Institutions

Evaluate one financial workflow, from authority to recovery.

Operations, risk, architecture and audit teams need to explain why an action was allowed, what was observed and what justifies recovery. Evaluate FxChain against your existing controls on one bounded case; native mechanisms and institutional integrations have separate evidence.

Discuss a workflow See where it lands in your stack

Audiences

Built with your constraints in mind

Financial institutions

Rails designed for institutional discipline: provable refusals, auditable state, no legal fiction about what the chain can and cannot do.

Keep executed, finalised, delivered, legal and complete as five facts your auditor can read.

Payment providers

A staged path toward a payments platform (FxPay) with explicit authority boundaries at every step.

Track unresolved payment evidence, the next justified action and the obligations that still need an owner.

Markets and trading

FxTrade is a planned trading platform. Its execution, venue and settlement integrations would need evidence beyond the local protocol mechanisms.

Order, fill and settle as separate evidence; conflicts explicit, never buried in sequencing.

Treasuries

A programmable banking platform (FxBank) explored in stages: sandbox first, regulated partners next, never a claimed license before it exists.

Give an agent a mandate, not a key: scope, amount, time - and a portable refusal when evidence is missing.

Enterprise

Integration requires its own checks: input provenance, connector behaviour, authority boundaries and recovery under the failures your team actually sees.

Integrate a discipline, not a dialect: one grammar of authority, evidence and refusal across your stack.

Two scenarios

What the discipline changes, on two familiar cases

Design scenarios, not services: they show what the protocol is built to establish and what it refuses to compress. No amounts, no dates - the shape of the record.

Tokenised delivery versus payment

A securities leg is recorded while cash delivery and legal effect still need confirmation.

FxChain is designed to keep five facts apart and to seal each of them: executed on the protocol, finalised by consensus, delivered on the external rail, legally effective, economically complete. A trade can be executed and finalised while the cash leg on a bank rail is still pending and the legal effect still unconfirmed - and the record says exactly that, instead of a single flag. Missing evidence is recorded as missing, never as false.

Executed on the protocol: yes
Finalised by consensus: yes
Delivered on the external rail: pending
Legally effective: unconfirmed
Economically complete: not yet

An agent under mandate

An AI agent holds a key to a treasury. The key says nothing about what it may do.

In the FxChain design the agent acts under a declared mandate - bounded in scope, amount and time - that is checked before anything executes. When the mandate is exhausted, the key is still valid and the authority is gone. When the evidence a claim needs is missing, the refusal is to be sealed as an artefact the institution can carry to its auditor: what was declined, at which boundary, for which reason.

Key valid: yes
Mandate in scope: no - exhausted
Outcome: refused before execution
Refusal: sealed, reasoned, portable
Money moved: none

What exists today is smaller: an internal lab on synthetic data, where every request must fit a mandate with a bounded action, a ceiling and an expiry, a refusal leaves no effect and every attempt leaves a receipt. The FxChain core is illustrated there, not executed: the supplier invoice review.

These design scenarios do not execute the FxChain core. The five downstream settlement facts are distinct from the six-axis project assessment and from a payment-evidence model that also records Signed. A technical signature alone does not establish legal consent. The linked supplier example uses synthetic data.

A bounded evaluation

Measure the decisions, including unresolved ones.

Agree the cases, observable evidence and expected outcomes before comparing FxChain with a competent conventional procedure.

Define the failure and the authority

Use a lost provider response, an exhausted agent mandate or restoration from an earlier state. Identify which actions are allowed, which facts remain unknown and which additional evidence could justify continuation.

Compare practical outcomes

Examine justified decisions, unresolved cases, false refusals, manual work, integration cost, added risks and evidence quality. Measure exposure and duration only where the inputs and observation make them meaningful. No customer result or active pilot is implied.

Where it lands in your stack

Map the questions to your operating systems

These nine examples are a discussion map, rather than a claim of complete institutional coverage. Existing systems already provide access controls, approvals, idempotence, logs and reconciliation; an evaluation must identify where their composition needs stronger evidence.

Payment hub and rail gateways

Evaluation question

A provider response may be missing or conflict with another record; an operator must decide whether to recover, wait, reconcile or escalate.

What stays apart here

The design keeps five downstream settlement facts distinct: executed, finalised, delivered, legally effective and economically complete. Their evidence and unresolved states remain separately inspectable.

Direction (research)Receipts, denials and capsules

Treasury and TMS automation

Evaluation question

An API credential may be broader or longer-lived than the authority intended for a particular agent action.

What stays apart here

A key is not a mandate. Authority is a capability with scope, amount and time that can only narrow; an exhausted mandate leaves a valid key with no power, and a refusal sealed for the auditor.

Local mechanism; institutional integration remains a research direction.

Local mechanism; integration unqualifiedThe authority algebra

Custody and securities settlement

Evaluation question

A "settled" flag hides a cash leg still pending on a bank rail and a legal effect nobody has confirmed.

What stays apart here

Delivery versus payment kept as separate facts, sealed separately, with the external rail an external claim bound by reconciliation rather than assumed.

Direction (research)FxBlock

Finance operations and reconciliation

Evaluation question

An operations team may need to join evidence from several systems while protecting the detail behind a refusal.

