Incentives — Research Map

Every load-bearing claim in the essay, followed by the research that supports or connects to it.

Load-Bearing Claims

Four roles are stable because agents control the information principals need
Completion-based pay biases toward action regardless of fit (the completion problem)
Time-based pay biases toward local optimization regardless of outcome (the scope problem)
Both persist because writing better contracts requires information the principal does not have
Software created systems of record without transferring information control
AI transfers information control from agents to systems
Discretion can be replaced by specification
The contract frontier expands when information becomes system-controlled
Frontier expansion triggers commoditization: legibility → comparability → competition → margin compression
Historical precedents confirm the pattern (Bloomberg, electronic trading, programmatic ads, legal tech)
The GP is a trust aggregator — captures rent on the bundle because the LP cannot disaggregate
AI makes per-component contribution observable, making the bundled fee contestable
The GP role gets repriced and narrowed to the illegible residual
Securitization 1.0 disaggregated the bank’s role by transferring credit information to the structure
In performing securitizations the servicer is a commodity; in NPL securitizations (ERLS NPL1) the servicer is critical and differentiated
Securitization 2.0: standardize the decision-making, not just the cash flows
The operating protocol as the securitized asset
Cognitive throughput is the binding constraint — removing it expands the market (Jevons dynamic)
Securitization creates moral hazard when origination separates from ownership
Transition pathway: monitoring → repricing → disaggregation
The Accountable Executor as a new stable equilibrium
The Direct Principal as a new stable equilibrium
The Accountable Intermediary as a new stable equilibrium
Enforceable risk requires proportional agency
The infrastructure owner captures disproportionate value
Judgment holders command higher premiums as everything around them commoditizes
The effects bifurcate: top GPs expand in volume, median GPs compress in both margin and volume
The 2008 lesson — specification replacing discretion creates correlated fragility
Goodhart’s law — monitoring systems become optimization targets
Incumbent absorption — GPs may internalize the tools and preserve the bundle
The central uncertainty: how much of GP value is transferable trust aggregation vs. genuinely illegible judgment

Claim-by-Claim Research Mapping

1. Four roles are stable because agents control information

Jensen & Meckling (1976) — Agency costs (monitoring, bonding, residual loss) exist because principals cannot costlessly observe agent behavior. The four roles are the equilibrium response to these costs — each bundle minimizes the sum of agency costs given a specific information regime. Change the information regime and the cost-minimizing bundles change.

Holmstrom (1979) — The optimal contract depends on what is observable. When the agent controls observation, the contract is constrained to crude forms (flat fees, bundled trust premiums). The four roles are what remains after information constraints eliminate all other configurations.

Hayek (1945) — Knowledge exists in fragments across individuals; no central authority can aggregate it all. The four roles are a capital-markets response to Hayek’s knowledge problem: the agent holds local knowledge the principal cannot access, so the contract must accommodate that gap.

2. Completion-based pay biases toward action (the completion problem)

Rau (2000) — M&A advisory fees are overwhelmingly completion-contingent, and acquirers with stronger completion-incentive advisors show weaker long-term performance. Direct empirical evidence that completion pay creates a measurable bias toward closing deals regardless of quality.

McLaughlin (1990, 1992) — Contingent fees are efficient responses to deal uncertainty, not pure market failures. But the efficiency argument depends on reputation effects and repeat relationships moderating the completion bias. In one-shot or low-reputation settings, the bias dominates.

Golubov, Petmezas & Travlos (2012) — Top-tier advisors partially offset completion bias through reputation. The bias is real but moderated by advisor quality and repeat-game discipline. This is why the essay says the bias is “persistent” rather than “universal” — sophistication and reputation reduce it without eliminating it.

3. Time-based pay biases toward local optimization (the scope problem)

Holmstrom & Milgrom (1991) — The canonical source for the multitask problem. When agents perform multiple tasks but only some are measurable, strong incentives on measured tasks distort effort away from unmeasured ones. The PM closes tickets because tickets are measured. Tenant retention is not. This is the formal mechanism behind the scope problem.

Sirmans, Sirmans & Turnbull (1999) — Property management fee structures have predictable behavioral consequences: base fees encourage different behavior than leasing commissions. Time-based property management fees specifically incentivize throughput (ticket volume, showing activity) rather than outcome (NOI, tenant quality). Direct empirical grounding for the scope problem in the real estate context the essay focuses on.

