Concepts of Open Commerce

Concepts of Open Commerce

Open commerce is the set of primitives required for buyers, suppliers, and their agents to transact across organisational boundaries under shared governance, decentralised identity, and cryptographic trust — independent of any single standard or platform.

This page defines those primitives. They stand on their own: you can read, cite, and reason about them without adopting SIGNET. SIGNET is the reference realisation of open commerce — the Canonical Data Model is what these concepts look like when made concrete, validated, and machine-checkable — but the concepts are the more durable thing. Where a primitive is realised in the CDM, this page points to it; the pointer is an illustration, not a dependency.

This page is non-normative. It frames the model; it does not define it. The normative artifacts are the JSON Schema (the source of truth), then the prose specification, then the rest of this wiki.

Why a concept vocabulary exists

A standard defines a format. A concept vocabulary defines what the format is for — the shared terms in which a whole category reasons about itself. Open commerce needs the second because its hard problems are not format problems: they are questions of authority (what may an agent do?), accountability (who decided, and can they show their working?), and auditability (can the history be trusted without trusting the holder of the history?). No existing procurement standard answers these, because they were designed for documents exchanged between people, not decisions taken by agents.

Defining this vocabulary raises a fair question: who appointed the definer? The answer is the same governance firewall that governs certification. The concepts are stewarded by Concert Foundation as a neutral body that holds no proprietary claim over them. Defining the vocabulary is not the same as controlling its implementations:

• The concepts and the CDM that realises them are dedicated to the public domain (CC0 1.0) — anyone may implement them, anywhere, with no attribution burden and no permission sought.
• Conformance is certified on identical terms for every implementer, by a public, machine-runnable suite (neutrality rules CN-1…CN-4). No subjective gate, no private suite, no preferential path.
• No commercial operator — including Score Networks — holds a privileged definition of any concept on this page, an early-access route to the spec, or a relaxed conformance bar.

A vocabulary that anyone may implement and that certifies everyone identically is open in the way that matters: stewardship without ownership. That distinction is the whole of the position — see Governance & Versioning.

The nine primitives

Open commerce is built from nine concepts that fall into three groups. One describes the network as a whole (governed network). Four describe what makes an agent's participation governed (accountable agent, bounded authority, decision-as-record, policy-as-data). Two are network-formation primitives — how parties come to transact at all (discovery, and consent over their data). And two are lifecycle primitives — how a transaction is closed and how its history stays trustworthy (settlement, event-sourced trust). Each is defined standard-agnostically, then mapped to its SIGNET realisation.

1. Governed network

A commerce network is governed — rather than merely automated — when authority, accountability, and auditability are structural properties of the network, not procedures bolted onto it.

Automation makes a network fast. Governance makes it answerable. The distinction is whether, for any action the network took, you can mechanically establish what authority permitted it, who is accountable for it, and whether the record of it has been tampered with. If those answers live in process documentation and human attestation, the network is automated. If they live in the data the network already produces, it is governed.

→ Realised in SIGNET as the four-layer Canonical Data Model: Foundation, Process, Agent, and Trust. The Agent and Trust layers are what make a SIGNET network governed rather than only interoperable.

2. Accountable agent

A synthetic agent is a first-class party whose every material action records who acted, under what authority, from what inputs, and why.

The defining concept of open commerce, and the one no existing procurement standard carries. Conventional automation treats an agent as invisible machinery behind a human's identity. Open commerce treats the agent as a named participant that can be held to account on the same footing as a person or an organisation — with the same standing to act and the same duty to explain. Accountability is not a property the agent has; it is a property of the record its actions produce.

→ Realised in SIGNET as the SyntheticAgent (a Party with partyType: agent, whose operatedBy field names the party accountable for it) and the Decision record its actions emit.

3. Bounded authority

Authority is data: an explicit, machine-checkable statement of what an agent may do, within what limits, and above which a human must approve.

