PolicyMint
Autonomous trading agents have historically chosen between two failure modes: execute freely with no guardrails (high-performance, high-risk), or sit behind policy engines that never actually trade (safe but inert). This gap is quantifiable — H1 2025 crypto theft alone reached $3.01 billion. PolicyMint closes both sides of the problem: it trades actively while proving every risk decision on-chain through the ERC-8004 Validation Registry, making institutional-grade safe execution achievable for the first time. How the Policy Engine Works Before every trade execution, PolicyMint runs a mandatory pre-flight evaluation against three core policy rules: Spend Cap — Per-transaction USD limit capped at ≤ $450 to stay below the RiskRouter's $500 hard ceiling Venue Allowlist — Only whitelisted trading venues (e.g., kraken-spot, uniswap-v3) are permitted to receive orders Daily Loss Budget — Cumulative drawdown tracked in real time; agent halts when the daily loss threshold (≤ 4% of capital) is breached Every trade intent is structured as an EIP-712 typed-data payload, cryptographically signed with the agent wallet, and bound to Ethereum Sepolia (chainId 11155111) via EIP-155. If the policy engine returns allow, the signed intent becomes the execution proof submitted to the ERC-8004 RiskRouter. If the engine returns block, the Kraken CLI call never fires and a human-readable block reason is logged.
Category tags:
"1. Application of Technology: 4.25 / 5 Justification: Very strong TypeScript backend engineering. They successfully implemented EIP-712 typed-data signing and EIP-155 bindings to Ethereum Sepolia. The evaluate routing and mapping architecture (evaluate.service.ts, trade-intent.mapper.ts) suggests a highly scalable node.js application suited for enterprise use. 2. Presentation: 4.5 / 5 Justification: Excellent. The PolicyMint_PRD files and the pitch deck present a highly professional narrative targeting institutional-grade safety. Sighting the "$3.01 billion in H1 2025 crypto theft" is a fantastic hook that immediately justifies why their hard ceiling risk router needs to exist. 3. Business Value: 4.5 / 5 Justification: Extremely high. Just like Wisdom Pivot and TreasuryShield, this project zeros in on the single biggest hurdle to AI trading adoption: fear of catastrophic loss. A compliance team can actually understand and approve a policy engine that hard-blocks trades over $450 mechanically. 4. Originality: 3.5 / 5 Justification: While beautifully executed, the underlying concept is standard enterprise compliance (whitelist venues, cap daily loss, limit transaction size). It is very similar to TreasuryShield's approach, though PolicyMint goes further by tying the policy approval directly into an EIP-712 cryptographically signed intent. ⚖️ Pros & Cons Pros: On-Chain Binding: Structuring the trade intent as an EIP-712 payload bound to Ethereum Sepolia means the agent's signature acts as a tamper-proof "ticket" to execute the trade, merging Web2 compliance engines with Web3 trust. Hard Math Caps: Enforcing a max $450 per-transaction limit to stay safely beneath the broader $500 hard ceiling is practically exactly how real quant funds throttle their API keys. Excellent Documentation: Providing extensive PRDs (Product Requirement Documents) inside the repository shows a highly mature product management lifecycle. Cons: Missing the "Brain": Like multiple other projects in this cohort, PolicyMint solves the execution risk half of the puzzle but doesn't actually provide a novel AI predictor or signal generator to find the trades in the first place. It is a safeguard waiting for a signal. Single Layer Focus: Focuses heavily on basic numerical caps (spend limits, loss limits) rather than complex market-regime detection or dynamic volatility scaling seen in projects like Quasar or kraken-interface."
Sanem Avcil
For Innovators & Creators