4. Both problems persist because better contracts require information the principal doesn’t have

Holmstrom (1979) — The informativeness principle: the optimal contract should incorporate every informative signal. But if those signals are agent-controlled or non-existent, the optimal contract is unreachable. The principal is stuck with a suboptimal contract because the information needed to write a better one is held by the agent or was never captured.

Grossman & Stiglitz (1980) — Information production is costly. Even if the principal could theoretically acquire the information, the cost may exceed the benefit. The current contract structure is optimal given the cost of information — not optimal in the abstract.

5. Software created systems of record without transferring information control

Walsh & Ungson (1991) — Organizational memory is distributed and retrieval fails when it matters. Software addresses storage but not retrieval at the moment of decision. The PM enters data into the property management system but the data is not synthesized or surfaced when the HVAC decision is being made. The retrieval failure persists in the new medium.

Argote (1999, 2013) — Information systems store information but rarely make it retrievable at the moment of decision. Knowledge is embedded in people, tools, and tasks — software captures the tools layer but not the people or task layers. The agent’s information advantage persists because the system captures structured data while the relevant context remains in the agent’s head.

de Holan & Phillips (2004) — Organizational forgetting includes both unintentional loss and failure to capture. Software addresses loss (data is stored) but not capture failure (the context behind a pricing decision discussed in a hallway was never entered). The deeper problem — that the data needed for monitoring often does not exist in structured form — is a capture failure, not a storage failure.

6. AI transfers information control from agents to systems

Hayek (1945) — If the economic problem is coordinating dispersed knowledge, a system that can aggregate unstructured, dispersed information (emails, notes, records) without agent curation is a new coordination mechanism. AI is a partial solution to Hayek’s problem in the micro-context of the firm: it aggregates local knowledge that was previously accessible only to the agent.

Nonaka & Takeuchi (1995) — Competitive advantage comes from converting tacit knowledge into explicit knowledge. AI accelerates the externalization step of the SECI spiral — converting tacit context (embedded in emails, conversations, decisions) into explicit, system-accessible form. The agent’s tacit knowledge advantage narrows as the system captures more of what was previously illegible.

Polanyi (1966) — “We know more than we can tell.” Polanyi sets a hard limit: some knowledge is genuinely tacit and resists codification. AI can capture much of what was previously uncaptured (unstructured data, contextual reasoning) but cannot capture embodied skill, taste, or judgment that the agent themselves cannot articulate. This is the boundary of the information transfer — and the boundary of the essay’s predictions.

7. Discretion can be replaced by specification

Grossman & Hart (1986) — Contracts are incomplete because not all contingencies can be specified ex ante. But the degree of incompleteness is a function of the cost of specifying contingencies. If AI reduces that cost (by monitoring compliance with specified criteria at scale), contracts can become more complete — replacing discretion with specification in domains where specification was previously too expensive.

Williamson (1975, 1985) — The choice between market governance (specification, price mechanisms) and hierarchy (discretion, authority) depends on transaction costs. If AI reduces the transaction costs of specification and monitoring, the efficient boundary shifts toward market-like governance — more specification, less discretion, more external contracting.

8. The contract frontier expands when information becomes system-controlled

Holmstrom (1979) — The informativeness principle implies that when new informative signals become cheaply available, the set of viable contracts expands. System-controlled information provides signals that were previously too expensive to produce. Contracts that were theoretically optimal but practically unenforceable become viable.

Gorton & Pennacchi (1990) — Technological progress reduced information asymmetries in the loan market, making previously non-marketable bank assets marketable. New contract types (loan sales) emerged. This is a direct historical precedent for frontier expansion: better information enabled new contracts.

Coase (1937) — Firm boundaries shift when technology changes the relative cost of internal vs. market coordination. The contract frontier is the contract-level version of Coase’s firm boundary. As monitoring costs fall, the set of viable external contracts expands, and activities that were previously internal (bundled into the GP role) can be contracted externally.

9. Legibility → comparability → commoditization → margin compression

Glosten & Milgrom (1985) — Bid-ask spreads reflect information asymmetry. When information asymmetry decreases (as Bloomberg made dealer prices visible), spreads compress. This is the market microstructure version of the commoditization mechanism: legibility destroys the information advantage that sustained pricing power.

Kyle (1985) — The informed trader’s profit depends on their information advantage. As the market becomes more transparent (lambda decreases), the informed trader’s edge disappears. Applied to the GP: as system-controlled information makes the GP’s performance legible, the GP’s information-based pricing power compresses.

