You're already doing it — you just can't prove it, move it, or sell it twice

Executive summary

To label beef "Product of USA" under the USDA/FSIS rule effective January 1, 2026, the animal must be born, raised, slaughtered, and processed in the United States — and you must keep records that support the claim and present them to FSIS on request. Every serious processor's first reaction is correct:

"We don't need a new system. We already collect this data."

You do. Audited process-verified programs, kill-floor and grading data, load sheets, supplier affidavits, a GS1 feed, maybe a DNA program. This paper concedes that fully. The problem is not missing data. The problem is that a system of record is not a system of proof — and the two things the market now rewards are proof and movement of proof:

• Your records prove things to you. Everyone else must trust your database or audit it after the fact. After the 2024 rule made a false claim affirmatively misbranded, "trust me" is a rising legal exposure.
• The fact that decides the claim — where the animal was born — is known by the rancher, not you. Today you capture it as a signed affidavit: a fresh assertion, only as strong as the weakest signer.
• Your record dies at the loading dock. It doesn't travel across the seams — feedlot commingling, the freight handoff — where a truthful-at-birth claim silently becomes false.
• Your data can't transact. It can't carry a price, a permission, or an isolation boundary as it moves.

This paper describes the complete alternative: an open data supply chain in which every party — field, ranch, feedlot, packer, carrier, retailer, and even the AI agents that buy and audit — holds a self-sovereign digital identity and signs the one claim only it can vouch for; in which information binds to the identity of the animal and transforms as it becomes a carcass, a primal, a strip steak; in which quantity is conserved so a claim can't be minted twice; and in which data can be transacted. It is not a database. It is a network of owned, verifiable records built on top of the standards the industry is already adopting (GS1 EPCIS, W3C Verifiable Credentials, DIDs) — which is exactly why it can be verified by an AI agent that has never integrated with you and does not have to trust you.

The ask: a scoped lighthouse pilot on the one leg AFG fully controls — its own freight — that proves the rails end to end and extends outward from there.

01

The objection, taken seriously: "We already do this"

Concede the strongest version, because it's mostly true. A modern packer runs USDA-audited process-verified programs; AFG synchronizes trade-item data through GS1 and moves its own freight on America's Service Lines; OWB documents origin "down to the field"; FPL runs DNA TraceBack. This is real, and it is more than most industries have.

So why isn't it enough? Five structural gaps that "more database" cannot close:

1. It can't be verified from the outside. Your records are evidence to you. Everyone else must trust or audit them — now a legal cost.
2. It's single-party. The birth fact lives with the rancher; you capture it as a PDF of someone else's say-so, unlinked to evidence.
3. It dies at the dock. Identity is lost at the commingling and freight seams — the single hardest fact to reconstruct, and the one your database cannot span.
4. It can't transact. Data in a database just sits there; it can't carry value, permission, or an isolation boundary as it moves.
5. It wasn't built for the reader that's arriving next — AI. Increasingly the party consuming supply-chain data isn't a person; it's a machine making a decision and driving a process — a procurement system, an export-compliance engine, a shopping agent. An AI can only act on data it can verify is real: give it a verifiable record and it checks origin trivially and at scale; give it an unverifiable one and it trusts blindly or ignores it. Every prior generation of traceability was built for humans and periodic audits — a large part of why fraud keeps rising even as we collect more data: we digitized the records without making them provable. The AI future doesn't merely prefer better provenance; it requires it.

The conclusion is not "your data is bad." It is that Product of USA and the market — and, increasingly, the machines that read for it — now ask for two things you don't have: proof anyone (a buyer, a regulator, an AI) can verify, and proof that moves.

02

Identity that transforms — and quantity that is conserved

Everything gets an identity, and identities roll up as product transforms — while quantity is conserved so nothing can be claimed twice.

A field has an identity. A calf is born with an identity incorporating the field, the dam, the ranch, and its 840 EID tag. Movement events extend the animal's identity. At the plant the animal becomes a carcass identity that rolls up the animal's; the carcass is broken into primals, then cuts — each resolvable, one hop each, back to the field.

