Calyx screens the documents that arrive at your firm and the parties behind them for markers inconsistent with what they claim to be — then reconciles both against records held outside the transaction, by authorities with no stake in it. The finding is sealed as evidence a third party can verify without our help.
Almost everything that arrives at a firm is one of two things: a document, or a party. They fail differently, so they're screened differently — but both are struck against the same body of external ground truth.
Deeds, guarantees, bank letters, proof of funds, notarial certificates, wire instructions, filings, invoices. Screened for markers inconsistent with the class of instrument they claim to belong to, then reconciled against the authority that issued or recorded it.
The entity, its registration and standing, the people named as principals, the address, the licensing the claimed activity requires, and whether what the party asserts about itself matches what the registers actually hold.
Corpus is the reference base — a versioned, sourced library of what a genuine instrument of each class actually contains, what a legitimate entity's public record actually looks like, and which authority holds the record that settles it. The screens strike the sample. Corpus is what it's struck against. It gets deeper every month, and it doesn't transfer to anyone else.
A wire instruction is a document and a counterparty at the same time. So is a payoff letter, an engagement solicitation, an invoice from a vendor you haven't paid before. Forcing those into one lane is how a screen misses — both run, and the finding carries both sections. Where the classification itself is uncertain, that uncertainty is recorded rather than resolved silently.
A convincing forgery defeats a careful reader — that's what makes it convincing. It cannot defeat the registry, the issuing bank, or the court docket, because those records were never the forger's to write. Every criterion pairs what the sample asserts with the record that settles it.
No single marker decides anything. Findings come from convergence — several criteria pointing the same way at once.
Does it behave like the instrument it claims to be — form, tool chain, internal structure, metadata consistent with its stated origin.
Does the named entity exist in the jurisdiction it claims, in good standing, with a formation date consistent with the history it asserts.
Do the people cohere — names, titles, credentials, and the authority to sign what they signed. Is anyone identifiable at all.
Where the claimed activity requires a credential, does it exist, is it current, and is it held in this entity's own name. This criterion carries the most weight, because it admits no innocent exception.
Do the figures and the shape of the transaction hold together, and which way is capital actually flowing.
Do domains, headers, routing and telephony match the claimed geography and identity. Does the digital footprint match the claimed operating history.
Is the evidence reachable through a channel the counterparty does not control.
Does the party appear on a restricted list. Name collisions are common, which is exactly why this one is resolved by a person rather than by a match score.
What the party or document asserts about itself, set against what the authority recorded. Founding year against registration date. Headquarters against address of record. Scale against filing history.
Urgency, secrecy, engineered deadlines, withheld payment instructions, pressure to bypass a step.
How much the markers converge is a different question from how much we could actually reach. Most systems average them into a single score and lose the distinction that matters most.
Convergence and weight of the markers present in what was screened. Is this what it claims to be?
Whether a prudent firm should rely on this for the commitment actually in front of them. A thing can be entirely genuine and still be a bad bet.
Completeness of the inputs. What was supplied, what was reachable, what was missing — reported honestly and never blended into the other two.
No disqualifying markers within the scope screened. Not a clearance, and the finding says so.
Markers present that a stated condition addresses — escrow, payment on delivery, a verified callback on independently obtained details.
Material deviations present. A person has to resolve them before anything is committed.
The screen could not be completed at usable confidence. Not a negative finding — an incomplete one, with the gaps named.
The evidence runs against what the document or party claims.
Scope is fixed before a verdict is available, and below a confidence floor the only available verdict is STANDBY. A system that will say not yet, and here is what I would need is worth more than one that always produces an answer.
Every screened document is hashed exactly as received, before anything is done to it. The finding is hashed too. Both go into a provenance record issued alongside it.
Anyone holding a copy can recompute the digest with shasum -a 256 and compare. A match establishes the file is byte-identical to the one screened. A mismatch establishes it is not, without revealing what changed. No key, no software, and no cooperation from Calyx is required — the check is repeatable by a stranger years from now.
Content integrity is fixed; time is asserted. A hash proves what a document contained, not when it existed. Where a qualified time source is required, the record is anchored to an RFC 3161 timestamp authority — and where it is not, the record says so on its face.
This is the load-bearing constraint, not a disclaimer at the bottom of a page. A screen that renders verdicts about people is one a court, a carrier, or opposing counsel can take apart. One that documents what it found and routes the decision is one they can rely on.
Every finding is reviewed and released by a named human operator before it is issued. Nothing auto-clears and nothing auto-accuses. Determinations route to the professional who owns them — counsel, the client's own accountant, a licensed broker, or the authorities.
Screening covers commercial entities and the principals named in connection with their business conduct. It is not a background check on individuals, and it is not used for employment, credit, insurance, or housing eligibility decisions.
Screening happens at intake — before the wire, before the closing, before the counterparty becomes a client. Findings are useful afterwards too, but they're cheapest before.
Corporate records, identity documents, bank references and agreements screened as a set, so the conflicts between them become visible.
Proof of funds, bank letters, guarantees and payoff statements — with the channel that produced them treated as part of the evidence rather than as a given.
Deeds, notarial certificates and identity evidence, reconciled against the recorded chain and the commission record behind the seal.
Demands, invoices, wire instructions and solicitations — envelope and instrument screened together, since the message carries the document.
The single highest-value trigger. A counterparty already engaged, sending new details — re-screened against the baseline, where the change itself is the finding.
Accounting, legal, title and brokerage firms running intake diligence on behalf of the clients who rely on them, with a record that survives the question later.
Screening runs as a service, or on the node — on hardware the firm owns, behind its own firewall, with confidential material never leaving the building.
A scoped screen on a real document or a real counterparty is the fastest way to see what this produces — the finding, the reasoning, and the record behind it.
Start the conversation