v1.0.0: contracts v3 live on Robinhood Chain, platform, packages, film and pitch

CargoFlow v1.0.0: evidence-gated USDG working capital, live on Robinhood Chain

Built for the Arbitrum Open House Singapore Buildathon. Testnet only, not audited; testnet USDG has no value.

Live: web app · API · API reference · contracts and services · MCP for Claude: https://cargoflow-mcp.adoranto737.workers.dev/mcp · Telegram: @Cargo_FlowBot

Contracts v3 (Robinhood Chain Testnet 46630, all source-verified)

• FinancingController 0x06DaF9462eCF2434ED0314a005Bf762cCDEd7Fe1, ReceivableVault 0x167783DB96E27f36f78C8E5F6f1575b0c45a8151, CoverPool 0x4e4f09Da01f466275b586b0cc32613a90b1B69e5, EvidenceRegistry 0x3930f06dC9Deb7b7587AD5a04B05350CaACc7BA2, PolicyEngine 0x74be1E30bEeDc004447F0427Dd6EBC62CCDabA25, ShipmentRegistry 0x2B9E2B70b6fF48DaD9847944bbEcE16c9f4396F3, Groth16Verifier 0xF00eE4c686cE4EaC0B161dEe757e19A1C600d0f6, DeviceRegistry 0xD176E4e02f97F12462e68014F92B2A13A664552e, EBLRegistry 0x72278056f6537e4F3437b96898BB439ab7BF68A9, CargoFlowAccess 0x6b334b4C73c27CB297470140c86311075408b050; settlement in Paxos USDG 0x7E955252E15c84f5768B83c41a71F9eba181802F.
• New: electronic bill of lading (ERC-721) bound to facilities under documents against payment, device registry with per-epoch source devices, parametric cover, cancel path, an OpenZeppelin Pausable brake on new risk only, place-based milestones, humidity and shock limits, default cover.
• Arbitrum Sepolia extensions: Fhenix ConfidentialInvoiceTerms 0x5c1C12448D27c2E8519c1E471078Bf42E685D207, GMX GMXHedgeVault 0xE0E90F3E57e3a040AD99FE4384bE96Dd97002f74.

Platform

• Evidence engine, zero-knowledge recovery prepared automatically (the exporter gets a "proof ready" alert and signs once), alerts on Telegram, email, Slack and webhooks plus an in-app notification centre.
• Device trust (Ed25519, P-256 secure elements with X.509, WebAuthn passkeys), GS1 EPCIS 2.0 export and import, fee guidance, OpenAPI 3.1, RPC failover (QuickNode, Alchemy, public).
• ZeroDev passkey smart accounts: a buyer signs in with a passkey and pays from a Kernel account (live on testnet, CF-SG-VAX-0401).
• Redesigned web app with the film's characters and a GSAP scroll story; 18 end-to-end and 296 unit tests.

Packages

• npm: @cargoflow/sdk, @cargoflow/mcp (25 tools, also hosted for claude.ai), @cargoflow/gateway
• PyPI: cargoflow

Assets attached

• CargoFlow-v2-1080p.mp4: the 5:30 product film (story, explainer, live website, claude.ai, Telegram alerts, passkey payment)
• CargoFlow-v2.srt: its captions
• CargoFlow-pitch.pdf: the 16-slide funding deck

Measured

366 contract tests · 25 circuit · 296 frontend unit · 18 end-to-end · SDK 92 · MCP 27 · gateway 36 · Python 25 · Fhenix 19 · GMX 20.

v0.8.0: testnet prototype

CargoFlow v0.8.0: testnet prototype

Evidence-gated working capital for physical trade finance, live on Robinhood Chain Testnet. Testnet only: not audited, testnet USDG has no value, and the Groth16 trusted setup is single-party. The dashboard (P6) is not built yet.

Highlights

• Full story run on the public chain. Fund, two milestones, thermal anomaly, pause, Groth16 recovery proof verified on-chain, remaining milestones, delivery, settlement: 22 transactions in 116 s with real USDG, the AI monitor on, every transaction linked in docs/runbooks/testnet.md. Run at 1/2000 scale (20 USDG facility, 50 USDG invoice) because the faucet drip was 100 USDG; the full-size 40,000 / 100,000 run is covered by the end-to-end test.
• Seven contracts deployed and source-verified on the Robinhood testnet explorer, with a separate wallet per role.
• Advisory AI monitor (Groq). It can only make an outcome stricter (pause or ask for more proof), through a key that holds no other role. It never sees telemetry-derived text, and the deterministic policy gate decides alone whenever the model is absent, slow or wrong.
• Self-healing backend. A reconciler resends missing commits, pauses and releases with backoff and a retry cap; cargoflow demo re-runs the whole story against any deployment.
• Stylus (optional, Arbitrum Sepolia). A Rust evidence engine held to the same vectors as the Go and Solidity versions. Measured on identical calldata: 8 readings cost 40,239 gas in Solidity vs 29,783 (17,378 cached) in Stylus; 128 readings 611,945 vs 31,511 (19,106 cached). One run, untuned Solidity, see the caveats in stylus/README.md.

Verification

174 contract tests (fuzz, invariants I1-I8, real-proof integration), 20 Go packages with -race including an end-to-end test through the running service, 25 circuit tests, Rust tests against the shared vectors, Slither triaged. CI green. make check, make bench, make slither.

Known limits

Single-party ZK setup, simulated telemetry, one backend instance per database, uncalibrated score weights, no frontend yet. See the README's "Honest limits".