Peter Kahl

Research Frameworks

A connected research programme on evaluative governance: how institutions constitute standards, delegate judgement, recognise capability, govern admission and conversion between recognition systems, and remain answerable for consequential decisions across AI, law, higher education, finance and public governance.

Research programme

The governance of evaluative power.

These frameworks are not separate intellectual projects. Together, they examine a recurring organisational problem: formal authority may remain assigned while effective control over judgement, evidence, standards, admission, conversion and evaluation has migrated elsewhere.

The programme therefore asks where a consequential result was actually determined, who constituted the standard that governed it, what may count under that standard, which recognised achievements may travel between institutions, whether the standard can be challenged or revised, and who remains capable of answering for the decision afterwards.

Framework 01

Evaluative Control

Evaluative Control identifies the power to determine the standards, categories, evidence and comparison rules through which persons, claims, risks, capabilities and possible outcomes are assessed.

It distinguishes formal authority over a decision from control over the evaluation that produces it. An organisation may remain legally or institutionally responsible for an outcome while depending upon evaluative machinery that it did not constitute and cannot adequately interrogate.

This is the programme’s principal diagnostic concept. It applies to boards, regulators, universities, legal services organisations and AI-enabled decision systems.

Framework 02

Delegated Discretion

Delegated Discretion examines how authority is conferred, bounded, exercised and supervised when judgement is entrusted to another person, institution, professional process or artificial system.

The framework asks whether the recipient merely applies an inherited standard or possesses discretion to constitute, modify or displace that standard. That distinction determines where responsibility should attach and whether the delegating institution retains genuine evaluative control.

Framework 03

Constitutive Trace and the Attribution Rule

The Constitutive Trace follows a consequential decision backwards, not merely to the actor who executed it, but to the point at which its operative standard was constituted.

Its attribution rule is that responsibility for a standard ordinarily sits where that standard was settled, not simply where it was later transmitted or applied. The framework distinguishes answerability for conformance from answerability for the standard itself.

It also identifies the possibility of a null terminus: a standard may govern consequential decisions even though no present actor or office is capable of answering for why it has the form it does.

Framework 04

Answerability Capture

Answerability Capture occurs when the procedures intended to make authority accountable are controlled by the same institutional or evaluative structure whose decisions are being questioned.

Reasons may be supplied, audits conducted and responsibilities formally allocated, while the underlying standard remains insulated from external challenge. The appearance of accountability is preserved, but the route by which the governing evaluation could lose is closed.

The associated principle is that distribution is not answerability. Dividing tasks or assigning nominal responsibility across a system does not establish that any actor can reconstruct and justify the judgement that determined the outcome.

Framework 05

The Answerability Fuse

The Answerability Fuse is a legal and governance trigger for the point at which a consequential decision remains operational but the judgement responsible for it can no longer be adequately reconstructed.

It shifts attention away from catastrophic failure alone and towards an earlier institutional threshold: the loss of a reliable account of who authorised the decision, what standard governed it, what evidence was material, and where responsibility remains located.

The proposed response is procedural rather than merely explanatory. Once the threshold is crossed, enhanced duties of preservation, reconstruction, disclosure, review and human responsibility should arise.

Framework 06

Objective-Layer AI

Objective-Layer AI distinguishes capability within an assigned objective from participation in the evaluation, revision or constitution of the objective itself.

Governance cannot therefore be confined to model outputs. It must also address the evaluative layer through which a system determines what counts as success, chooses between competing outputs, modifies its strategy and, in more advanced systems, alters the practical meaning of its governing objective.

The framework is particularly relevant to frontier and agentic AI, where output review alone may fail to reveal how the system’s effective evaluative criteria are changing.

Framework 07

Continuity–Rupture–Generativity

Continuity–Rupture–Generativity distinguishes three dimensions of intelligent action: continuity across time, rupture from an inherited trajectory, and the generation of a materially new course of action.

The framework separates agency from autonomy. A system may pursue objectives and generate novel outputs without possessing the continuity required to maintain, recognise and revise a course of action as its own over time.

It provides a more discriminating basis for analysing artificial agency, autonomy, deliberation and responsibility than capability measures based principally on output sophistication.

Framework 08

Representational Sealing Theory

Representational Sealing Theory examines the condition in which a generated representation becomes inaccessible to inquiry into the generative conditions that produced it.

A represented order may appear complete and intelligible while excluding the temporal, causal or procedural history through which it was composed. The resulting seal is not merely informational opacity; it is a structural limitation on what can be recovered from within the representation itself.

The framework applies to generated environments, institutional records, automated explanations and AI systems whose outputs conceal or compress the process by which their apparent world was formed.

Framework 09

The Recognition Game and the Game of Games

The Recognition Game analyses how institutions allocate credibility, status and opportunity through metrics, credentials, rankings, supervisory dependence and incumbent standards of legibility.