Meat lineage is a directed graph, not a row, with two operations relational schemas handle badly — and one property v1 was missing and this version makes central:

• Split (one-to-many): one animal → dozens of cuts; each child inherits and narrows provenance.
• Combine (many-to-one) — honestly. Ground beef is continuous, commingled flow; there is no clean per-animal edge into a grinder lot. So we model two regimes: discrete/whole-muscle items (a strip steak) as per-item identities with real lineage, and fungible/commingled streams (grind, trim) as lot-level identities governed by mass balance over a time window. "All-domestic" for a ground lot is a property of the input pool, bounded by segregation discipline. This is where Product-of-USA claims live or die: one imported animal in the pool breaks the lot.
• Mass balance (conservation): every identity carries a balance; each transformation debits inputs and credits outputs under a yield factor; a claim propagates only up to available balance; a transfer/consumption registry prevents replay. Without this, "sell it twice" is a double-spend; with it, the network is auditable and non-forgeable at scale.

The "born in the USA" claim attaches to the animal's identity and travels with it — inherited, not re-asserted — all the way to the cut. A stack of affidavits can never do this.

03 · the most important thing this paper says

Verifiable is not the same as true — and why that still wins

A signature does not prove a claim is true. It proves who asserted it and that it wasn't altered.

A rancher can cryptographically sign a false birth location in three seconds. So we do not claim "trustless truth." We claim something narrower and genuinely powerful — the upgrade from an anonymous affidavit to a record that is:

• Attributable & non-repudiable — a named, accountable legal entity signed it and cannot later deny it.
• Tamper-evident — any downstream alteration is detectable.
• Physically bound — anchored to the 840 EID tag, and for high-value claims to DNA, the one biometric that survives transformation and re-links a carcass (or ground product) to its source without trusting a tag that comes off at slaughter. DNA is only as trustworthy as its baseline: the origin DNA reference must itself be provenanced — captured and signed into a data wallet bound to that animal at the moment of sampling, so it becomes the immutable record every later test is checked against and can't be quietly swapped or re-baselined.
• Accountable — a false signed claim is attributable misrepresentation with a liable signer, not deniable paperwork.
• Graded by depth of measurement (L0–L3) — affidavit L1, EID-anchored L2, DNA-anchored L3 — published as a machine-readable field so a buyer or agent reasons about claim strength and prices accordingly. We never present a claim as stronger than its evidence.

This honesty is not a hedge — it is the reason the design survives contact with a skeptic, a regulator, and a plaintiff's lawyer. It also names the boundary of what technology can do (the oracle problem), which §7 confronts directly.

04

A network of owned records, not a trusted database

Four properties make it a network rather than a database — each a specific mechanism, not conferred by the word "network":

Multi-party contribution. Each party signs only the claim it can honestly make. The record is assembled from independently-signed pieces, each checkable on its own. No operator has to collect — or be trusted with — everyone's data.

Data isolation. Contribution without exposure. Attribute selective disclosure (reveal "born USA," hide the herd book) is available today (BBS+/SD-JWT). Aggregate proof ("this lot is all-domestic without revealing whose animals") is a zero-knowledge predicate over a committed set — harder, on the roadmap, paired with physical segregation. Isolation is the direct answer to the cattle-ID revolt, where the objection was never traceability but who holds the data.

Transactability with non-replay. Because each record is a signed object owned by an identity and governed by the mass-balance ledger, it can carry value and permission as it moves without being double-spent.

A trust anchor. Verifying a signature proves a key signed — not that the signer is authoritative. So the network includes an issuer-accreditation registry: who may assert which claim types, rooted in an authority a verifier already believes. An agent's verification = valid signature + accredited issuer + not revoked.

A good ERP is a fine place to keep your data. It is not a place where other parties contribute, where isolation is guaranteed, where quantity is conserved, or where data transacts.

06

It speaks AI at every step

The center of gravity of who reads food data is shifting from humans to machines. Rootz already sees this on freight.rootz.global (~940K requests/week; the real audience is agents — a signal of machine demand).

The legacy way every agent is dropped into a supply chain assumes a single trusted owner reached through an out-of-band integration (EDI, an API behind a login, "trust my DB"). That does not compose for autonomous agents transacting across organizational boundaries — because trust and integration don't scale. A network of self-describing, independently verifiable signed events lets an agent transact with a party it has never integrated with and does not have to trust — verification replaces both the integration and the trust relationship.

Relational data answers "what does the owner say?" A signed, standards-based event graph answers "what can I verify myself?"

Four concrete commitments

1. Every event is a self-describing, standards-based, signed object (EPCIS 2.0 + VC) an agent consumes without an integration project.
2. Every identity is resolvable and every claim checkable — Digital Link / DID → an OpenID4VP presentation; verification = valid signature + accredited issuer + not revoked + fresh.
3. Every agent is itself an identity. The buying/auditing agent joins as a DBA-class CorpID with a verifiable delegation credential; its actions re-enter the graph as signed events. The same primitive that proves the steak's origin proves which agent verified it, under whose authority, when — the endgame ties to Rootz attested inference.
4. The network is the API — an MCP tool surface (discover / verify / quote / buy / subscribe), a machine-payable rail so an attestation fee clears agent-to-agent, and the delegation credential that authorizes the spend.