An agent without bounded authority is either a liability or a toy. The concept that resolves this is the mandate — a declaration, readable by both the agent and an auditor, that enumerates the agent's permitted capabilities, the hard constraints it must respect, and the thresholds above which autonomy yields to human approval. An agent acting outside its mandate is, by construction, non-conforming — not because a rule was broken after the fact, but because the authority to act was never present. Bounded authority is what lets agent autonomy and human oversight coexist instead of trading off.

→ Realised in SIGNET as the Mandate: permittedCapabilities, hard constraints, and approvalThresholds that force human-in-the-loop above a ceiling. The agent demonstration shows it running — a €12M award bounded by a €10M autonomous ceiling requires human approval before it can proceed.

4. Decision as record

A decision is an artefact, not a log line — a single object recording what was decided, by whom, under which mandate, from which inputs, under which policy, with what rationale, and with what human approval.

The difference between a system that did something and a system that can account for what it did is the decision record. An event log says a thing happened; a decision record says why it was permitted, what evidence it rested on, which rule governed it, and who — human or synthetic — stands behind it. This is the backbone of accountability, and it is precisely the artefact that regulation increasingly demands of automated and AI-assisted decisions. The record is the same whether a human or an agent decided; that symmetry is the point.

→ Realised in SIGNET as the Decision object — see the worked award decision.

5. Policy as data

The rule a human reads is the same rule the agent executes. A policy carries both an executable form and a human-readable form of the identical rule, and both are mandatory.

When the rule an agent enforces is code, and the rule an auditor reviews is prose, the two drift — and the drift is invisible until it causes harm. Open commerce closes the gap by making policy a dual-form artefact: the executable expression and the plain-language statement are two faces of one object, versioned and governed together. "Rules as code" becomes auditable rather than opaque. The eligibility threshold, the evaluation weighting, the approval routing, and the agent's own constraints are all this kind of object.

→ Realised in SIGNET as the Policy object, whose expression (e.g. Rego) and humanReadable statement are both required — see the worked evaluation policy.

6. Consent as data

The right to access data is separable from the assertion that the data exists. Access is a granted, bounded, revocable artefact — not an assumption.

Open commerce does not assume a central warehouse that holds everyone's data and adjudicates who may see it. Instead, the existence of a document and the right to read it are different facts: a record can assert "this evidence exists" while the data itself stays with its owner, released only under an explicit grant to a named party, for a stated purpose, for a bounded time. Data sovereignty becomes a property of the model rather than a policy promise.

→ Realised in SIGNET as the Consent grant and the Document.accessGrant field that points to it (the Solid consent pattern, design principle 1.5).

7. Discovery

An agent can find a counterparty and verify what it claims to be able to do across an organisational boundary, without a central registry — capability is advertised, and the advertisement is checkable.

The previous primitives describe how a governed transaction works once the parties are already in a room together. Discovery is how they get into the room. In a network with no central broker, an agent must be able to locate who to transact with, learn what another agent or organisation can do, and verify those claims against evidence rather than taking them on faith — all without asking a registry's permission. Two things make this open rather than closed: capability is advertised at a resolvable, self-hosted location the party controls (no gatekeeper owns the directory), and an advertised capability or attribute is verifiable — a claim of "ISO 27001 certified" or "can evaluate submissions" can be checked against a credential or a declared interface, not merely asserted. Discovery without verification is a phone book; discovery with it is the basis of trust between strangers.

→ Realised in SIGNET as the AgentCapability model (declared, discoverable capabilities aligned to the A2A Agent Card), advertised at the agent's self-hosted agentCard location (/.well-known/agent.json), with claims about a party verified through Credentials — optionally via selective disclosure, so a capability can be proven without exposing the whole certificate. The Catalogue is the analogous advertisement for goods and services rather than agent capability.

8. Event-sourced trust

The current state of any object is the projection of an append-only, hash-chained event stream. The audit trail is the system of record; tampering is evident rather than prevented by trust.

