Arcane

Arcane is an agentic legal prediction-market exchange that converts public corporate litigation into tradable, auditable probability markets. The platform lets users view ongoing public-company legal cases, review docket activity and legal catalysts, and trade YES/NO contracts on objectively resolvable outcomes such as whether a court grants an injunction, whether a company settles before trial, or whether an appeal is affirmed by a specific deadline.

The project combines a working frontend, FastAPI backend, SQL database, agent research pipeline, LMSR automated market maker, wallet-connected trading flow, and Arc-based USDC settlement architecture. Autonomous agents discover public cases, summarize docket activity, extract market-moving events, compare precedent, estimate probabilities, and support resolution workflows. These agents can be paid through nanopayments for research tasks such as docket summaries, legal-risk scores, probability forecasts, liquidity quotes, and resolution checks.

Arcane uses Arc as the stablecoin-native settlement layer. Prediction-market collateral is denominated in USDC, escrowed through smart-contract logic, and conditionally released to winning YES/NO holders after public legal outcomes are verified. Ledger adds a human-in-the-loop approval layer for high-risk autonomous actions such as large trades, spending-limit changes, and market resolution. The project also includes an optional Aqua collateral-yield extension, allowing eligible locked USDC behind long-duration legal-event positions to earn yield while users wait for court outcomes.

The goal is to create programmable legal-risk infrastructure: a market layer that prices public litigation risk, an agent layer that improves legal-information discovery, and a stablecoin settlement layer that makes legal-event markets transparent, auditable, and capital-efficient.

Arcane is built on a Python FastAPI backend and a vanilla HTML/JS frontend, powered by an ensemble of 9 specialized AI agents. The agents (CaseScout, PrecedentAgent, ProbabilityAgent, etc.) autonomously ingest live litigation data from the CourtListener API, analyze dockets, and output quantitative probability forecasts. These forecasts are priced using a custom Logarithmic Market Scoring Rule (LMSR) AMM engine built directly into the backend, allowing users to trade against the AI's predictions in real-time. For settlement, we wrote and deployed ArcaneSettlement.sol to the Arc Testnet using Foundry. The backend acts as a relayer, taking user-signed EIP-712 intents and executing buy() calls on-chain to escrow Circle Testnet USDC, enforcing a strict 24-hour dispute window for market resolutions. To secure high-stakes actions, we integrated the Ledger DMK to build a dynamic risk policy engine. Trades under $100 use standard MetaMask signatures, but any trade over $100—and all admin resolutions—strictly require Ledger hardware approval. The backend dynamically generates EIP-712 clear-signing artifacts so the user's Ledger device displays the exact Market, Side, and USDC Amount, preventing blind-signing attacks. The most notable implementation is our machine-to-machine economy utilizing the Circle Agent Stack. We built a custom x402 nanopayment middleware that intercepts internal agent API calls. When a user requests research, they authorize a payment via EIP-3009. The Orchestrator agent then programmatically batch-settles micro-fees in USDC through the Circle Gateway to pay sub-agents for specific tasks—like paying $0.005 to parse a PDF or $0.01 to run a forecast. This creates a fully functional, self-sustaining agentic economy where AI researchers are compensated deterministically for their compute.