Compare · approaches, not logos
The honest comparison. Including where it loses.
Mirror is not the most powerful tool in this table. It is the one two compliance teams can run tomorrow, and the one whose answer a regulator can check without trusting anyone. Here is the trade, stated both ways - and a filter instead of a score, because a comparison should tell you what fits your constraints, not rank anybody.
Pick what you cannot live without
Six approaches, one question: which fit?
Select the requirements your programme cannot bend on. Approaches that fail any of them fade out; the ones that depend on how they are built stay, marked. Nothing here adds up to a score.
Nothing selected: every approach is shown.
| Requirement | Mirror | General MPC | General ZK proofs | Trusted data pool | Document-fingerprint registry | Each bank alone |
|---|---|---|---|---|---|---|
| The operator learns no raw valueA registry holds document fingerprints, not values - but they are identifiers. | yes | yes | yes | no | depends | n/a |
| No third party ever holds the data | yes | yes | yes | no | no | n/a |
| Runs without special hardware or a key ceremony | yes | no | depends | yes | yes | yes |
| The verdict is signed and verifiable offline by anyone | yes | depends | yes | no | no | no |
| Refuses to clear on the stakeholders' word aloneMirror's independence gate is enforced in code; the others can be built to, but it is not the primitive. | yes | depends | depends | depends | n/a | no |
| Detects cross-institution mispricing | yes | yes | yes | yes | no | no |
| Catches one cargo financed under different documents | yes | depends | depends | yes | no | no |
| Catches one document financed twice | yes | yes | depends | yes | yes | no |
| Compares exact, continuous valuesMirror compares bands, so even a clear is same-band agreement. | no | yes | yes | yes | no | yes |
| Works without a pre-agreed corridor tableMirror needs a table both banks ratified first. | no | yes | yes | yes | yes | yes |
"Depends" means the approach can be built to meet the requirement but it is not a property of the primitive itself; "n/a" means the requirement does not apply, so it never excludes an approach. "Trusted data pool" is any arrangement where a third party holds the parties' raw data and compares it; "each bank alone" is ordinary monitoring inside one institution.
Where Mirror loses
Said first, so the rest can be believed.
Bands, not prices
A clear means the two legs landed in the same band of the agreed table. A band is a range, so even agreement is consistent with a price difference inside it; the table's signed geometry certificate states the widest one it can hide.
A table first
Two banks must ratify a corridor table before a single trade is compared. MPC and zero-knowledge approaches can compare values without one.
A witness to clear
Two banks agreeing is not a clear: a ratified witness with nothing to gain - customs, a terminal, an assayer - has to agree too. No witness is connected today, so the live rail clears nothing yet.
Not zero-knowledge
The operator is blind; the counterparty is not fully. The two banks share the per-shipment secret, and a band has few values, so a counterparty can work out the other's band. That is stated in the spec, not hidden.
Against the incumbent model
Where a document-fingerprint registry stops.
The established way to catch cross-bank trade fraud is a registry of hashed documents: submit a fingerprint, get flagged if the same document was financed before. It is the benchmark for one job, and structurally blind to two others.
Exact-document duplicate detection at global scale, on rails banks already trust, with no table to agree. If the fraud is the same invoice financed twice, a registry catches it and has the distribution to do so everywhere.
A fingerprint matches a document, not a price. Over- and under-invoicing uses a unique, legitimate-looking document for every leg, so no fingerprints collide. Mirror bands the price against a ratified table, so a 2.9x mispriced trade reads divergent even when every document is one of a kind.
One lender finances a bill of lading, another a warehouse receipt, for the same full container. Different documents, no match. Mirror's cargo key binds the ISO 6346 container number to its port and load date, so the two lenders meet on the cargo itself.
A registry is a central data-holder: one breach exposes everyone who joined. Mirror's operator holds opaque commitments and signs verdicts anyone can check offline - there is no pot to seize.
When to choose something else
Good advice includes the other answer.
If your constraint is one of these, Mirror is not your first tool.
- You need exact values compared, not bands - and both parties can run an MPC deployment or accept a proving system. Take that route; bands will frustrate you.
- You only need to catch the same document financed twice - a document registry already does it at scale.
- You can accept a trusted third party holding the data - a pooled model compares more, at the cost of a honeypot and the transfer friction that comes with it.
- You need a clear today, with no witness in your corridor - Mirror will say undetermined until a party with nothing to gain agrees.
Still here?
Then check it, don't take it.
Verify a real signed verdict in your browser, read the refusals that make the design hold, or see what a corridor costs.