Grossman & Stiglitz (1980) — When prices become more informative, the return to private information acquisition falls. Applied to trust aggregation: as system-controlled information makes GP performance legible, the return to “knowing a good GP” (the LP’s core allocation decision) falls. The GP’s information rent compresses.

10. Historical precedents confirm the pattern

Glosten & Milgrom (1985) + Kyle (1985) — Together provide the theoretical framework for what happened in dealer markets (Bloomberg) and electronic trading: information transfer from intermediary-controlled to system-controlled compressed spreads and shifted value to infrastructure owners. The theory predicts the empirical pattern described in the essay.

Gorton & Pennacchi (1990) — The loan sales market emergence is a direct precedent: information technology made bank assets marketable, disaggregating the bank’s bundled role. The parallel to GP disaggregation is structural, not analogical.

11. The GP is a trust aggregator

Kaplan & Schoar (2005) — GP performance persists across funds, but capital flows don’t fully respond — top GPs don’t raise enough to compete away returns, poor performers keep raising. The LP cannot efficiently disaggregate GP skill from luck or from individual components. Trust aggregation is the mechanism that explains sticky capital: the LP pays for the bundle because they cannot price the parts.

Metrick & Yasuda (2010) — Two-thirds of GP revenue is fixed (management fees), insensitive to performance. The GP’s economics are driven by fund size, not by alpha generation. This is the financial anatomy of trust aggregation: the fee structure rewards commitment to the bundle rather than demonstrated per-component value.

Phalippou & Gottschalg (2009) — Average PE fund underperforms net of fees. Total fee burden exceeds 25% of invested capital. The industry’s opacity allows underperforming managers to continue raising. This is the cost of trust aggregation borne by LPs — the information disadvantage is large enough to sustain negative-alpha managers at scale.

12. AI makes per-component contribution observable

Holmstrom (1979) — The informativeness principle: when new signals about effort become available, the optimal contract changes. System-controlled information provides signals at the component level (sourcing quality, asset management execution, underwriting accuracy). The optimal contract shifts from bundled trust premium to component-level pricing.

Holmstrom & Milgrom (1991) — When measurement improves on specific tasks, incentives can be sharpened on those tasks without distorting effort on others — because the others are now also measurable. Component-level observability enables component-level contracting without the multitask distortion that previously made it counterproductive.

13. The GP role gets repriced and narrowed to the illegible residual

Polanyi (1966) — Tacit knowledge resists codification. The GP’s residual — sourcing relationships built on personal trust, judgment on novel situations, taste — is genuinely tacit in Polanyi’s sense. It cannot be transferred to a system because the GP themselves cannot fully articulate it. This is what makes the residual durable: it is not just hard to measure, it is fundamentally illegible.

Nonaka & Takeuchi (1995) — Some knowledge resists externalization. The GP’s judgment on when to deviate from the playbook is embedded in context and experience that the SECI spiral cannot fully convert to explicit form. The narrowing of the GP role converges on exactly the knowledge that resists the tacit-to-explicit conversion.

Teece (1980, 1982) — Economies of scope are conditional: bundling activities is only valuable when they genuinely share inputs or capabilities. If the observable components of the GP’s bundle can be provided more cheaply by specialists, the scope economics that justified the bundle no longer hold. The GP narrows because the bundling rationale disappears for the legible components.

14. Securitization 1.0 disaggregated the bank’s role

Gorton & Pennacchi (1990) — The loan sales market is the precursor to securitization. Banks retain larger fractions of riskier loans to maintain monitoring incentives (skin in the game). The disaggregation of origination from ownership requires that the originator retains risk — otherwise moral hazard emerges. This is the theoretical foundation for the 2008 lesson.

DeMarzo (2005) — Securitization strategy depends on information asymmetry: pooling with tranching creates information-insensitive senior tranches that are highly liquid. The senior tranche is the contract frontier expansion — a previously impossible contract (tradeable, rated exposure to a mortgage pool) becomes viable because the structure transfers information from originator-controlled to structure-controlled.

Keys, Mukherjee, Seru & Vig (2010) — Causal evidence that easier securitization led lenders to screen less carefully, producing 10-25% higher defaults at the FICO 620 threshold. The moral hazard is concentrated in low-documentation loans where soft information matters most. This is the empirical confirmation of the disaggregation risk: when origination separates from ownership, quality degrades on the dimensions that require judgment.

