Capital markets have a small number of recurring roles. Everyone recognizes them: the person who owns the capital, the person who decides where it goes, the person who closes the deal, the person who does the scoped work. These groups appear across every asset class, every geography, every era. A pension fund in Oslo and a family office in São Paulo both delegate to allocators who hire bankers who retain lawyers. The labels vary. The structure does not.

The standard explanation is institutional convention — “that’s just how it works.” A better explanation is that these roles are not arbitrary groupings. They are the only groupings that hold together, forced into existence by one structural fact: the people doing the work control what the people paying for it can see.

Each role bundles multiple functions and captures rent on the entire package because the principal cannot see which component generated value. This essay shows why only four bundles are stable under that information regime and why other combinations decay toward them. When that control transfers to a system, the landscape reshuffles.

How to read this with the companion incentives essay:

  • This essay is the static map: the four stable bundles under agent-controlled information.
  • The incentives essay is the dynamic shock: what changes when information control transfers to systems.
  • Read this for equilibrium logic; read the other for repricing and the transition path.

I. Contracts, Not Titles

The groups are not defined by job titles or org charts. They are defined by contracts. Every contract in capital markets is built from four primitives:

Primitive Options
Pay basis Time, completion, or outcome
Risk exposure None (flat fee), partial (bonus/clawback), or residual (equity-like)
Decision rights Policy-bounded or discretionary
Information rights Agent-controlled, principal-controlled, or system-controlled

These are what actually drive behavior. Same title, different contract, different behavior. A property manager on a flat fee optimizes for ticket throughput. A property manager with NOI participation optimizes for tenant retention and cost control. The title is identical. The contract is not. When behavior surprises you, the contract changed even if the title did not.

Stability rules (quick heuristics):

  • Risk requires discretion.
  • Outcome pay requires verifiable information.
  • Completion pay cannot carry full residual risk.
  • Time pay plus discretion requires tight monitoring or it drifts.
  • Agent-controlled information makes outcome pay and risk hard to enforce.

Note on information rights: principal-controlled information is usually periodic, audit-based, and still dependent on the agent’s systems. It improves oversight but does not change who controls the narrative. System-controlled information is continuous, independent, and attributable; it is the regime shift that opens new bundles.

II. The Four Bundles

Four bundles of these primitives recur across capital markets:

Providers (Capital Owners)

Primitive signature: Residual pay, full risk, delegates decisions, information access depends on monitoring capacity.

Examples: Pension funds, sovereign wealth funds, endowments, family offices, individual LPs in a fund.

Providers bear the ultimate upside and downside. They delegate because they lack the time, expertise, or operational capacity to make every decision. The provider’s delegation generates the need for the other three bundles.

Allocators (Decision Makers)

Primitive signature: Outcome pay (carry, bonus), partial to residual risk (GP commit, clawbacks), discretionary within mandate, high information but constrained by bandwidth.

Examples: Private equity GPs, hedge fund managers, venture capital partners, real estate fund managers, CEOs, CIOs at endowments.

Allocators decide where capital goes on behalf of providers. The allocator is the clearest example of a trust aggregator. A fund manager bundles deal sourcing, underwriting, asset management, disposition, and investor reporting into a single role.

The LP pays for the entire bundle — typically 2 percent management fee plus 20 percent carry — because they cannot see which component generated alpha. Did the fund outperform because of superior sourcing? Better underwriting? Skilled asset management? Fortunate timing on exits? The LP cannot disaggregate the contribution, so the allocator captures rent on the whole package.

This is not a market failure. It is the rational outcome when the agent controls the information about their own performance.

Wait — are CEOs really the same as fund managers? Yes. Look at the contract, not the title. A CEO decides where the company’s capital goes: hiring, capex, M&A, R&D, buybacks. They have discretion within a board-defined mandate. Their pay is outcome-linked — equity, bonus, termination risk. The board is the provider; the CEO is the allocator. The fact that one sits in an office tower and the other on a trading floor is institutional decoration. The primitive signature is the same.

Middlemen (Transaction Intermediaries)

Primitive signature: Completion pay, low risk (reputation only), discretionary on process, privileged information during deal window.

