Fragmented stack

The stack count

Take a single trade. $400K of nitrile gloves from a manufacturer in Anhui to a distributor in Texas. NIH-published distribution research finds 17 to 20 parties touch a typical international shipment. Industry analysis from before the TradeLens shutdown put it closer to 30 organisations: buyer, seller, both banks, possibly a correspondent, the carrier, the freight forwarder, customs brokers at both ends, port authorities, the insurer, the inspection body, the chamber of commerce, the destination regulator, and the internal teams behind each.

Each party operates on its own systems. McKinsey’s 2025 DHL Trade Atlas survey of logistics providers: 34% run 8 or 9 different technology solutions in their transportation tech stack alone. 37% run 5 or more solutions in warehousing. That is the provider side. The bank, carrier, and customs sides carry their own counts.

Page 02 was about paper, the medium. This page is about the systems, the structure. They are different problems. Going from paper to PDFs delivered by email solves nothing if the PDF has to be re-typed into 9 different systems on the way to the buyer. Digital is not integrated. The 17 to 30 parties were always there. The fragmentation just shifted from paper attachments to email attachments to incompatible APIs.

The standards graveyard

The standards-body answer to fragmentation was Electronic Data Interchange. ANSI chartered the X12 committee in 1979. The UN chartered UN/EDIFACT shortly after. The two have been the foundation of B2B trade messaging for 40+ years and they do not interoperate.

X12 (300,000+ companies, North America) and UN/EDIFACT (Europe and Asia) use different syntax, segment structures, and field codes for the same trade event. A purchase order is an 850 in X12 and an ORDERS in EDIFACT. Crossing the divide requires a Value-Added Network or a translation layer that introduces lag, cost, and its own errors. On top of these foundational standards sit a dozen industry-specific ones: EANCOM, HIPAA, ODETTE, Oracle Gateway, RosettaNet, SAP IDoc, SWIFT, Tradacoms, VDA, VICS. SWIFT itself is migrating to ISO 20022, another layer added through 2027 without removing the legacy MT messages it sits next to.

A company sending a $400K trade today routinely emits the same business event in four to six different formats: an internal ERP record, an EDI 850 for one partner, an EDIFACT ORDERS for another, a PDF attachment for a third, a CSV upload for a fourth, a custom JSON payload for a fifth. The data is identical. The representations are not.

The SWIFT silo

Banks have their own fragmentation problem layered on top of everyone else’s.

SWIFT moves messages. The money rides separately, through correspondent-banking relationships on different rails. Technical documentation puts it plainly: “Messaging functions independently from the actual settlement of funds.” A single letter-of-credit transaction can use 12 or more distinct SWIFT MT message types (MT700 issuance, MT701 continuation, MT705 advice via correspondent, MT707 amendment, MT710 third-bank advice, MT720 transfer, MT734 refusal, MT742 reimbursement, MT759 ancillary, MT760 SBLC, MT767 SBLC amendment, MT798 corporate-to-bank, MT799 free format).

For multi-bank trades (the common case in emerging markets where the issuing bank has no direct relationship with the seller’s bank), all message references must match at every level or the credit is voided. Each bank runs its own trade-finance system: Finastra, Surecomp, Misys, Oracle Banking, in-house mainframe. None of them share a database. Same business event, six representations, six chances to drift. And the actual settlement moves on a correspondent-banking chain that may or may not match the SWIFT message routing. Reconciling the two is its own job, performed by humans in back offices.

Why consolidation fails

The obvious response to fragmentation is consolidation: build one shared platform, get everyone on it, win. The industry has tried this repeatedly. The most ambitious attempt was TradeLens.

Maersk and IBM launched TradeLens in 2018 on IBM Hyperledger Fabric at $25 per container per journey. By mid-2019 the platform had 300+ industry members, 5 of the 6 largest ocean carriers (Maersk, MSC, CMA CGM, Hapag-Lloyd, ONE, ZIM), and 600+ ports and terminals connected. The technology worked. In November 2022, Maersk and IBM announced discontinuation. Maersk’s official reason, verbatim: “the need for full global industry collaboration has not been achieved.”

