Positional Causal Boundary

In plain language: “What the position confers” — the reach the Agent inherits by occupying a specific node in a Topology.

Definition

An Agent’s Positional Causal Boundary is the part of its Causal Boundary that the Agent has because of where it sits in a Composition’s Topology. It is the reach the position confers — the set of causal capabilities the Agent has only because of its current position, not because of anything about the Agent itself. A different position would confer a different set — though the two sets may overlap, since positions in the same Topology often share some of the same reach. A VP’s authority to approve budgets and reassign headcount is positional; a biological cell’s ability to influence specific neighbors in a specific tissue is positional; an AI agent’s access to production systems in a specific deployment is positional. The Positional Causal Boundary is part of the Causal Boundary; together with the Intrinsic Causal Boundary it sums to the full Causal Boundary.

The Positional Causal Boundary at Capacity is a property of the position, not of the Agent occupying it. The same VP chair confers the same budget authority regardless of who sits in it. When the Agent leaves, the Agent loses that Positional reach, but the Topology retains it — the next Agent inheriting the position gets the same Positional Capacity, or the organization reshapes to redistribute the reach across other positions. This is what distinguishes Positional from Intrinsic. Intrinsic reach comes from the Agent itself — the Agent’s own body, skills, and capabilities. Positional reach comes from the Topology — it exists at the position before the Agent arrives and remains at the position after the Agent leaves.

What the Positional Causal Boundary is not:

• Not the full Causal Boundary. The Causal Boundary is the Agent’s total reach. Positional is the Topology-given part of that reach; Intrinsic is the Agent-given part.
• Not what the Agent actually does with the position’s reach — that is Actual Output, which depends on the Agent’s own capabilities across all three Boundaries. Two different Agents at the same position have the same Positional Capacity but may produce very different Actual Output. One may under-realize the position’s reach because his Cognitive Boundary doesn’t extend to the domain (he has the authority but doesn’t understand how to exercise it). Another may exercise reach beyond what he can model because his Frame exceeds his Cognitive Boundary — he thinks he understands the downstream effects of his decisions when he does not, and the Positional reach carries those decisions through to consequences no one anticipated. The Positional Capacity is the same in both cases; what the Agent does with it is not.
• Not limited to what the position was designed to confer. The designer of a position — the CEO who creates a role, the engineer who configures a deployment — has a Target Surface in mind: the effects the position is expected to produce. But the actual Positional Causal Boundary may be larger than the designer’s intent. Organizational authority has structural consequences beyond the job description. A VP of Product who controls the feature roadmap can also, as a consequence, determine which customer promises the sales team can keep — an effect that shapes revenue even though it is nowhere in the VP’s job description. An AI agent with write access to a code repository can also, as a consequence, deploy a change that breaks a production system the deployment designer never intended the agent to touch. The designer’s intent is the Target Surface; the Positional Causal Boundary is the actual reach, and the gap between them is the Franz Ferdinand Effect applied to organizational and system design.
• Not fixed for the Agent. Positional Causal Boundary changes when the Agent’s position changes — a promotion, a transfer, a redeployment. The change is tied to the Agent’s position in the Topology, not to any change in the Agent itself. This contrasts with Intrinsic, which changes only when the Agent itself changes.

Relations

Positional Causal Boundary is one of two sub-forms of the Causal Boundary; the other is Intrinsic Causal Boundary. Together they sum to the full Causal Boundary. The diagnostic value of separating position-given reach from Agent-given reach is high across domains — for an executive evaluating what survives a role change, for a deployment engineer evaluating what an AI agent can affect, and for an alignment researcher distinguishing what a model can do by its own capabilities from what its deployment context enables. The Causal Boundary definition describes the total; the Intrinsic and Positional definitions decompose it by source.

Positional is not a subset of Intrinsic, nor vice versa. They are complementary parts of the Causal Boundary, distinguished by source. Some effects are purely Positional (the authority to bind a company to a contract), some purely Intrinsic (making a signature with your hand), and some draw on both (actually signing a binding contract) — but even in the combined case, the two sources are analytically separable.

