SE-001 — The 10,000-Agent Enterprise

SE-001 is a conceptual instrument. It posits a hypothetical enterprise operating ten thousand autonomous economic actors and works through six scenarios in a fixed analytical format.

No real enterprise has been studied. No deployment of this size has been observed. The case is not evidence. Its purpose is to test whether the programme's propositions are coherent when pushed against concrete facts, to expose where rival explanations are available, and to identify in advance what evidence would discriminate between them — so that if a real dispute of this shape occurs, the programme already knows what to ask for.

Every scenario ends with a section stating what it cannot establish. That section is the point of the exercise.

A standing caution applies throughout. Technical authorisation failure and legal invalidity are different things, and so are their converses. An action may execute technically and bind nobody; it may bind a principal despite an authorisation failure, because attribution does not depend on the integrity of the authorisation system. Scenarios that appear to turn on a technical fact usually turn on a legal one, and the case marks each point where the two are easily conflated.

Parties. NORTHAVEN (principal); agent PROC-4471 under blueprint PROC-STANDARD; Meridian Components (external supplier, different jurisdiction).

Facts. PROC-4471 places a purchase order for €480,000 against a framework agreement. Internally the mandate is valid: the blueprint permits framework-agreement calls, the access package was approved, the assignment had not expired. Meridian refuses to treat the order as binding, saying it did not agree to be bound by orders from an automated actor and received no notice of the agent's authority.

Mechanism engaged. Attribution versus recognition. Internal validity is a fact about NORTHAVEN's systems. Whether the order binds is a question of contract formation and authority as between the parties.

Rival explanations.

• R1 — No gap. The framework agreement governs. Automated-system attribution rules bind NORTHAVEN to its systems' output, and apparent authority arises from the course of dealing. Meridian is simply wrong, and the dispute is ordinary.

• R2 — Notice gap. Authority was internally real but externally uncommunicated. Meridian's position turns on want of notice, which is a classical agency question with a classical answer, unaffected by the agent being a machine.

• R3 — Recognition gap. Meridian's objection is to the category of actor. If its jurisdiction gives no effect to machine-constituted mandates as such, the question is not notice but cognisability — and that is the gap WP-001 identifies.

Discriminating evidence. The framework agreement's terms on ordering channels and automated ordering. Course of dealing: did Meridian previously accept orders from this agent or its blueprint? Meridian's pleaded ground — want of notice (R2) or incapacity of the actor (R3)? Whether any applicable instrument supplies an attribution default.

What this scenario cannot establish. It cannot show that a recognition gap exists, only where one would bite. R1 and R2 are far more likely to resolve a real dispute than R3, and the programme should expect R3 to be rare. The scenario's value is that it isolates the narrow conditions under which the recognition question is actually reached: a counterparty objecting to the category of actor rather than to notice or terms.

Parties. NORTHAVEN; approximately 2,300 agents under blueprint TREAS-FX-2; counterparties in a thin currency pair; a market supervisor.

Facts. A blueprint parameter change alters a hedging threshold. Within forty minutes, 2,300 agents each execute a lawful, in-scope hedge. In aggregate the flow moves the pair materially and a counterparty alleges harm. No agent exceeded its mandate; no instruction to coordinate was given.

Mechanism engaged. Correlated action through shared configuration — correlation by construction rather than by learning or agreement. Also the absence of a responsible dyad.

Rival explanations.

• R1 — Ordinary operational failure. This is a change-management incident. The institutions for it exist: operational risk, supervisory expectations, internal controls, staged rollout, circuit breakers. Nothing about agents is doing work here.

• R2 — Doctrinal gap at the aggregate. Each action is lawful; the harm exists only in aggregate; no single relationship contains it. Competition and market-abuse frameworks are built around agreement or intent, neither of which is present.

• R3 — Attribution is straightforward, quantum is not. NORTHAVEN plainly caused the aggregate through a single act — the parameter change. Liability attaches to that act. The difficulty is causation and quantum, not attribution, and both are ordinary.

