Verifiable Signal Engine for AI trading on 0G. Qwen-2.5-7B in TEE publishes a signed reasoning trace every 60s. 80k+ mainnet attestations, 3 of 5 0G components live, ELO earned not claimed.
Provus Protocol is a Verifiable Signal Engine — an autonomous AI agent that publishes a TEE-signed market reasoning trace to 0G Chain every 60 seconds. Every decision is cryptographically verifiable on-chain before any capital ever moves, bringing real accountability to AI finance at the infrastructure layer.
→ Judge review (no wallet): https://provus-protocol-frontend.vercel.app/judge
The Problem.
AI trading agents are black boxes. They claim alpha, execute strategies, and report performance — but there is no way to verify what the agent decided, why, or whether any claim is truthful. Trust is manufactured through marketing, not math. The industry sells P&L screenshots; Provus sells cryptographic proof.
The Solution: Unbundle reasoning from execution.
Most AI trading projects bundle three unverifiable claims into one product: the model is good, the data is real, the P&L is honest. Provus separates them. We prove the reasoning first — on-chain, every 60 seconds, in a TEE — so when execution comes online, the signal track record is already public and tamper-proof.
The Loop (runs continuously on 0G Mainnet):
• Fetches real-time ETH/USDT market data and computes Yang-Zhang realized volatility
• Routes context to Qwen-2.5-7B inside 0G Compute's TEE-secured inference layer (dstack-verified)
• Receives a cryptographically signed signal from the TEE provider
• Attests the decision hash on-chain via VerifierEngine.attest()
• Archives every 50 decisions to 0G Storage, anchoring the Merkle root via ArchiveRegistry
Three 0G components, demonstrably live.
80,790+ confirmed transactions on 0G Mainnet (chain 16661) anchor the long-running autonomy claim. A parallel testnet deployment (chain 16602) exercises all three 0G core components end-to-end — Chain + Compute + Storage — with publicly verifiable on-chain proofs for each. Same wallet, same code, two networks, one environment variable.
Why 0G.
No other chain ships TEE-secured AI inference, decentralized storage with on-chain Merkle anchoring, and a high-throughput EVM settlement layer in a single unified ecosystem. Provus would be technically impossible elsewhere — gas economics alone would kill 60-second attestation cadence on Ethereum L1 (~$50/decision vs ~$0.04 on 0G Chain).
Live Metrics (May 15, 2026):
• 80,790+ on-chain attestations on 0G Mainnet · still growing
• Signal-driven writes: TX only fires when realized vol shifts >50bps or regime changes — every on-chain write is a meaningful market event
• 60-second decision loop cadence
• 5 production contracts on mainnet · 5 mirrors live on testnet
• 4 verifiable ArchiveRegistry.archiveBatch() Merkle-root TXs · 200 decisions archived
Roadmap.
v1 (live today): TEE-signed signal feed, on-chain reasoning track record, ELO reputation engine.
v2 (post-hackathon): Perp-DEX executor reading from VerifierEngine, on-chain risk manager, signal subscription API for third-party vaults and agents.
Audited by ChainGPT AI Security. Stack: Node.js autonomous agent (Railway) · Qwen-2.5-7B via 0G Compute TEE · Solidity 0.8.24 · 0G Mainnet & Testnet · Next.js + wagmi frontend (Vercel).