The gap between an Agent’s Positional Causal Boundary and its Actual Output is one form of Trapped Intelligence: Positional reach the Agent is not using. The cause of the gap is diagnosable across Boundaries. If the Agent doesn’t receive the information that would trigger action (Computational — the signal isn’t reaching him), the intervention is at the Computational layer. If the Agent receives the signal but doesn’t understand the domain well enough to act on it (Cognitive — he has the authority but not the knowledge), the intervention is at the Cognitive layer. If the Agent understands the situation but lacks the personal capability to execute (Intrinsic Causal — he has the positional authority and the domain knowledge but not the skill), the intervention is at the Intrinsic layer. Same symptom — the position’s reach is under-realized — three different Boundary-level causes, three different interventions.

Target Surface is the framework object that names the designer’s intent for a position. The gap between Target Surface and Positional Causal Boundary at Capacity is one diagnostic of organizational design risk — the position enables more (or different) effects than the designer intended. The gap between Target Surface and Actual Output is the diagnostic of Agent-position fit — the Agent is not producing what the position was designed to produce.

A Causal-side Tool (gun, lever, broadcast antenna) extends the Causal Boundary; whether the extension is Intrinsic or Positional depends on whether the Agent has access to the Tool independently or only through Topology position. A personal firearm extends Intrinsic reach. A corporate broadcast channel the Agent uses only as the company’s spokesperson extends Positional reach — the access stays with the role when the Agent leaves.

Frame applies. The Agent’s self-model of his Positional Causal Boundary can be wrong in either direction. An executive may overestimate his Positional reach — thinking he can influence the board’s strategy when his position carries no mechanism to do so. An executive may underestimate his Positional reach — not realizing that his budget decisions structurally shape the company’s talent trajectory for the next two years. The mismatch between Frame and Positional Boundary is one source of the Franz Ferdinand Effect: the Agent exercises Positional reach he can’t cognitively model, producing consequences he cannot predict.

Actual Output cuts across both Intrinsic and Positional parts — at any given moment, what the Agent is in fact producing draws on both substrate-given and Topology-given reach. The distinction between Intrinsic and Positional is analytical; the Agent’s actual effects are a single stream.

Example — CEO

A CEO creates a new VP of Product role and writes the job description: own the product roadmap, coordinate with engineering and sales, approve feature priorities, and manage a team of twelve product managers. This job description is the CEO’s Target Surface for the position — the effects the CEO intends the position to produce.

The Positional Causal Boundary of the role is larger than the job description. The VP of Product, by virtue of sitting at the intersection of engineering and sales in the org chart, also has the structural ability to influence which customer feedback reaches engineering (because product managers control the backlog), to shape which deals the sales team can credibly promise delivery on (because feature timelines run through his team), and to affect the company’s technical debt trajectory (because roadmap prioritization determines what gets built now versus later). None of these effects are in the job description. All of them are real consequences of the position’s reach in the Topology — the Positional Causal Boundary exceeds the Target Surface. This is the Franz Ferdinand Effect at the organizational-design level: the CEO designed the role to do X, Y, Z, but the role’s structural position enables W, V, and U as well, and nobody modeled those.

Now the CEO hires two candidates sequentially for the role. The first VP has deep technical knowledge but limited political instincts. His Intrinsic Causal Boundary includes the ability to evaluate architectures, write clear specifications, and debate engineers on technical merits. His Positional Capacity is the same as anyone else in the chair — the budget authority, the headcount, the org-chart position. But his Actual Output under-realizes the Positional reach: he uses the technical parts of the role effectively but doesn’t exercise the cross-functional influence the position structurally enables, because his Cognitive Boundary doesn’t extend to organizational politics. The position could do more; the Agent doesn’t know how to use it.

The second VP has strong political skills but limited technical depth. His Intrinsic Causal Boundary includes the ability to read a room, build alliances, and negotiate across departments. Same Positional Capacity — same chair, same authority. But his Actual Output is shaped differently: he exercises the cross-functional influence aggressively, reshaping how engineering and sales interact, but his roadmap decisions are technically weak because his Cognitive Boundary doesn’t extend to architecture. He approves a feature prioritization that creates six months of technical debt he cannot model — the Franz Ferdinand Effect, amplified by Positional reach. The position carries the decision through to forty engineers’ work, and the VP’s Frame told him the decision was sound.

