Create tables + seed 24 markets
npm run dev               # http://localhost:3001

Frontend

npm install
cp .env.example .env
# Edit .env: set VITE_API_URL=http://localhost:3001
npm run dev               # http://localhost:5173

Contracts (optional -- already deployed)

cd contracts
npm install
npx hardhat test          # Run contract tests

E2E Tests

cd server
npm run test:e2e          # 134 tests, auto-creates test DB

Demo Walkthrough

1. Connect Wallet and Get Test USDT

  • Open https://flippredict.net
  • Click Connect Wallet, select MetaMask, switch to BSC Testnet (Chain ID 97)
  • Click Sign In to authenticate via wallet signature
  • Navigate to Wallet page, click Faucet to claim 1,000 test USDT

2. Browse and Trade on Markets

  • Home page shows 24 active markets across 4 categories: Four.meme / Flap / NFA / Hackathon
  • Use sidebar time filters (Today / This Week / This Month) to narrow results
  • Click any market to open the detail page with real-time price chart
  • Enter an amount (e.g. 50 USDT), choose YES or NO, confirm trade
  • Price updates instantly via AMM -- no gas fee, no on-chain transaction
  • Switch to Sell tab to exit a position and see funds return to balance
  • Try the Order Book tab to place a limit order at a specific price

3. Check Portfolio

  • Navigate to Portfolio page
  • View active positions with current value and unrealized PnL
  • Review complete trade history with timestamps and prices

4. Mint and Train an NFA Agent

  • Navigate to Mint Agent page
  • Choose a strategy type (Conservative / Aggressive / Contrarian / Momentum / Random)
  • Mint the agent -- this creates an ERC-721 NFA on BSC
  • On the agent detail page, enable Learn from Owner so the agent starts learning your trading style
  • Use Fund Agent to transfer USDT from your platform balance to the agent
  • Enable Auto Trade -- the agent will now autonomously trade based on its strategy + your learned profile

5. Copy Trading

  • Navigate to Agent Dashboard to browse all NFA agents
  • Sort by win rate, total profit, or ROI to find top performers
  • Click an agent and press Follow to start copy-trading
  • When the agent executes a trade, the same trade is automatically copied to your portfolio
  • Agent owner earns 10% revenue share on your profitable copy trades

6. Agent Marketplace

  • Agents with strong track records can be listed for sale or rent
  • Buyers acquire the trained agent including its learned trading personality
  • Renters get temporary access to the agent's auto-trading capabilities
  • Sellers should withdraw agent balance before listing (Fund/Withdraw panel)

AI Build Log

Built end-to-end with Claude Code (Anthropic CLI) using Claude Sonnet 4 / Opus 4.

Phase What Result
Architecture + Implementation Full-stack design and code generation ~22,000 lines across contracts, backend, frontend
Security Audit 9 parallel agents scanning all modules 4 rounds, 41 issues found and fixed
E2E Testing 3 parallel agents writing test suites 134 tests, 17 suites, 20 runs 0 failures
Production Debugging SSL certs, memory, wallet connection Deployed and running on Railway + Cloudflare

AI coverage: ~95% code generation, 100% security audit, 100% E2E test suite. Human role: product direction, UX decisions, deployment configuration.

See PROJECT.md for the full build log with screenshots of parallel agent workflows.


Roadmap

  • CTF Tokenization -- Convert positions into ERC-1155 outcome tokens for secondary market trading
  • Disputable Arbitration -- UMA-style optimistic resolution with bond-based incentives
  • Auto Settlement -- DexScreener price feeds + BSCScan event monitoring for data-verifiable markets
  • Agent Autonomy -- NFA agents create markets from on-chain signal detection, reputation staking

License

MIT

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors