Status: architecture note for partners and evaluators. Mirror is a layer, not a silo. It does one thing no tool in a bank's trade stack can do alone — reconcile a trade across two institutions without either seeing the other's data — and it composes with the tools that already sit on either side of that step. This note maps the seams. It names categories of partner and, illustratively, representative vendors; it asserts no partnership Mirror has not signed.
The shape of the seam#
Everything Mirror touches is either produced inside a bank before the commitment or consumed inside a bank after the verdict. Raw data never crosses the middle. The three seams:
INSIDE BANK A THE BLIND MIDDLE INSIDE BANK B
┌────────────────┐ ┌──────────────────────────┐ ┌────────────────┐
│ doc extraction │──fields──▶│ bander: band + commit │ │ doc extraction │
│ physical feed │──leg────▶ │ (party-side, on a laptop)│ │ physical feed │
└────────────────┘ └────────────┬─────────────┘ └────────────────┘
│ opaque commitment only
┌──────▼───────┐
│ blind operator│ reconciled / divergent / undetermined
└──────┬───────┘ + signed, offline-verifiable
│ verdict
┌───────────▼────────────┐
│ screening / case mgmt │ (inside each bank)
└────────────────────────┘1. Upstream — document & data extraction (feeds Mirror)#
Category: trade-document digitisation and compliance-examination platforms that already read the raw documents inside one bank (OCR, NLP, LLM extraction, UCP/ISBP rule engines). Representative vendors: Cleareye.ai, Traydstream, Conpend, Surecomp.
The seam. These tools extract structured fields — unit price, quantity, Incoterms, HS code, control-list hits — from LCs, invoices and bills of lading, inside the bank. Mirror's party-side bander takes those fields and produces the salted band commitment locally; nothing raw leaves. The extraction vendor never has to become privacy-preserving, and Mirror never has to build document OCR. Each does the half it is good at.
Why it's a partnership, not a competition. An extraction tool is structurally single-bank — it reads this bank's documents. It cannot see the counterparty bank's price and cannot catch duplicate financing across institutions. That is precisely the step Mirror adds. Two banks each running an extraction tool can reconcile through Mirror without either tool ever exchanging data.
2. Side — physical-leg & counterparty intelligence (feeds the physical checks)#
Category: maritime-behaviour, vessel-ownership, sanctions and supply-chain-graph intelligence. Representative vendors: Windward, Sayari, Kharon, Pole Star.
The seam. These platforms own the observable physical and ownership data — voyages, AIS gaps, ship-to-ship transfers, beneficial-ownership graphs. Mirror's physical-leg check binds money to matter by comparing what the two banks independently assert about a shipment; it does not, and should not, aggregate a global vessel feed. Where a bank already subscribes to a physical-risk feed, that feed supplies the attributes the bank commits; Mirror reconciles whether the two banks' independent assertions agree, blindly.
Why it's a partnership, not a competition. These are aggregation-and-scoring products — the opposite of privacy-blind by design; their value is in centralising observable data. Mirror does not compete on owning that data and should not try to. They are the more plausible partner (and, candidly, acquirer) than rival: they supply the leg, Mirror supplies the blind cross-bank reconciliation of it.
3. Downstream — screening, monitoring & case management (consumes Mirror)#
Category: AML transaction-monitoring, sanctions/PEP screening and case-management suites that already run inside the bank against its own data and watchlists. Representative vendors: ComplyAdvantage, Napier, SymphonyAI Sensa-NetReveal, Feedzai.
The seam. Mirror emits a signed, three-state verdict — reconciled / divergent / undetermined — and nothing else: no score, no rank, no filing recommendation. A divergent verdict is a first-class, offline-verifiable reason to look, routed into the case-management and monitoring system the bank already runs, where the human decision (review, escalate, file) is made under the bank's own policy and its anti-tipping-off obligations.
Why it's a partnership, not a competition. These suites detect within one bank. They have no cross-institution primitive and no privacy-blind multiparty step. Mirror does not make filing decisions and does not want to — it structurally refuses to. It hands the downstream system a defensible cross-bank signal that system could not have produced itself, and stops there.
What Mirror deliberately is not#
Mirror is not a document reader, not a data aggregator, not a risk-scorer, and not a case-manager. Each of those is a mature product category with strong incumbents, and each is the wrong place for a rail whose entire value is that it sees nothing. Mirror is the one blind step in the middle. The integration story is that it makes the tools on both sides worth more without asking any of them to change what they are.