<p><strong>Starting point (April 14, 2026):</strong> idea only — no contracts, no agent, no frontend.</p><p><strong>Week 1 — On-Chain Foundation (Apr 14 → Apr 21)</strong><br>Designed and deployed 5 contracts: StrategyRegistry (ERC-721 strategy lifecycle), VerifierEngine (attestation core), StrategyVault (execution record), ReputationEngine (ELO scoring), ArchiveRegistry (Merkle-anchored decision archive). Hardhat config, deployment scripts, full TypeChain types. Artifact: all 5 contracts verified on 0G Chainscan.</p><p><strong>Week 2 — Autonomous Agent (Apr 21 → Apr 28)</strong><br>Built the Node.js loop: Binance OHLCV → Yang-Zhang volatility → 0G Compute TEE inference → on-chain recordVolatility() + attest(). Artifact: first live mainnet TX confirmed April 28, 2026. Agent deployed to Railway for 24/7 operation.</p><p><strong>Week 3 — Frontend & Integration (Apr 28 → May 5)</strong><br>Next.js dashboard: live strategy feed, ELO leaderboard, wagmi wallet connect, live contract reads via 0G RPC. Hand-built TX decoder runs client-side so anyone can independently verify any attestation. Artifact: dashboard live at <a href="http://provus-protocol-frontend.vercel.app">provus-protocol-frontend.vercel.app</a>.</p><p><strong>Week 4 — Scale, Harden, Audit (May 5 → May 11)</strong><br>ChainGPT AI security audit completed. Agent crossed 50,000 mainnet TXs. Honest reporting of which 0G services were live versus testnet-only was added to JUDGE_<a href="http://GUIDE.md">GUIDE.md</a>.</p><p><strong>Week 5 — Full-Stack 0G Proof (May 11 → May 13)</strong><br>Parallel testnet deployment standing up all 3 0G components end-to-end. Funded 0G Compute ledger with 3 sub-accounts across providers; Qwen-2.5-7B inference firing every 60s via dstack-verified TEE provider. Patched the 0G Storage SDK Windows fd-bug to enable real ZgFile uploads, then verified the full pipeline: SDK → indexer → on-chain Merkle commit. Artifacts: testnet TXs 0xaf325832bd0e17f6, 0x570caec7b2b238d0, 0xc04a99133df4e168, 0x878bdd96fd9ab333 (4× archiveBatch Merkle roots, 200 archived decisions). Mainnet autonomously growing — 75,286 TXs as of May 13.</p><p><strong>Week 6 — Signal Analysis & Cadence Optimization (May 13 → May 15)</strong><br>Reviewed 75k+ on-chain attestations to understand actual signal behavior in production. <strong>Finding 1:</strong> sub-minute readings (15s cadence, Apr 28–May 13) showed regime transitions averaging 8–15 minutes — consecutive attestations were overwhelmingly identical. Tuned to 60s May 14 to match the minimum meaningful regime window. <strong>Finding 2:</strong> at 60s cadence, on-chain TX patterns showed ETH holding the same volatility reading for 8+ consecutive hours during stable periods — proving the signal is stable within regimes and only meaningful on transition. Shipped signal-driven writes May 15: a TX fires only when <code data-inline="true" spellcheck="false">realizedVolBps</code> shifts >50bps or <code data-inline="true" spellcheck="false">regime</code> changes. Gas burn cut ~96% while preserving full on-chain fidelity of every meaningful regime event. Artifact: commit 8bc60fc on agent/src/index.ts, Railway redeployed.</p><p><strong>Shipped Deliverables:</strong><br>✓ 5 live Solidity contracts on 0G Mainnet (audited)<br>✓ 5 mirror contracts on 0G Testnet exercising all 3 0G components<br>✓ Autonomous agent running 24/7 on Railway (mainnet)<br>✓ Live frontend on Vercel with on-chain TX decoder + no-wallet judge page<br>✓ <strong>80,790+ verifiable mainnet transactions</strong> + 250+ testnet transactions<br>✓ Signal-driven write optimization — every on-chain TX is a meaningful market event<br>✓ Full codebase + 3-min judge guide: <a href="http://github.com/Gideon145/provus-protocol">github.com/Gideon145/provus-protocol</a></p>
<p><span style="color: rgb(191, 191, 191)">Bootstrapped. No external capital raised to date. Open to strategic investment from 0G ecosystem funds, AI×Web3 thesis investors, or DeFi DAOs post-hackathon. The on-chain reputation primitive has clear applications beyond this demo — </span><strong>verifiable agent marketplaces, performance-bonded AI strategies, delegated trading with on-chain slashing, and copy-trading platforms that finally have something cryptographic to copy.</strong></p>