15. Commodity servicer vs. differentiated servicer

Gorton & Pennacchi (1990) — In performing loan markets, the servicer is standardized and replaceable because the work is routine collection and waterfall administration. The information needed to service a performing loan is fully captured in the structure. But non-performing loans require judgment — foreclosure timing, workout negotiation, asset disposition — that the structure cannot specify. The distinction between commodity and differentiated servicing maps directly to information transferability.

Bolton, Freixas & Shapiro (2012) — Credit rating quality matters most in complex, information-sensitive contexts. Rating agencies (the verification infrastructure for securitization) are more rigorous when the asset is complex because naive investors cannot substitute their own judgment. The ERLS NPL1 deal is an example: DBRS interrogated servicer quality deeply because the NPL portfolio’s value depended entirely on servicing judgment that the rating methodology alone could not verify.

16. Securitization 2.0: standardize decision-making

Williamson (1975, 1985) — The shift from discretion to specification is a shift from hierarchy-like governance to market-like governance. Securitization 2.0 is a Williamsonian regime change: activities that were previously governed by authority and discretion (the GP’s judgment calls) are re-governed by specification and monitoring (the operating protocol).

Grossman & Hart (1986) — Contracts become more complete when the cost of specifying contingencies falls. AI reduces this cost for operational real estate decisions. The “operating protocol” is a more complete contract than the traditional GP mandate — it specifies decision criteria that were previously left to residual control rights.

17. The operating protocol as the securitized asset

DeMarzo (2005) — What makes a security marketable is information insensitivity: the buyer does not need to know as much as the seller. If an operating protocol is specified, monitored, and verified by a system, the investor can evaluate the protocol rather than the operator. The asset becomes more information-insensitive — and therefore more liquid — because the operator’s judgment is no longer the primary risk factor.

18. Cognitive throughput is the binding constraint; removing it expands the market

Jevons (1865) — The Jevons Paradox: efficiency improvements in resource use can increase total consumption because lower per-unit cost makes the resource economical for new uses. Applied to GP cognitive throughput: if AI makes deal evaluation cheaper per unit, more deals get evaluated, more capital gets deployed, and the market expands. The binding constraint was attention, not capital.

Brooks (1975) — Communication overhead grows combinatorially with team size. The GP cannot simply hire more people to process more deals because the coordination cost of a larger team grows faster than the throughput gain. AI breaks this constraint because it scales monitoring without adding communication channels.

Dunbar (1992, 1993) + Hill & Dunbar (2003) — Relationship-based coordination has a hard cognitive ceiling (~150 stable relationships). The GP’s deal flow is partly constrained by how many relationships the team can maintain. AI can extend the information layer (synthesizing context from past interactions) but cannot extend the relationship layer (trust built through repeated personal interaction). This is another way of identifying the illegible residual: the relationship constraint is cognitive and personal in a way that AI cannot substitute.

19. Securitization creates moral hazard when origination separates from ownership

Keys, Mukherjee, Seru & Vig (2010) — The cleanest causal evidence: loans just above the FICO 620 securitization threshold default 10-25% more than those just below, because lenders screened less when they knew the loan would be sold. The moral hazard is worst where soft information (judgment) matters most and where hard metrics (credit scores) create a false sense of safety. Direct empirical support for the 2008 lesson objection.

Gorton & Pennacchi (1990) — Banks retain more of riskier loans to signal monitoring commitment. Skin in the game is the structural solution to securitization moral hazard. Securitization 2.0 must preserve meaningful risk exposure on the components that require judgment — the essay’s point about preserving risk on the judgment residual.

Bolton, Freixas & Shapiro (2012) — CRA ratings inflation is pro-cyclical: quality degrades during booms when naive investors are prevalent. The verification infrastructure itself is subject to incentive problems. Applied to Securitization 2.0: the system that monitors the operating protocol is not immune to gaming, bias, or failure — especially during market expansions when scrutiny relaxes.

20. Transition: monitoring → repricing → disaggregation

Gorton & Pennacchi (1990) — The loan sales market evolved in phases: first banks gained the ability to evaluate each other’s loans (monitoring), then loan pricing became standardized (repricing), then origination separated from ownership (disaggregation). The three phases in the essay mirror the historical evolution of securitization.