07

When it breaks — and when someone lies

A complete picture must survive the adversarial case.

• When a party lies. The system does not magically detect a false birth claim — but the signer is named, accountable, and liable; the claim is graded L1 unless corroborated; a buyer can require higher. Lying is materially harder and costlier than an anonymous PDF, and provable after the fact.
• The oracle / tag-swap problem (the deepest question in traceability). Tags mis-scan, get swapped, and come off at slaughter. The answer is DNA as the L3 anchor — a biometric that survives transformation and re-links carcass and ground product to source, independent of any tag.
• Lost/compromised keys; farm sold; signer out of business. Identities support rotation and revocation; historical signatures remain verifiable against the key valid at signing time; verifiability does not depend on the original signer staying online.
• Recall / invalidation. A Status List + claim-invalidation propagation flags every inheriting credential when an upstream claim is revoked — wired to real recall workflows.
• Plant-scale signing without slowing the line. ~100K+ events/day sign off the critical path (async, HSM, Merkle-batch). And plainly: a plant signing its own roll-up buys attribution and tamper-evidence, not independent verification. The real trust gain is cross-party — the rancher's birth signature the packer can't forge.

08

The end-to-end system (AFG & OWB, walked honestly)

1. Field & birth (external to AFG). The ranch, as a business with its own DBA identity (CorpID), signs the birth event, binding the field and the 840 tag (L1–L2; L3 with DNA). AFG doesn't own cow-calf and buys commingled cattle — so this leg is populated first through existing source-verified premium programs where the birth ranch is known, degrading gracefully to affidavit-level where it isn't.
2. Raising & feedlot. Movement and segregation events are signed; isolation lets the packer verify "domestic, never commingled" without seeing the ranch's book.
3. Slaughter & processing (AFG plant). AFG signs the carcass roll-up and cut splits under the mass-balance ledger; the animal→carcass link is corroborated by DNA where value warrants.
4. Freight (America's Service Lines). The BOL is issued as a control-based electronic record. Where parties use a negotiable document of title under UCC §7-106 in a Revised-Article-7 state, control = legal possession. For the non-negotiable straight bills typical of trucking, the value is tamper-evident custody integrity — still the cleanest demonstrable win, because AFG owns the carrier.
5. Retail (OWB brand). The tray resolves — Digital Link → OpenID4VP — to the selectively-disclosed, level-graded chain back to the field, checkable by an inspector, a buyer, or a shopping agent.

09

The real prize — proof of quality, which is what drives margin

"Product of USA" looks like a compliance sticker, and even USDA has framed it mostly as origin. The bigger message — the one largely missed — is quality, and the proof of it. What this system really does is modernize the food supply chain the way manufacturing industries already think: where documented quality is not an annoyance to be tolerated but the very thing that commands price and margin.

Origin is only the beginning. The same signed, level-graded records that prove where an animal was born can document how the product was made — the inputs used on the field, the sanitation of the slaughterhouse, the cold chain that proves the meat never rose above a set temperature, and every process step to the shelf.

Provable quality, end to end — a data supply chain bound to the food.

Quality is what drives margin — and Wagyu is the proof. Wagyu commands its premium not because of a country sticker but because its quality is documented and provable — lineage, feeding, grading, the whole chain. A provable-quality data supply chain does that for the rest of American product: it turns "Made in USA" from a slogan into a checkable quality guarantee, which is exactly what raises the world's willingness to pay and defends the margin American producers earn. The premium and the proof travel together.

Higher-value product naturally attracts imitation, so verifiable quality is also its own protection — an AI, a buyer, or a regulator can confirm the proven chain and flag anything that doesn't verify. But that is the byproduct, not the headline. The headline is simpler: quality you can prove is quality you can charge for.

How the value is captured (B2B first)

• Win and keep contracts. Verified end-to-end quality becomes a spec that wins/retains retail, foodservice, and export accounts — and makes your retailer win the shopping agent.
• Export access and premium. Machine-checkable quality is what strict export regimes demand and what global premium buyers pay up for.
• Owned brands acquire subscribers. For owned premium brands only, a consumer — or their agent — can subscribe to a proven-quality stream and buy direct: a niche upside that strengthens, not bypasses, the core channel.