Examples: Investment bankers, commercial real estate brokers, placement agents, M&A advisors, mortgage brokers.

Middlemen are paid to close transactions. They do not own the long-term outcome. Neither side shares long-term risk with them because tracking attribution back to the middleman after close is too expensive. They bundle market knowledge, counterparty relationships, and process management into a single service — trust aggregation in a narrower, more transactional form.

Mercenaries (Task Intermediaries)

Primitive signature: Time or deliverable pay, no risk, policy-bounded, limited information.

Examples: Law firms, accounting firms, appraisers, property managers on flat fees, environmental consultants, title companies.

Mercenaries are paid for scoped work regardless of outcome. No risk means no reason to grant broad discretion. Limited discretion means limited information needs. Time-based pay follows because there is no measurable outcome to tie compensation to at the task level.

III. Why These Four and Not Others

The four bundles are not categories imposed from outside. They are the only combinations that hold together. Each bundle is self-reinforcing: its primitives lock each other in place. Other combinations create tensions that push the contract back toward one of the four.

Two instabilities demonstrate the pattern:

Discretion without risk exposure. If an agent can make consequential decisions and bears no downside, they will take excessive risk, shirk on effort, or optimize for personal convenience. The combination is unstable. It drifts toward allocator (add risk exposure to make discretion tolerable) or toward mercenary (remove discretion and tighten policy). The middle ground does not hold — unless something else substitutes for skin in the game.

Completion pay with meaningful risk exposure. If a middleman bears real capital risk extending beyond the transaction, the contract generates an attribution problem. Did the deal fail post-close because of the middleman’s actions or because the market turned? The two time horizons — transactional pay and ongoing risk — are incompatible at full risk. So risk exposure drops off, leaving the standard middleman bundle. Partial risk with a bounded accountability window (holdbacks, clawbacks) can bridge the gap only if the attribution problem is solved.

The same pattern repeats across the full primitive space. Outcome pay without information is unenforceable — the agent games the metrics. Time pay with broad discretion enables free-riding — the principal cannot distinguish good judgment from bad. These failures are what make the four bundles attractors rather than conventions. Every unstable combination decays toward one of the four existing bundles.

An exhaustive analysis of all 36 primitive combinations confirms this. The Contract Stability Atlas maps every combination — click any cell to see its primitive signature, why it holds or fails, and where it drifts.

Of the 18 with agent-controlled information, exactly four are self-sustaining. The other 14 fail for a version of the same structural problem: the agent controls the information the principal needs to enforce the contract. There is no hidden fifth or sixth bundle under the current information regime — but there are combinations in the system-controlled half of the space that become viable under a different one.

IV. What Holds the Landscape in Place

Agent information control is what holds the landscape in place.

The GP controls the quarterly reporting narrative. The middleman carries the relationship graph in their head. The mercenary controls the timesheet. The principal cannot cheaply verify decisions, effort, or contribution — because the agent is the gatekeeper of the relevant information.

This is the specific mechanism that forces the landscape into four points. When the agent controls the information, risk is unenforceable (the agent can obscure outcomes), outcome pay is gameable (the agent can shade the metrics), and discretion is unverifiable (the principal cannot tell whether it was exercised well or poorly). Remove agent information control and these instabilities disappear — which is why system-controlled information opens up new regions of the primitive space.

Monitoring cost is the reason agent control persists. The principal wants better information. They cannot afford it. A pension fund with five investment staff covering forty GP relationships has roughly two hours per week per relationship. The GP controls the information flow because the LP cannot afford to verify every decision.

Several reinforcing mechanisms — centralized liability, regulatory structure, reputational insurance, discretion as default — keep the landscape in place. These reinforce each other: information control centralizes liability, liability hardens regulation, regulation increases reliance on reputation, and reputation keeps discretion as the default. Weaken one and the others compensate. But weaken the information control itself — transfer it from agent to system — and the whole structure shifts.