Argote (1999, 2013) — Organizational learning follows experience curves, but knowledge transfer across contexts is slow and lossy. The transition pathway is slow because each phase requires institutional learning — LPs learning to use monitoring tools, the market learning to benchmark GP components, lawyers learning to draft new fund structures.

21. The Accountable Executor as a new equilibrium

Holmstrom (1979) + Holmstrom & Milgrom (1991) — Together establish that outcome pay for bounded-task workers is only viable when attribution is granular enough to measure their contribution. System-controlled information provides this attribution. The Accountable Executor is the contract that becomes optimal under the informativeness principle once task-level measurement exists — it was always theoretically superior, just practically unenforceable.

Sirmans, Sirmans & Turnbull (1999) — Property management contracts already show variation in incentive structures (base fees, leasing commissions, renewal fees) with predictable behavioral consequences. The Accountable Executor is the next step: a PM paid on system-verified NOI contribution, with partial risk and continuous monitoring. The contract structure already exists in embryonic form.

22. The Direct Principal

Kaplan & Schoar (2005) — The largest, most sophisticated LPs (sovereign wealth funds, mega-endowments) already invest directly, bypassing GP intermediation where they have sufficient scale and capability. The Direct Principal equilibrium exists at the top of the market. AI extends it downmarket by providing the information infrastructure that previously required a large internal team.

23. The Accountable Intermediary

Rau (2000) + McLaughlin (1990, 1992) — Completion-based pay creates bias, moderated by reputation in repeat games. The Accountable Intermediary adds a structural enforcement mechanism (system-tracked post-close outcomes with holdbacks/clawbacks) that substitutes for reputation. It works in high-volume contexts where the system can build attribution data — not in bespoke, one-off transactions where the sample is too small.

24. Enforceable risk requires proportional agency

Grossman & Hart (1986) — The party with residual control rights must have corresponding investment incentives; otherwise the allocation is inefficient. Imposing risk on an agent without corresponding decision rights is a misallocation of control — the agent bears consequences for decisions they cannot influence, which is either refused or gamed.

Jensen & Meckling (1976) — Agency costs arise from the separation of ownership and control. Reuniting risk and control reduces agency costs. But forcing risk onto an agent without control is worse than separation — it creates punishment without incentive. The constraint is structural, not just practical.

25. The infrastructure owner captures disproportionate value

Glosten & Milgrom (1985) + Kyle (1985) — Market microstructure theory shows that the entity controlling price discovery and information flow captures economic rent. Bloomberg, exchanges, and electronic trading platforms extracted value by becoming the information infrastructure. The same mechanism predicts that whoever controls the monitoring/attribution infrastructure in PE captures disproportionate value.

Grossman & Stiglitz (1980) — When prices become more efficient through better information, the return to private information falls but the return to the information infrastructure rises. The platform that produces the information captures the Grossman-Stiglitz rent that was previously captured by informed agents (GPs).

DeMarzo (2005) — Intermediaries with securitization capability grow faster by leveraging capital through information-insensitive senior tranches. The infrastructure that enables securitization (standardization, verification, monitoring) captures value through throughput, not through risk-taking. The Fannie Mae analog: setting the standard and taking economics on every conforming transaction.

26. Judgment holders command higher premiums

Polanyi (1966) — Tacit knowledge is permanently resistant to codification. The individuals whose contribution is genuinely tacit — not just uncaptured but uncapturable — become scarcer as everything around them commoditizes. Their pricing power increases because substitution is structurally impossible.

Nonaka & Takeuchi (1995) — The SECI spiral converts tacit knowledge to explicit, but some knowledge resists externalization permanently. The judgment residual is defined by what the spiral cannot reach. As the system captures more, what remains unconverted becomes more valuable by exclusion.

27. The effects bifurcate: top GPs expand in volume, median GPs compress

Kaplan & Schoar (2005) — GP performance persistence is concentrated in the top quartile. This supports the bifurcation: top GPs have genuine judgment that survives the repricing — their commodity components compress but they capture premium on the judgment residual across a larger portfolio as cognitive throughput ceases to be the bottleneck. Median GPs, whose persistence is weak or absent, have less judgment to fall back on.

Jevons (1865) — The Jevons Paradox predicts that removing the cognitive throughput constraint expands total market activity. But the expansion surplus flows to whoever controls the scarce input. For top GPs, the scarce input is judgment — they benefit from volume expansion. For median GPs, nothing they provide is scarce once trust aggregation is contestable. The infrastructure owner captures the expansion surplus on commodity activity — the Bloomberg pattern.