Same position. Same Positional Capacity. Different Agents. Different Actual Output — because each Agent’s Intrinsic capabilities and Cognitive Boundary shape how the Positional Capacity is realized in practice.

The diagnostic value for the CEO: when a position is under-performing, the framework separates two questions that are easy to confuse. First: is this the right role? If the Positional Capacity itself is wrong — the role doesn’t have the authority it needs, or it has authority in the wrong areas — that’s a Topology problem, and the fix is redesigning the position. Second: is this the right person in the role? If the Positional Capacity is right but the Realization is wrong — the person has the authority but isn’t using it, or is using it in ways that produce unmodeled consequences — the cause is in the Agent’s own Boundaries, and the intervention depends on which Boundary is the bottleneck.

Example — Research

An AI agent — a language model plus its operational harness (system prompt, provisioned tools, memory, constraints) — has an Intrinsic Causal Boundary defined by what it can produce from its own capabilities: generate text, reason over inputs, follow instructions, use the tools built into its harness. This Intrinsic Causal Boundary travels with the agent across deployments.

The same agent deployed in two different contexts has two different Positional Causal Boundaries. Deployed as a customer service chatbot, the agent’s Positional reach includes: responding to customer queries, issuing refunds through the payment API, escalating tickets to human agents, and updating customer records in the CRM. Deployed as an autonomous coding assistant, the same agent’s Positional reach includes: reading and writing files on a developer’s machine, running shell commands, creating pull requests, and deploying code to staging or production environments. The Intrinsic capabilities are unchanged — same model, same harness. The Positional reach is entirely different because the deployment Topology determines which systems the agent can touch.

The Positional Causal Boundary at Capacity is a property of the deployment, not of the agent. The APIs are provisioned, the permissions are set, the system integrations exist before the agent is plugged in. A different agent deployed in the same infrastructure inherits the same Positional Capacity — same API access, same permissions, same system connections. What differs is what the agent does with that access, which depends on the agent’s own Intrinsic capabilities and Cognitive Boundary.

Under-realization applies: the agent may have access to Tools it never uses because its Frame doesn’t include them — the tools are provisioned in the deployment but not described in the agent’s instructions, so the agent doesn’t know they exist. This is the exact parallel to the SaaS Trapped Intelligence case from the Cognitive Boundary definition: the capability is there, the access is there, but the Agent’s Frame doesn’t include it, so the Positional reach goes unused. A configuration change — adding the tool to the agent’s instructions — moves the tool inside the Frame and the Positional reach expands at Realization immediately, without any change to the underlying Positional Capacity.

The Franz Ferdinand Effect applies with particular force. A deployment engineer configures an autonomous coding agent with write access to a production repository, intending the agent to make small, well-scoped changes. The deployment’s Positional Causal Boundary includes everything the agent can do with that write access — including deploying a breaking change, overwriting critical configuration, or triggering a cascading failure in dependent services. The engineer’s Target Surface was “small, well-scoped changes.” The actual Positional Capacity is much larger. Whether the gap produces damage depends on the agent. A more capable agent with better Cognitive modeling of downstream effects will exercise the Positional reach cautiously, recognizing what it doesn’t understand. A less capable agent whose Frame exceeds its Cognitive Boundary — one that believes it understands the deployment architecture when it does not — will exercise the same Positional reach without recognizing its own limits, producing consequences neither the agent nor the deployment engineer modeled. The less capable agent is not acting maliciously; it simply cannot see the gap between what it thinks it understands and what it actually understands. The damage comes from the structural mismatch between Frame and Boundary, not from intent.

The practical implication for deployment architecture: constraining the Positional Causal Boundary — sandboxing the agent in a container, restricting communication to specific interfaces, limiting which systems the agent can write to — is a Topology-level intervention that reduces the maximum blast radius regardless of which agent occupies the position. This is the deployment equivalent of the CEO who narrows a role’s authority to contain risk: the Positional Capacity is deliberately made smaller so that even an Agent with poor Frame-Boundary alignment cannot produce catastrophic effects.