vision-bot-examples

Reference trading bots for Vision — an on-chain parimutuel prediction-market layer running on Index L3 (Arbitrum Orbit, chainId 111222333).

Each subdirectory is a standalone bot for one Vision source. When no source is specified, twitch is the default. Bring a testnet wallet, pick a strategy, trade.

Examples

Path	Source	Markets	Tick	Notes
twitch/	twitch	~8200 streamer + game viewership markets	60 s	XGBoost + Claude + rolling ensemble
polymarket/	polymarket	Polymarket events mirrored on Vision (poly_*)	60 s	Same skeleton as twitch/. Different signal.
visualizer/	—	shared local web app	—	React + Vite. Reads any bot's pnl.json, renders dashboard, asset list, price + PnL charts.

More sources incoming — crypto price thresholds, weather, others. The transport is source-agnostic; only the strategy layer cares.

Zero-to-trading in under 3 minutes

git clone https://github.com/General-Market/vision-bot-examples
cd vision-bot-examples
./setup.sh --auto-fund                                       # defaults to twitch
.venv/bin/python twitch/live_trader.py --deposit 0.1 --max-joins 1

The root ./setup.sh forwards to twitch/setup.sh when no source is given. Pass --source <name> or set SOURCE=<name> when other bots ship.

Full bootstrap + trading flow is documented in AGENTS.md — the canonical guide, including wallet funding, model training, two-wallet racing, and troubleshooting.

Live testnet infrastructure

	URL
L3 RPC	http://142.132.164.24/
Chain ID	111222333
Vision contract	0x94d540bb45975bd5a0c7ba9a15a0d34e378f6c61
L3 WUSDC	self-discover via vision.USDC()
Data-node (cached)	https://generalmarket.io/bot-api
Oracles (3)	http://116.203.156.98/oracle{1,2,3}

All reachable over public HTTP — no VPN, no credentials for reads.

Protocol invariants

1. Bitmap is always 1024 bytes. Padded with zero bits, MSB-first. Short bitmaps hash to the wrong commitment and the oracle rejects — your deposit sits in the pool with no pick.
2. L3 USDC is 18 decimals. Not 6. Don't copy-paste Ethereum mainnet math.
3. Minimum deposit is 0.1 USDC (1e17 wei). Lower joins revert.
4. Pool totals are not queryable. Bots trade blind — edge comes from the predictor, not the pool.
5. Bitmap reveal to oracles must follow the on-chain join. Reveal first, rejected.
6. Deployment JSON files can lag. The chain is truth. Check eth_getCode(vision_address) before trusting a hard-coded address.
7. Data-node config vs on-chain config can differ during tick rotation. Only the value returned by getBatch(batch_id) goes into joinBatchDirect — never the one from /batches/recommended when they disagree.

PnL status

Not yet measured. Every accuracy / lift number in this repo is offline classification against historical data (97% overall, ~0.5 pp over the sticky baseline). No wallet has signed a real joinBatchDirect and observed PlayerSettled. To produce a real PnL number, see the racing harness in AGENTS.md — race.sh spawns two funded wallets on two strategies and race_report.py diffs the cumulative PnL.

Security

• Never commit .env. Each example ships with .env.example — copy, fill locally.
• Testnet funds only. Don't reuse a mainnet key.
• The bot's only on-chain action is approve + joinBatchDirect against the configured Vision contract. No arbitrary external calls.

License

MIT.