Phalippou & Gottschalg (2009) — The median PE fund already underperforms net of fees, sustained by information opacity. When system-controlled information makes performance transparent and per-component attribution feasible, the median GP’s trust-aggregation rent becomes visible and contestable. The compression is largest for managers whose alpha was primarily rent, not judgment.

Metrick & Yasuda (2010) — Two-thirds of GP revenue is fixed fees insensitive to performance. If observable components get repriced to commodity rates and only the judgment residual commands premium pricing, the fixed-fee component collapses for managers who cannot demonstrate judgment-driven alpha. Top GPs with demonstrable skill retain pricing power on the judgment component; the median GP’s fixed-fee economics are the most exposed.

28. The 2008 lesson — specification replacing discretion creates fragility

Keys, Mukherjee, Seru & Vig (2010) — Causal evidence that rule-based securitization criteria (FICO cutoffs) were gamed and produced worse outcomes than discretionary screening. When specification replaces judgment, the specification becomes the attack surface. Errors in the specification are systematic and correlated — every loan above 620 was treated identically regardless of soft information.

Bolton, Freixas & Shapiro (2012) — Verification infrastructure (CRAs) failed pro-cyclically. The monitoring system itself was subject to the same incentive problems it was supposed to solve. A biased monitoring system in Securitization 2.0 would make the same error across every asset simultaneously — correlated failure is the structural risk.

29. Goodhart’s law — monitoring becomes optimization target

Holmstrom & Milgrom (1991) — Strong incentives on measured dimensions distort effort away from unmeasured ones. When the monitoring system defines what is measured, agents optimize for the system’s metrics. The realistic frame is an arms race between measurement and gaming — the thesis holds only if measurement iterates faster.

Cyert & March (1963) — Organizations satisfice against aspiration levels and respond to performance feedback with problemistic search. Agents optimizing for monitoring metrics is a rational satisficing response: meet the measured threshold, then allocate remaining effort to personal utility. The Goodhart dynamic is a predictable consequence of any monitoring regime.

Bolton, Freixas & Shapiro (2012) — CRA conflicts stem from the issuer-pays model: the entity being evaluated pays the evaluator, creating structural misalignment. Ratings inflation is pro-cyclical and persistent, not self-correcting. Applied to the monitoring infrastructure: if GPs are the paying customers, the platform faces the same issuer-pays dynamic — optimizing for what GPs want to see rather than what LPs need to know. Who the infrastructure serves (GPs vs. LPs) determines whether the Goodhart problem is mitigated or reproduced at the platform level. This is the deepest version of the objection and connects directly to the 2008 lesson.

30. Incumbent absorption — GPs internalize the tools

Teece (1980, 1982) — Economies of scope are conditional: if the GP can integrate the information tooling into their existing bundle, the scope economics may favor internalization over disaggregation. Large GPs with the resources to build or buy monitoring infrastructure may preserve the bundled structure by absorbing the technology, capturing the efficiency gain internally rather than ceding it to an external platform.

Coase (1937) — Whether the information infrastructure sits inside the GP or outside depends on relative coordination costs. If the GP can operate the system internally at lower cost than contracting with an external platform, the firm boundary expands to include the infrastructure. The structural change then occurs within firms (the GP becomes more efficient) rather than between them (the GP gets disaggregated).

31. How much is trust aggregation vs. illegible judgment — the central uncertainty

Kaplan & Schoar (2005) — GP performance persistence is real, which means some component of the GP’s value is genuine skill. But the persistence data cannot distinguish which component of the bundle drives it. The 80/20 vs. 40/60 question is empirically open — Kaplan & Schoar show skill exists but cannot decompose it.

Polanyi (1966) — The boundary between codifiable and tacit knowledge is not sharp. Some knowledge that appears tacit may become explicit with better tools; some that appears codifiable may resist capture in practice. The boundary moves as information technology improves, which is why the answer to the central question will only become visible as system-controlled information is actually deployed.

Phalippou & Gottschalg (2009) — If the median GP underperforms net of fees, a significant portion of current GP economics is rent, not judgment-driven alpha. This suggests the transferable fraction is larger than the industry claims — but the top-quartile persistence documented by Kaplan & Schoar suggests the judgment fraction is real for the best managers. The distribution is bimodal: mostly rent for the median, mostly judgment for the top.