SportiQue

THE PROBLEM: WHO NEEDS DATA, AND WHY For example, AI teams may need datasets to train models; labs and drug companies may need records to study treatment outcomes. Buyers pay for agreed use of data they lack, for a defined purpose and period.

Each deal repeats field and period matching, provenance and quality checks, permission and payment. This makes small exchanges difficult. The owner decides whether to accept.

WHAT WE BUILT AgentRail gives buyers and providers agents to coordinate each deal. A buyer's agent identifies gaps in its database, queries a shared offer book anchored by a Merkle root, compares candidates and negotiates within the owner's budget. Code checks signed quality profiles and committed samples before payment.

The accepted bundle is delivered encrypted. Successful XRPL EscrowFinish publishes opening key K and pays the seller. Code checks data; XRPL enforces escrow conditions. Recipient sealing restricts opening; resale follows the license.

The LLM proposes; code rechecks and can only narrow permissions. Owner-signed spending limits are checked before payment. A multi-signed treasury prevents the agent from increasing its own budget.

Two separate synthetic runs: the rule-based market completed 109 paid trades, 80 without an owner decision; six mismatched offers were refused before payment and four post-opening failures became refund claims. The LLM replay contains 120 trades, 85 settled. These are separate runs, not production outcomes.

WHY SPORTIQUE We are SportiQue, building a network where individuals collect, use and share their own health data, control permissions and review use history. UniQdata starts with the individual's personal health database. UniQLab supports research after the individual chooses to participate.

We face the same coordination problem: people need clear participation terms; researchers need suitable fields, periods and provenance. Repeated contributions must stay aligned with permission, delivery and compensation. AgentRail is our hackathon prototype for that problem.

APPLYING IT TO HEALTH DATA For example, a lab studying recovery may request three weeks of activity records. It proposes the purpose, fields, period and payment; each person accepts or declines. Their agent can handle later requests only within that approved scope.

• T0-T3 sensitivity rules cover 46 columns, GA4GH DUO purposes, retention and recipient trust.
• Approval binds recipients, projects, purpose, fields and period. An institution relationship can cover several sub-projects only within that scope. Changed scope returns to the person.
• Genomes, CT and MRI require fresh approval every time, even for a trusted lab.
• Intake signatures and committed samples detect specified mismatches; they do not guarantee scientific usefulness.
• k=5 cohorts, study pseudonyms and identifier controls reduce re-identification risk; they do not eliminate it or replace consent.
• Raw health data stays off-chain. Delivery is sealed to its recipient; the store remains private.
• Withdrawal stops future delivery and revokes the study's MPT license. Unreleased escrow can be refunded after CancelAfter. Paid compensation stays with the person; decrypted copies cannot be recalled.

A model of 50 synthetic people and six studies compared 11,600 item-level checks with 233 package signatures; 1,792 requests were handled within prior consent. Modeled median request-to-payment time fell from 12 hours to under a minute, assuming an 18-hour average owner response. The units differ: these are workflow comparisons, not measured customer or clinical outcomes. A paid pilot must test continued participation, usable records and repeat purchasing.

EVIDENCE AND LIMITS Standalone Testnet evidence covers Escrow, token Escrow, MPT/Clawback, Credentials/DepositPreauth, multisig, Batch, AMM and x402, including expected refusals. T54 submits x402 payments. QuickNode confirms transactions; its unsupported feature RPC is not amendment evidence.

Consent lanes address sensitive data, third-party provision and cross-border transfer. Code rechecks scope. Credentials attest an issuer's claim; the IRB decides approval. Live use needs local legal/security review.

Synthetic people, records and amounts. Hands-on: mock ledger/server demo keys. Before/after apps: illustrative. Separate XRPL Testnet proofs use app-issued USD standing in for RLUSD. Production device custody and AP4M settlement remain unproven. Not a medical device; no diagnosis.

NEXT Our goal is to start in Asia and expand to the US, Singapore and beyond, supporting data exchange between agents under each owner's terms.

TRY / PROOF Hands-on (mock): https://agentrail-try-production.up.railway.app/try?code=-DrGPuS8I0dn0nj3 Key reveal: https://testnet.xrpl.org/transactions/6A96310816B9EC9F7B445C113FFBB719E794A588AC20BE06D71C086C435706C4 Refund: https://testnet.xrpl.org/transactions/D2575FD948326043D2642CB6C470387FD911939DEC3F3E03FC26EA6C02EF51E6