What stays apart here

Receipts of what execution observed; denials with a public reason class and claimant-scoped detail; a refusal that travels to the auditor with its rule, its missing evidence class and its remediation path.

Local mechanism; institutional integration remains a research direction.

Local mechanism; integration unqualifiedReceipts, denials and capsules

Compliance and screening

Evaluation question

A decision may depend on a changing policy, an external assessment and the time at which each was applicable.

What stays apart here

A compliance determination is an external claim with a source, an epoch, a disclosure class and a revocation; the protocol verifies its presence and binding, never performs legal judgement; bitemporal lineage answers "why was rule R active at coordinate C".

Direction (research)Privacy by disclosure class

Internal control and change management

Evaluation question

A green pipeline alone may leave the deployed software version, effective controls and reviewed evidence unresolved.

What stays apart here

The acceptance discipline, practised on every crossing: falsifiers, negative controls, hostile review; and Sovereign Assurance & Attestation (SAA), closed on 28 September 2026: the version at the head of the line, its controls and its evidence are on the record.

Practised on the lineThe acceptance discipline

Risk, limits and collateral

Evaluation question

Risk inputs can be incomparable, stale or incorrectly treated as permission to broaden an action.

What stays apart here

Risk is attributed evidence that may constrain exercise, never expand a grant; integer tiers, not floating-point formulas in consensus; incomparable dimensions stay incomparable; every change of power carries its duty.

Direction (research)The authority algebra

Data platforms and regulatory reporting

Evaluation question

A reporting projection can drift from its underlying record or lose the assumptions needed to reconstruct it.

What stays apart here

Canonical truth and projections are different species: every view is reconstructible and non-authoritative by type; state itself is a projection over sealed evidence, and a bounded preserved history has been reconstructed through separate oracle and engine paths under internal review.

Local mechanism; institutional integration remains a research direction.

Local mechanism; integration unqualifiedThe replay domain

Incident management and audit

Evaluation question

An incident correction needs a retained reproducer and provenance, so that later changes can be checked against the same failure.

What stays apart here

The engineering discipline calls for reproducible failures, negative controls and retained regression evidence. Corrections are additive to sealed history; the dossier states the scope in which that rule was exercised.

Practised on the lineThe evidence algebra

Why this is a system, not a feature list

The common design links a bounded capability, evidence, a retained refusal and distinct settlement facts. Reusing those mechanisms may reduce semantic gaps between systems. Qualification of one local mechanism does not qualify the next row, its connector or their composition.

Statuses identify local mechanisms, engineering practice or research direction. Institutional integration, an active pilot and a measured operating benefit each require their own evidence.

By question

How this differs, asked the way a risk committee asks it

Six questions to test against a competent existing procedure on the same inputs and failure cases. The baseline varies by implementation; these examples do not characterise all ledgers or providers.

Question	Common practice	FxChain	Status
Is settlement one word or five facts?	Some workflows expose one status; a competent baseline can also maintain distinct execution, delivery and reconciliation records.	Five facts kept apart and sealed separately; missing evidence recorded as missing, never as false.	Direction (research)
Is a refusal an error or a result?	Error codes and audit logs can retain refusal reasons; their portability and disclosure controls depend on the implementation.	A typed result. Proven locally: every organ refuses at an explicit, typed boundary. A sealed denial with a public reason class and claimant-scoped detail, past the admissibility threshold, is direction.	Documented local scope
Is authority a role or a capability?	Roles, scoped credentials and approval workflows can bound authority; their composition must be checked for the chosen operation.	Capabilities, not roles. Proven locally: a named, counted authority exhausted inside its act, and a machine-checked algebra in which composition cannot escalate. Scoped, expiring, revocable capabilities with provenance are direction.	Documented local scope
Is finality a number or a function?	Confirmation policies may differ by chain, asset, transaction value and external settlement dependency.	Declared planes derived from the effects a claim declares, each with its own window and caps.	Direction (research)
Is state a substance or a projection?	Ledgers, event logs and derived views assign authority differently; the chosen baseline must state which record controls a decision.	A projection over sealed evidence; any cache is non-authoritative by type; the core stays small.	Direction (research)
Does the history ever get rewritten?	Systems differ in how they retain corrections, reorganisations and migration lineage.	For the sealed local subjects, corrections are additive: frozen bytes and their refusal records remain preserved.	Documented local scope

The systems the architecture programme reads as frames of reference - public chains, permissioned ledgers and messaging rails - are named in its sources, not characterised here. The organ-level version of this table is on Technology.

Boundaries

The boundary of an evaluation

An institutional relationship starts with the limits stated plainly.

This portal offers no live financial service. A technical profile or product design does not establish regulatory authorisation.
No custody, no accounts, no payments are offered on any surface.
There is no token and no investment product of any kind.

Engagement today means design conversations and staged partnerships - in writing, with explicit boundaries.

Bring a workflow, its constraints and one failure case. The discussion identifies the decisions to justify, the evidence required and the properties still to be evaluated.

Dedicated pages per audience (compliance, treasury, infrastructure) ship in a later phase, each under the same discipline.

Bring one workflow and one failure case.

Define the action, the authority, the observable evidence and the point where recovery becomes difficult. We can discuss a bounded comparison with your operating procedure.

Discuss an evaluation The roadmap