The analyst consensus is sharper. The technology worked. A platform owned by one carrier cannot be the platform for all carriers. Maersk’s commercial incentive to own the rail was structurally incompatible with the other carriers’ need to trust it. The same pattern repeated in the same 18 months: We.Trade (12 European banks + IBM, insolvent May 2022), Marco Polo Network (R3/TradeIX, 30+ banks, $85M cumulative losses, insolvent Feb 2023), Contour (8 major banks on Corda for digital LCs, shutdown Nov 2023). The Australian Securities Exchange parallel blockchain-settlement project was cancelled in the same November-2022 window for “technology, governance and delivery challenges.”

Surviving competitor GSBN (COSCO, Hapag-Lloyd, PSA, Hutchison, Shanghai Port) still operates but has not achieved industry-wide adoption either, a platform backed by COSCO is the Chinese shipping platform, not the neutral one. The lesson the industry took from this graveyard: no actor in trade can be the neutral party. A platform built by a bank for banks belongs to that bank. A platform built by a carrier for carriers belongs to that carrier. The trust problem is structural, not technical.

The rekey tax

While the industry argued over which consortium platform would win, the rest of the supply chain kept paying the rekey tax.

Shipamax (acquired by WiseTech Global in 2022) ran timed studies on manual document processing. Each document received as an email attachment takes 10 to 15 minutes (up to 60 minutes for larger documents) for a professional data-entry clerk to fully process. 90% of data entry in logistics still happens manually. Tradeverifyd’s 2026 supply-chain survey: 69% of compliance and supply-chain teams spend 11 or more hours every week on manual data translation. Conexiom: customer-service and inside-sales reps spend 20 to 40% of their time on manual order handling. The error rate runs up to 4%, and DataMondial finds correcting an error after the fact costs five times the original entry time.

This is the trap EDMA Group lived in. For each $400K nitrile-glove order we ran, the same shipment data was re-entered across 9 to 12 systems: our ERP, the supplier’s spreadsheet, the bank’s LC application portal, the customs broker’s export-filing system, the carrier’s booking portal, the freight forwarder’s TMS, the insurer’s certificate generator, the inspection body’s report system, the chamber of commerce’s certificate-of-origin tool, the destination customs system, and back into our accounting. None of them talked to each other. We employed people to rekey data. They missed numbers. We paid for the corrections. The protocol thesis didn’t come from reading the McKinsey report. It came from being the people doing the rekeying.

What it actually costs

The fragmentation tax adds up to a number roughly the size of a national economy. US business logistics costs reached $2.58 trillion in 2024, 8.8% of national GDP (Tradeverifyd 2025). Industry estimates of the integration tax inside that number range from 15 to 30% of total logistics overhead.

Disruptions reveal the cost most clearly. The average organisation reports $184 million per year in losses from supply-chain disruptions, with the per-day cost averaging $1.5 million (Supply Chain Dive). Less than 8% of businesses feel fully in control of supply-chain risks; 63% report greater-than-expected losses when disruptions occur. Every visibility gap is a fragmentation gap. When the LC fails, the bank doesn’t know the goods left the factory. When customs flags a shipment, the buyer doesn’t know until the demurrage clock starts running. Page 02 documented the cost of paper. This is the cost of operating across siloes once the paper is gone.

The opportunity

The TradeLens lesson suggests a different architecture.

A platform sits on top of existing systems and asks them to defer to it. Whichever party owns the platform owns the data, the rules, and the leverage. Carriers will not defer to a Maersk platform. Banks will not defer to an HSBC platform. The structural neutrality problem rules out platform-as-solution for an industry in which no participant is neutral.

A protocol sits underneath existing systems and lets them keep their own state. Its job is to ensure that when a shipment leaves the factory, an evidence-bearing message is signed and propagated such that the bank, the customs authority, the carrier, the buyer, the insurer, and the inspection body all see the same canonical event without any of them having to defer to the others. This is the architectural inversion behind TradeOS: not a competing platform, but a shared evidence layer underneath the existing platforms. Proof of Verification consensus requires every claim about a real-world event to be signed by independent verifiers, hash-matched to the same evidence, and committed exclusively once network-wide. That is the Settlement rail, with EDSD stablecoin and EMT milestone tokens on the EDMA L2. Page 04 walks through why the trust primitive itself needs to change. Page 05 walks through what becomes possible. We have been building this for two years.