Its claim is not merely that individual metrics are imperfect. Multiple recognition systems interact: education, publication, employment, funding, professional status and institutional prestige convert value between one another. Together they form a game of games in which the standards of admission and conversion can become more consequential than the underlying capability being assessed.

Epistemic clientelism describes one resulting pathology: access to recognition becomes dependent upon proximity to incumbent evaluators, institutions or authorised forms of thought.

Framework 10

Admissibility, Conversion and the Evaluative Exchange Rate

Admissibility and Conversion identify the two gates through which institutional recognition systems are joined into a wider evaluative order. The admissibility gate determines what may enter a recognition process as a contribution of the relevant kind. The conversion gate determines which recognised achievements may travel into another institutional field, and at what rate.

The framework relocates analysis from competition within a single institution to the machinery that connects institutions. Its central claim is that consequential evaluative power may sit neither with the institution in which recognition was earned nor with the institution in which it is later used, but in the rules governing admission and conversion between them.

Evaluative crowding occurs when the available category-space becomes congested or too narrow to admit a capability for which no authorised form exists. Evaluative capture arises at a second order when those who control a gate also control what may count as evidence that its governing rule has failed.

The constitutional question is therefore by what authority these gates operate, to whom their holders answer, who has standing to contest them, and which institution may reopen their rules. An answerable evaluative order requires provenance, attribution, standing, revision and remedy.

Framework 11

Financial Machine Learning, Evaluative Crowding and Strategy Saturation

This framework examines how apparently independent financial machine-learning systems can become structurally similar because they are trained on overlapping data, organised through comparable representations, optimised against related objectives and evaluated through common performance criteria.

The resulting problem is not merely model correlation. It is evaluative crowding: a narrowing of the effective strategy-space as diverse systems are channelled through the same admissible features, labels, benchmarks, risk measures and optimisation conventions. Formal plurality may therefore conceal substantive convergence.

Strategy saturation describes the market-level consequence. As increasingly similar systems identify and exploit the same apparent opportunities, returns are compressed, positions become crowded and individually rational optimisation may produce collective instability. Diversification at the level of firms or models may fail where the evaluative architecture is shared.

The framework also addresses agentic asset allocation, where artificial systems do not merely execute a fixed strategy but select, revise and allocate capital between strategies. Governance must then examine who constituted the objectives, constraints and comparison rules through which the system determines which strategies deserve capital.

Framework 12

Evaluative Sovereignty and Innovation Governance

Evaluative Sovereignty concerns the capacity of an organisation or governing order to control and revise the standards by which new capability is recognised and recognised value is converted between institutional settings.

Innovation governance fails where institutions can apply inherited standards but cannot lawfully reopen them when genuinely new methods, instruments or forms of competence appear. It also fails where standards are reopened informally, without a traceable act of authority or an actor capable of answering for the revision.

The central design requirement is lawful standard-reopening: novelty must be capable of challenging an incumbent evaluative rule through a route that preserves provenance, authority and responsibility. The same requirement applies where an institution controls whether recognised capability may cross into another field and the rate at which it counts.

Framework 13

Standards That Can Lose and Epistemocracy

Standards That Can Lose proposes that an authoritative evaluative standard should remain capable of defeat by evidence arising from the people, claims or capabilities it governs.

The associated model of epistemocracy does not mean government by those already certified as knowing. It means governance in which epistemic authority remains institutionally answerable and no recognised evaluator possesses finality over the standard it applies.

The framework combines fiduciary answerability, epistemic plurality, distributed credibility, independent standing and structured routes through which an incumbent recognition standard may be challenged and defeated.

Framework 14

Epistemic Fiduciary Governance

Epistemic Fiduciary Governance treats the exercise of authoritative judgement as a form of entrusted power where another person’s interests, opportunities or recognised standing depend upon the quality and integrity of that judgement.

Epistemic humility is therefore not merely a personal virtue. It becomes an institutional discipline: a requirement to recognise uncertainty, disclose the limits of one’s knowledge, preserve routes of challenge, and avoid converting possession of evaluative authority into presumed epistemic superiority.

The framework connects fiduciary theory, epistemic justice, university governance, professional responsibility and AI-mediated decision-making.

Synthesis

Governance in which authority remains capable of giving an account of its judgement.

The common concern is not transparency in the abstract. It is whether consequential judgement remains attributable, reconstructable, contestable and revisable when it is mediated by professional categories, institutional procedures, admission rules, conversion rules, metrics, infrastructures, models or autonomous technical systems.

The programme proceeds from a single practical proposition: an organisation does not retain meaningful control merely because responsibility remains formally assigned. It retains control only where it can identify the operative standard, determine who constituted it, reconstruct how it governed the result, and preserve a lawful route through which that standard may be challenged or changed.

The same proposition applies to the wider order formed between institutions. A recognition system is not answerable merely because each participating institution can explain its own procedure. The admissibility and conversion rules connecting those institutions must themselves possess provenance, an attributable holder, contestable standing, a route of revision and an effective remedy.

View publications View curriculum vitae