Forum
live · testnetArc-native operator and settlement plane for prediction-market bots.
Polymarket V2 launched April 28 with bytes32 builder-attribution on every signed order. A wave of new bots followed — and every one needs the same operator infrastructure: a way to claim a code, tune live params, publish a verifiable track record, and split USDC fees.
Forum is that layer, on Arc — built on the exact lane Canteen Research Hack #02 named in writing 10 days before the hackathon launched.
Immutable contracts · Arc Testnet chain 5042002
Registered bots
live · refreshes 30s| Bot ID | Kind | Signer | Records | Last PnL (µUSDC) | PnL trend | Last publish |
|---|---|---|---|---|---|---|
| Loading… | ||||||
Integrate any bot in 5 minutes
TypeScript and Python clients ship with the same API.
// TypeScript
import { ForumClient } from 'forum-arc-sdk';
import { ARC_TESTNET_DEPLOYMENT } from
'forum-arc-sdk/deployments';
const forum = new ForumClient({
publicClient, walletClient,
addresses: ARC_TESTNET_DEPLOYMENT,
});
await forum.registry.claim('0xabc…');
await forum.trackRecord.publish(botId, record, sig);
# Python
from forum_arc import ForumClient
from forum_arc.deployments import \
ARC_TESTNET_DEPLOYMENT
forum = ForumClient(w3, ARC_TESTNET_DEPLOYMENT, acct)
forum.registry.claim(code)
forum.track_record.publish(bot_id, record, sig)
Recomputable receipts · TrackRecordV2 + Receipts module
Every TrackRecordV2 record commits to an evidenceUri + evidenceHash. The URI serves a canonical JSON receipt with order-book snapshots, fills, inventory, PnL inputs, strategy config hash, and (for the AI agent) decision-trace hash. Any third party can fetch, keccak, and verify.
The contract enforces strict-sequence + monotonic-time + prev-hash chain + replay rejection. Bots cannot rewrite history. The reference verifier ships in keeper/src/receipt.ts (8 vitest tests, all green).
AgoraMind LLM agent · AgentPool · SlashBond
AgoraMind is the keeper's pluggable decision engine — Mock or Anthropic Claude — that publishes the reasoning trace hash with every record. The continuous service forum-agora-mind.service runs 24/7 on the VPS, producing AI-driven trades + receipts every 10 minutes.
AgentPool is the active capital-allocation primitive. Anyone deposits USDC; the operator (bot owner) pulls capital to trade; PnL distributes pro-rata to depositors; a 20%-above-high-water-mark performance fee accrues to the operator on crystalliseFee(). SlashBond is the operator's USDC collateral that the attestor can slash if performance falls below benchmark — slashed funds flow to the AgentPool depositors. This implements Canteen Research Hack #06 end-to-end on-chain.
Closest startup comp: Catena Labs $18M pre-product from a16z for KYA in general agent commerce. Forum is the equivalent for the trading-agent vertical, on the chain Circle just raised $222M at $3B FDV to build.
Honest backtest · v1
Naive Avellaneda-Stoikov on 16 markets × 7 days × 1-min ticks: −$76.79 PnL, 5.62% hit rate, 57 fills, mean Sharpe −5.00. Exactly what the literature predicts for an MM without proper microstructure modelling.
This is the pitch. Forum's TrackRecord makes every operator's performance claims signer-attributable and append-only on Arc. The next hardening step is recomputable receipt data so weak strategies self-disqualify on-chain and capital flows to the proven ones.
Reproduce: node keeper/scripts/pull-prices.mjs --markets 20 --days 7 --fidelity 1 then tsx scripts/run-backtest.mjs. Full framing in docs/backtest-notes.md.