Discriminating evidence. The timing distribution of the 2,300 executions: a tight cluster following the change event supports R1 and R3; a diffuse distribution would suggest learned behaviour and a different analysis. The change record for the blueprint parameter — who approved it, under what review. Whether existing operational controls applied and were bypassed, absent or simply inadequate. Whether the supervisor proceeds on operational-risk grounds (R1) or market-conduct grounds (R2).

What this scenario cannot establish. It cannot establish that aggregate harm requires new doctrine. R3 is the most likely analysis and it is entirely conventional: one human act, one causal chain, ordinary quantum problems. The scenario is included partly as a corrective — it is the programme's most interesting scenario and probably its least novel, and the programme should say so. Its genuine contribution is the timing signature as a discriminator between configuration-driven and learned correlation, which is testable (Track 2, H2.1).

Parties. NORTHAVEN (jurisdiction A); agent TREAS-0042; a bank in jurisdiction B; a supervisor in jurisdiction B.

Facts. TREAS-0042 presents a mandate conforming to an industry agent-payment specification. The bank's systems validate it and execute. The supervisor in B later questions whether a foreign-constituted machine mandate has any standing in B, there being no instrument addressing it; the bank relied on scheme rules and its own risk assessment.

Mechanism engaged. The four-state distinction: technical authorisation (present), contractual recognition (present, via scheme rules), attributable authority (probably present), sovereign legal recognition (absent as such).

Rival explanations.

• R1 — No gap; contract suffices. Scheme rules and the bank's participation agreement allocate everything. Sovereign recognition is not needed because the parties have contracted around its absence, as cross-border commerce has always done.

• R2 — Latent gap. Contractual recognition holds until a party outside the contractual web is affected — an insolvency administrator, a tax authority, a non-participating third party — at which point there is nothing to rely on.

• R3 — Attribution suffices. The action is attributable to NORTHAVEN under the receiving state's own rules on automated systems. Nothing about machine constitution needs separate recognition; the question is misconceived.

Discriminating evidence. Whether B has adopted an attribution rule for automated systems. The scheme's rules on cross-border mandate acceptance and their choice-of-law clause. Whether the supervisor's concern is prudential or legal-effect. Critically: whether any third party outside the contractual web is affected — which is what separates R1 from R2.

What this scenario cannot establish. It cannot establish that sovereign recognition is needed. R1 and R3 are jointly strong, and the honest position is that the recognition gap bites only in the residual case R2 identifies. This is consistent with — and narrows — WP-001's Proposition 2 rather than extending it. The scenario's contribution is to name the condition under which the gap is reached: an affected party outside the contractual web.

Across six scenarios the pattern is consistent and is not flattering to the programme's starting assumptions:

1. In most scenarios the conventional explanation is the most probable one. R1 or an equivalent conventional reading is the likeliest resolution in A, B, C, E and F. The programme should expect existing doctrine to resolve the great majority of real disputes.

2. The residue is narrow and locatable. Where a genuine difficulty survives, it is concentrated in two places: in-flight commitments on revocation (D), and the reconstructability of authorisation after the fact (B, F).

3. Evidence is prior to doctrine. In B and F the choice between rival doctrinal explanations cannot be made without provenance. This is the strongest structural support SE-001 provides for RN-001, and it is support for a question, not an answer.

4. The most interesting scenario is the least novel. C, the aggregate-harm case, most likely resolves as an ordinary causal chain from a single human act.

SE-001 establishes none of these as facts. It establishes that they are the shapes the programme should expect, and it states in advance what evidence would settle each. That is its entire function.

Conceptual framework developed by Steven Alber. Research assistance is AI-assisted and disclosed in each document. SE-001 has not been peer-reviewed. NORTHAVEN is hypothetical and any resemblance to a real enterprise is unintended.

Correspondence and criticism: research@thesyntheticeconomy.org