ReliefMesh

Judge guide

What ReliefMesh does, what is real, and where to see it.

Why this matters

ReliefMesh does not just create a procurement plan. It determines whether that plan can survive disruption, shows exactly what breaks it, calculates the cheapest resilient alternative, and automatically rebuilds the plan when reality changes.

Live disaster → supplier quote → Plan V1 → continuity analysis → resilient alternative → real supplier change → V1 stale → approval revoked → Plan V2 → fresh approval → PO.

Walk the loop on a real mission

Nothing here is pre-seeded. Create a coordinator account, start a mission from a live disaster, and invite suppliers by email; this page then checks each step against that mission's stored records. The status page shows every integration's live check right now.

What is live?

  • Disasters: NASA EONET events, NWS alerts, NASA FIRMS fire detections, OpenFEMA declarations
  • Supplier screening: OFAC SDN list, SAM.gov registration, USAspending awards, openFDA recalls
  • Supplier discovery: Apify Google Places runs (leads only, never auto-trusted)
  • Suppliers, quotes and quote changes: submitted by real supplier accounts (USER_SUBMITTED)
  • Email: RFQs and POs sent from procurement@reliefmesh.us (DKIM/SPF/DMARC pass), plus each organization's own Gmail, Calendar, Drive and Slack
  • Plans, approvals, purchase orders and the audit chain: stored in MongoDB Atlas

What is sandbox?

  • CRS identity verification of supplier signers: CRS sandbox with CRS test identities only (SANDBOX)
  • x402 evidence payments: test USDC on Solana devnet (DEVNET TEST PAYMENT); no real money, no supplier is paid

What is hypothetical?

  • Every Continuity Engine result: N-1 supplier-loss scenarios, price / delay / capacity breakpoints and the resilient alternative (HYPOTHETICAL STRESS TEST)
  • These never change a quote, plan, approval or order, and are marked STALE the moment a real input changes

What uses AI?

  • Quote drafting: Gemini turns a pasted supplier quote into a draft the supplier reviews; every value is cited from the text
  • The Fetch.ai agent in ASI:One: explains stored results in plain language and starts allowed actions

What does NOT use AI?

  • Policy eligibility (deterministic rules over cited evidence)
  • The optimizer and every continuity number (MILP re-solves on exact quote copies)
  • Approval, purchase orders and payments (humans only; the agent can never approve, order or pay)