Product of USA is the reason to start. Proving the quality behind it — the margin-driver USDA's own framing underplays — is the reason it compounds.

10

Business model — and "your proof survives us"

• Large producers (AFG scale). Enterprise deployment on their own CorpID infrastructure, signing off the critical path. Pricing: platform license + per-volume signing — usage of the protocol, not rent on your data.
• Small producers (ranch, co-op). Low-friction per-head signing from a phone; the fee is trivial and earned back when the packer pays for the birth attestation or through direct followers.
• Portability. Verification depends on open standards and public keys, not on Rootz staying alive. Records are exportable; a producer can leave and take their identity and signed history. No lock-in — which is what makes an industry-wide, competitor-safe network possible.

A quantified ROI model is built per-account in the AFG/OWB proposal; this paper sets the frame, the proposal sets the numbers.

11

How this aligns with existing law

Every design choice is aligned to where the law already is. Stated defensibly:

• Product of USA (9 CFR 412.3). Signed, verifiable, level-graded records give a complete substantiation file to present to FSIS on request. FSIS — not the record — judges sufficiency; affidavits remain permitted, so the value is better, faster, more defensible evidence, not automatic compliance.
• eBOL (UCC §7-106). For a negotiable document of title in a Revised-Article-7 state, control = legal possession. For the non-negotiable straight bills typical of trucking, the benefit is evidentiary custody integrity, not transfer of title. (Article 12 excludes documents of title; no federal eBOL statute; no US MLETR — UK ETDA / Singapore are cross-border partners.)
• Cattle-ID (APHIS EID; R-CALF v. USDA). Producer-owned identity is the data-ownership layer on top of the official 840 tag — it answers "who owns the data" without displacing the official-ID requirement.
• Liability (Thornton v. Tyson, 2022). Thornton upheld preemption and dismissed the suit because FSIS pre-approved the label. The 2024 rule changes the analysis — a false claim is now misbranded, so a state claim mirroring the federal standard may survive as a parallel claim (untested, but it raises exposure). Auditable records are the strongest evidence to defeat such a claim on summary judgment — not a Rule 12(b)(6) pleading argument. For the meat label, FSIS (not FTC) has primary jurisdiction.
• Mandatory-COOL (S.421, H.R.5818). Low-cost universal provenance attacks the compliance-cost asymmetry the WTO faulted in US–COOL — a strong candidate for the WTO-compliant method S.421 directs USTR to develop.

12

Governance — an open protocol, not a private platform

The paper's own argument is that single-vendor databases fail politically. So the network cannot be one. The resolution is a layered separation:

• An open protocol — formats (EPCIS/VC/DID), resolution, presentation, revocation, and the issuer-accreditation trust registry — that anyone may implement, shepherded through the standards bodies already in play (W3C, GS1, NIST NCCoE).
• Rootz as first/reference implementation and initial registrar — revenue on protocol usage, never on locking up data.
• A governance path (consortium/foundation) so a second implementation can interoperate and no single party — including Rootz — is a choke point. Data ownership, exit, and portability guaranteed in the spec.

This is the honest answer to "how is your open network not just a new central database?"

13

We already have the components

Composition, not a moonshot:

• CorpID — the sovereign corporate identity factory: the farm/plant/carrier/agent DBA identity.
• Proof-of-origin / proof-site — the attestation substrate.
• Data-wallet / Government-Data-Wallet — custody + policy: data with origin and permission built in.
• Freight "Speak AI" — the agent-native MCP envelope, live today.
• Measured AI — the brand and the L0–L3 discipline that keeps every claim honest.

New is the composition into a food data supply chain and the go-to-market around Product of USA.

14

The ask — a scoped, honest pilot

We do not ask AFG to boil the ocean or pretend a single-enterprise pilot proves a multi-party network. We propose to prove the rails on the leg AFG fully controls, then extend — the honest answer to a network's cold-start problem:

• Phase 1 — the freight/eBOL leg (AFG + America's Service Lines). Signed, tamper-evident, §7-106-aware custody from plant to retail on AFG's own carrier. Real, novel, demonstrable, entirely inside AFG's control.
• Phase 2 — birth attestation via a source-verified premium program (OWB or an AFG premium line), where the ranch is known — proving the cross-party signature that is the real trust gain.
• Phase 3 — open the trust registry and standard to more ranchers, carriers, retailers, and their agents.

You're already doing it. Now you can prove it to anyone, move the proof across the dock, and turn it into contracts — to FSIS as compliance, and to your customers as an advantage their competitors can't verify.