Independent research

The constraint was never capability. It was legibility.

I work on the structures institutions build around information they cannot read — in market risk, transaction banking and private credit — and on what becomes unnecessary once that information is computable. Each piece here is offered for discussion, with the underlying working papers linked where they exist.

Research tracks

Market risk

Regulatory risk architecture — stress testing, trading-book capital and limit frameworks — and what a single continuously computed risk state would replace.

1 piece(s)

Transaction banking & treasury

Payment flows as information — provenance, purpose and forward implication — and why fungibility is a compensation rather than a fact.

1 piece(s)

Private credit & tokenization

Continuous risk pricing of confidential-borrower assets through attested factor reporting; observability separated from the reporting calendar.

1 piece(s)

Constrained AI & governance

Separating interpretation from calculation, provenance as a precondition, and uncertainty that survives to the output.

1 piece(s)

Food systems & supply chain resilience

Localisation, future foods and supply-chain resilience.

0 piece(s)

Climate science

Climate science research track.

0 piece(s)

Real estate economics

Real estate economics research track.

0 piece(s)

Papers & points of view

point of-view · Draft for discussion · September 2026

The compensating structure

Why most institutional AI work is tooling, and what the alternative actually looks like.

Most institutional AI work is tooling: it makes an existing process cheaper or faster and leaves the shape of the business unchanged. The alternative starts from one question — what does this let us stop doing? — and the answer is usually a compensating structure: a body of process, infrastructure and convention that exists only because a particular piece of information could not be read. Three cases from unrelated businesses share the identical shape.

Market riskTransaction banking & treasuryPrivate credit & tokenizationConstrained AI & governance

How it connects

Three businesses with nothing in common — different clients, regulators, infrastructure and people — and the same structure in each: information the institution already held, a workaround for the inability to read it, and a workaround that is now the largest single line of cost in that part of the business.

If the structure is general, the capability that dissolves it is transferable: the same engineering, the same governance argument, the same sequencing problem, applied to a different corpus. That is what ties the tracks above together, and it is a very different investment case from three unrelated projects.

This section will grow into an interactive map of the tracks. For now it is prose.