A mutable database asks you to trust whoever holds it. Open commerce replaces that ask with a structural guarantee: every material change is an immutable event, each event hash-chained to its predecessor, so that altering any point in the history breaks the chain and is detectable by anyone. State is derived by replaying the stream, which means the history cannot quietly disagree with the present. Audit stops being something you conduct and becomes something the network simply is.

→ Realised in SIGNET as the Event (append-only, previousEventHash chained) and the Provenance embedded in every Event and Decision (design principle 1.7).

9. Settlement

Commerce is the making and discharging of commitments. A commitment, once made, has a lifecycle the network tracks to fulfilment — with evidence — and the loop is closed explicitly, not assumed.

A transaction is not finished when it is agreed; it is finished when what was promised has been delivered, evidenced, and accounted for. Open commerce treats this as a first-class concern: every obligation arising from an agreement has a state the network can observe — outstanding, met, breached, or waived — and reaching the "met" state is backed by evidence, not by a party's say-so. This is the primitive that distinguishes a network that records intentions from one that records outcomes. It also closes the accounting loop: the invoice that demands payment, and the structured record that the demand was satisfied, are the final discharge of the commitments the rest of the model set up. Settlement is where governance is vindicated — the point at which the network can show not only that a decision was authorised and accountable, but that the thing decided actually happened.

→ Realised in SIGNET as the Obligation object (a deliverable or milestone with a status of pending/met/breached/waived and evidence of fulfilment), embedded in the Contract, and the EN 16931-aligned Invoice that closes the financial loop — itself convertible to Peppol BIS Billing and natively ViDA-aligned (see EN 16931 & ViDA E-Invoicing). The loop is closed explicitly, not assumed: Obligation.dischargedBy references the settling artefact, and Invoice.settles references the obligation it discharges, so "this obligation was discharged by that invoice" is a link you can mechanically traverse — not an inference. The transition is recorded as an immutable obligation.discharged Event, so discharge is auditable, not merely declared. (Invoice.settles is SIGNET-original and is omitted on the Peppol BIS projection, so closing the loop costs nothing in ViDA convertibility.)

The two cross-cutting commitments

Two commitments run through all nine primitives and are what make this open commerce rather than a well-designed platform.

Profile-and-bridge, not fork. Open commerce adds what existing standards lack — governance, agency, trust — and maps without loss to and from what they already define. It does not replace OCDS, EN 16931, UBL, or Peppol; it profiles them and fills their gaps, and its original contributions stay cleanly separable so they never entangle the standards they bridge. This is the anti-lock-in commitment: adopting open commerce never strands your data inside it. → Standards Mapping.

Vocabulary is public. The concepts and the model that realises them are public-domain; only the marks ("SIGNET", "Concert", "SIGNET Certified") are administered, and only so that a conformance claim means something. You never need permission to use the vocabulary — only to claim certification against it. → Governance & Versioning.

Relationship to the standards landscape

Open commerce concepts sit deliberately above the established procurement standards, not beside them. OCDS defines the lifecycle; EN 16931 defines invoicing semantics; UBL and Peppol BIS define document syntax; the EU eProcurement Ontology defines cross-cutting semantics. Each is authoritative in its layer — and each leaves governance, agency, and trust open. Open commerce occupies that open space and bridges down into the others, which is why a SIGNET document can always be projected to a conforming OCDS release or a Peppol BIS invoice, and why the Agent Layer simply drops away when it does.

For the field-by-field crosswalk, see Standards Mapping. For how the concepts support real regulatory obligations — UK Procurement Act 2023, EU ViDA, EU AI Act — see the Regulatory & Compliance Map.

Where to go next

• Introduction & Concepts — what SIGNET is and the eight design principles that realise these primitives.
• Architecture Overview — the four-layer model in detail.
• Agent Layer — primitives 2–5 made concrete.
• Regulatory & Compliance Map — the concepts mapped to the obligations adopters already carry.