The problem Secretariat solves

Thoroughbred racing is a massive $115 billion industry, but it has a high barrier to entry, strong cultural and economic exclusivity, and the data that drives decisions (pedigrees, performance history, injuries, and breeding outcomes) is aggressively guarded. This data opaqueness is by design: owners and trainers intentionally hide medical records in restricted "Sales Repositories" to protect their betting edge and prevent algorithmic devaluation at auction. Valuations are often negotiated privately, and bloodlines live in disconnected databases or even paper records. This severe informational asymmetry makes it difficult for owners, breeders, and fans to price horses fairly, make breeding and investment decisions, and compete against legacy breeders and LLCs.

Secretariat is a high-fidelity DeFAI database and marketplace on the ADI Foundation network that turns registered racehorses into on-chain intelligent NFTs (iNFTs) with a verifiable pedigree and an AI-driven valuation agent. By replacing "handshake-based" valuations with cryptographic truth, people can use it to:

Track provable lineage and history: A 5-generation family tree, trait evolution, and network-effect events are all recorded immutably on-chain.

Get real-time valuations that solve the privacy paradox: Instead of demanding raw veterinary X-rays that ruin a trainer's betting edge, our 0G Labs verifiable AI ingests the physical data but outputs ONLY macro-health percentages (e.g., "Respiratory Health: 80%", "Breeding Potential: 90%") to the public market. Only the verified iNFT owner can decrypt the granular "why"—the specific words, explanations, and audit trails behind those scores. This maintains the industry's traditional "Repository Shield" for private data while giving investors the mathematical proof they require.

Make smarter breeding and investment decisions through incentivized honesty. The platform provides AI recommendations and projected ROI for breeding pairs. To navigate firewalled medical data, we pivot middlemen from "trust" to "cryptographic liability". Bloodstock Agents act as "Certified Curators" who receive an immutable 5% sourcing fee, but must stake $ADI tokens to certify a horse's medical standing. If subsequent public race performance or state regulatory filings contradict their report, their stake is programmatically slashed. Democratize access through fractional ownership: The platform offers SEC-compliant fractional equity via bankruptcy-remote Delaware Series LLCs on a KYC-gated ledger. We standardize financial primitives around breeding revenue (stud fees) and performance-based value changes, protected by a mathematically locked OpEx buffer to ensure the biological asset is always safely funded.

In short, it’s a “Bloomberg Terminal for Biology” with accessible data, automated valuation, and decision support, all tied to a verifiable on-chain asset rather than paywalled spreadsheets and closed networks. We provide investors with the mathematical models they demand, give regulators the transparency they require, and offer owners a valuation premium for verified biological integrity.

Challenges we ran into

One of our biggest hurdles was a visibility bug. After breeding, the on-chain transaction succeeded and our breed event executed, but the new horse still didn’t show up in the user’s Portfolio/LiveFeed, making it look like the mint failed. Tracing the full path from contract to RPC to frontend revealed a multi-step pipeline problem: cached contract reads (Wagmi/React Query) were serving stale state, our refresh logic had a race condition between refetching the horse list and checking ownership, some RPC responses lagged 1–2 seconds immediately after confirmation, and a hardcoded tokenID scan cap silently skipped higher IDs. We resolved this by invalidating queries on breed success, serializing the refetch order, adding a short polling retry loop with debug tracing, and removing the scan ceiling with a direct on-chain fallback. We refactored the minting code and set up logs/tracing so we could see exactly where the execution flow was stalling.

Separately, we ran into contract-level breeding reverts with opaque errors that initially made it seem like the contracts themselves were broken, when in reality we were missing prerequisites. After debugging the revert conditions, we found two root causes: our deploy/seed scripts weren’t pre-verifying deployer/test wallets (they never called KYCRegistry.verify()), so the breed() function consistently failed eligibility checks, and our UI’s “Approve ADI” step wasn’t reliably executing our adi approval function before purchasing breeding rights, leading to insufficient allowance reverts. We fixed this by updating the seed scripts to verify test accounts, implementing a proper approval flow with allowance checking, and adding clear error toasts for each failure mode (KYC not verified, insufficient allowance, breeding rights not purchased), which made the entire breeding experience predictable and demo-ready.

Another issue in general is the lack of mass data when it comes to pedigrees and biometric statistics. There are no public API's or free databases. This makes it difficult to populate our registry with many horses. We are currently using demo horses alongside some horses that we do have data for. Overall, this is a problem we can fix through licensing and future adoption of our product.

Use of AI tools and agents

Secretariat uses AI/agent components to recommend breeding decisions, update valuations, and safely automate on-chain actions.
Breeding Advisor (recommendation agent): Given a mare, it returns the top 3 stallion matches with scores + explainability. Scoring blends heuristics (trait vector similarity, pedigree synergy, trait complement, budget/fee penalty, injury/form risk) and an XGBoost model that predicts expected offspring value. Output includes risk flags (e.g., high fee, injury, weak match). The UI calls /breeding/recommend, shows the “why,” and lets the user proceed manually or delegate execution.

Horse Valuation Agent: Computes a horse’s USD/ADI valuation with a clear breakdown (racing performance, breeding value, and modifiers like age/health/status). It’s used both for on-demand valuations in the app and as the intelligence inside the oracle workflow.

Oracle pipeline (event → valuation → chain): When real-world events occur (race result, injury, news), the system fetches current on-chain state, runs the Valuation Agent to propose an updated value, and commits the update on-chain (optionally persisting an auditable payload).

AgentExecutor (guardrailed automation): If the user enables automation, they sign an EIP-712 BreedingPlan (mare, chosen stallion, max budget, max stud fee, deadline, optional allowlist). The on-chain executor verifies constraints and then purchases breeding rights (if needed) + breeds to mint offspring. No funds move without a user signature, and execution is strictly limited by the plan.

How they work together: recommend (Advisor) → user reviews “why” → optional signed plan → on-chain execution (Executor) → valuations update over time via the oracle pipeline (Valuation Agent).