Normalise. Fingerprint.
Match. Explain.
No machine learning you cannot interrogate, no score without reasons. The registry does four deterministic things and shows its working at every step.
What happens between submit and verdict.
These steps run in order because each one depends on the last. The whole sequence completes before your officer sees a result.
The record is submitted
An officer enters the instrument through the member portal, or your core system posts it over the API. Identifying fields are mandatory; commercial terms are not collected.
collateral · cheque
References are normalised
Identifiers are upper-cased and stripped to alphanumerics, so MAEU-1194837 and maeu 1194837 collapse to the same value. Company names lose legal suffixes and generic trade words before comparison.
→ VERTEX MILLS
A fingerprint is computed
The normalised identifying fields are joined into a canonical string and hashed with SHA-256. Two records that describe the same underlying document produce the same fingerprint, regardless of who typed them or how.
9f2c8a…41be
The registry is searched
An exact fingerprint match is conclusive. Where it does not fire, hard identifiers are compared individually and fuzzy similarity on party names and cargo description adds weighted signal, producing a 0–100 score against the closest existing record.
score 97 / 100
A verdict is returned with reasons
The score maps to a band, and the band arrives with the specific signals that produced it. An officer who cannot explain a decline should not be making one.
3 reasons cited
The case is worked and logged
Suspect and duplicate results open a fraud alert with an owner. Whatever the officer decides — confirm, dismiss, escalate — the disposition and the reasoning join the audit log.
audit entry sealed
What carries weight, and why.
Hard identifiers dominate because they are the fields a duplicate cannot change without breaking the underlying transaction. Fuzzy signals exist to catch the near-misses that hard identifiers alone would let through.
Hard identifiers — dominant weight
Fields that identify a specific transaction and cannot be varied without changing what is being financed.
- Bill of lading number
- Container number
- Shipping bill number
- Commercial invoice number
- Document fingerprint
Fuzzy signals — supporting weight
Comparisons that survive small deliberate changes, so a re-typed exporter name does not defeat the check.
- Exporter and importer name similarity
- Cargo description similarity
- Cargo value proximity
- HS code agreement
- Time between the two registrations
Clear. No meaningful overlap with any existing record. The registration proceeds and becomes searchable for the next member.
Suspect. Shared signal that could be coincidence or could be the same document. An alert opens for manual review before funds move.
Duplicate. Treated as the same underlying document. The submitting member sees which member holds the earlier record and when it was registered.
Thresholds are member-configurable. The values above are the registry defaults; your risk function can tighten or widen them for your own organisation without affecting anyone else's.
The check no single institution can run.
A factor looking at an invoice cannot tell whether a bank has already lent against the shipment behind it. A bank looking at a shipment cannot tell whether the receivable has been sold. Linking the registries is where most of the value sits — it turns two partial views into one complete one.
- Invoices are linked to their underlying shipment where identifiers permit
- Financed status propagates from the bank registry to the factoring check
- Collateral charges surface against the borrower, not just the asset
- Cheque history follows the drawer across every cashing member
What onboarding actually involves.
Order matters here too — due diligence before access, pilot before production.
Institutional due diligence
We verify your entity, licences and regulatory standing, and you review our security posture, data handling and legal terms. Nobody gets registry access on a trial signup.
2–3 weeks
Agreements and data terms
Membership agreement, data processing agreement and the acceptable use terms that govern what members may do with what they see.
1–2 weeks
Roles, scoping and training
Your administrators are provisioned first and create their own officers. Roles map to the authority those officers already hold inside your institution.
1 week
Pilot on live traffic
Run the check alongside your existing process without acting on it, so your risk team can calibrate thresholds against outcomes you already know.
4–6 weeks
Production and reporting
The check becomes a control in your credit process, alerts route to named owners, and reporting lands with your risk committee on a schedule you set.
ongoing
Bring us a document that got past you.
The fastest way to understand the check is to watch it run on a case your team already knows the answer to.