Some combinations in the contract space stay broken even with perfect monitoring. Full residual risk requires discretion and outcome pay; full risk with time pay is punishment without agency. Completion pay is incompatible with full risk because the time horizon of accountability exceeds the deal window. These hard boundaries limit how far the frontier can expand.

What would falsify this:

  • Information control transfers to systems but bundle pricing does not change.
  • System-controlled attribution expands, yet new contract types do not appear in those domains.
  • Principals gain component-level visibility but still pay undifferentiated trust premiums.

V. Bundle Drift

Contracts are not static. The same role can drift toward different bundles over time. Toward mercenary — when the principal tightens policy constraints, reduces discretion, or shifts to time-based pay. Toward allocator — when the agent gets more discretion, outcome-based pay, or risk exposure. The job title stays the same. The contract changes. The behavior follows.

One person can occupy multiple bundles simultaneously. A founder deploys risky capital as CEO (allocator), protects personal wealth as an individual (provider), and pays lawyers by the hour (buying mercenary services). A wealth manager may be paid like a mercenary (AUM fee, no downside) while claiming allocator status. The contract, not the claim, reveals the actual role.

Quick diagnostic (use in the field):

  1. Who bears the downside?
  2. When do they get paid?
  3. Who controls measurement?
  4. Where does discretion live?
  5. What just became measurable or auditable?

VI. The Conditions Are Changing

If agent information control is the mechanism that forces the landscape into four points, and if that control is transferring from agents to systems, the landscape shifts.

This is not speculative. System-controlled information — continuous, auditable, attributable — is becoming feasible in structured, high-volume workflows. When context moves from agent memory to system of record, the principal can verify what happened at the edge without relying on the agent’s reporting. The instabilities that pushed every contract toward one of the four attractors stop operating the same way.

Trust aggregation becomes contestable. If the principal can observe per-component contribution, the bundled fee that currently captures rent on the whole package becomes contestable. The allocator’s 2/20 reflects the LP’s inability to disaggregate sourcing, underwriting, asset management, and reporting. Transfer information control to a system and each component becomes separately observable, separately priceable, separately contestable.

The role does not disappear. It gets repriced and narrowed to the components where information remains genuinely agent-held.

Concrete example: an LP gains system-level visibility into property-level NOI drivers and can attribute performance to leasing, capex timing, and operating costs. The LP keeps the GP for sourcing and structuring but replaces the property manager with a specialist paid on renewal outcomes and benchmarks asset management fees against peers. The GP bundle narrows to origination and judgment on exceptions; the observable components reprice or migrate.

The shift does not happen all at once. The first change is not disaggregation but transparency: the same bundle, with system-controlled information replacing the agent’s information monopoly. The allocator’s role is intact. Their information advantage is not.

As the principal gains visibility, each observable function gets benchmarked against alternatives, and fee pressure follows. Only then does disaggregation become rational — commodity functions migrate to specialized providers while the allocator narrows to the judgment components where information stays with the agent.

Specific combinations in the system-controlled half of the primitive space become stable for the first time — each one viable because system-controlled information resolves the instability that killed it under the old regime:

  • The Accountable Executor (outcome pay, partial risk, policy-bounded, system-controlled info) previously collapsed to time pay because outcome attribution at the task level was impossible; with system attribution, the contract holds.

  • The Direct Principal (residual pay, full risk, retains decisions, system-controlled info) previously failed because the principal made worse decisions than a competent allocator without adequate information; system-controlled information closes the gap.

  • The Accountable Intermediary (completion pay, partial risk, system-tracked post-close outcomes) previously shed risk because post-close attribution was too expensive; system tracking makes holdbacks and clawbacks enforceable in high-volume contexts.

These are not speculative additions to the landscape. They are the combinations that the exhaustive analysis found unstable under agent control and stable under system control.

The four roles are not permanent features of capital markets. They are the product of a specific information regime — one where agents control what principals can see. When the regime changes, the attractor landscape reshuffles.

What the new landscape looks like — which roles get repriced, how the disaggregation parallels the history of securitization, who captures value, and where the framework might be wrong — is the subject of the companion essay on incentives.

Research Map: Sources and claim-by-claim citations for this essay