Methodology, we compute the determination, we don't generate it

How the platform works

Agents do the work. Your determination stays consistent.

Foundation models are probabilistic: read the same document twice and the phrasing can come back differently. The determination itself runs through a deterministic rule engine, the risk ontology, the transaction graph, the regulatory register, the control library, and the historical outcomes and loss data it's checked against, so the same facts produce the same result every time, reproducible and versioned, whichever model read them in.

Probabilistic

Frontier models

Whichever foundation model is best this quarter does the reading and drafting.

Deterministic

Arqis risk logic

Risk ontology · transaction graph · regulatory register · control library · historical outcomes · insurance appetite · loss data.

Autonomous work

Specialised agents

Contract · Counterparty · Regulatory · Credit · Technology · Insurance · Monitoring.

Decision layer

Humans, by exception

Challenge agents, reconcile conflicts, quantify exposure, determine controls.

The model reads the document; the rule engine makes the determination. Foundation models can change without changing your risk record or your workflow.

DecideCited, test by test

Every test names the rule and its version, the facts it relied on, and the result, so the reasoning survives a rule change or the person who ran it.

VerifyConditions become controls

Each condition is classified by how it can actually be evidenced, so a condition on paper turns into something the monitoring layer can check for real.

EscalateWe compute, your counsel judges

A contested interpretation goes to your own counsel with the full trace attached, so they start from our working rather than a blank page.

We determine what each regime requires; we don't decide what should be permitted. Arqis is convening an advisory council of former regulators and policy specialists across jurisdictions to review that methodology, not any customer's commercial decisions, since a council that reflected only one government's view of what's permitted would defeat the point of a cross-border standard.

Why it holds up

Every number traces back to a clause, a rule and a date.

You will be asked to defend these numbers in front of a credit committee, an auditor, an insurer and the counterparty across the table. Generated prose cannot be defended: ask twice and you get two answers, neither citing the rule it came from.

The probabilistic default

Generated prose, no citation
No confidence attached
Answered once, never revisited
Silent when the rule moves
Silent when the facts move
One interpretation, asserted as fact

What Arqis computes

Rule cited and named, per test
Effective date and confidence stated
Re-run on every rule change
Re-run on every fact change
Breach detected at breach, not at audit
One neutral determination, same for every side

What you can defend

A trace you can walk back to the regulation
A position that stays true, or says it no longer is
Something every counterparty can work from
Judgment that stays with your lawyers
Contested calls escalated, not guessed
Reasoning exposed, never just asserted

A worked example: one transformer supplier slips 120 days, and five counterparties feel it. Arqis returns $28.5M of revenue exposure inside a day, the termination right it triggers, the financing milestone it affects, and the four things you can still do about it.