Skip to content
Marcus Gawronsky
Marcus Gawronsky

Project case study

Pricing Perspective

A quantitative-finance research programme connecting language-model representations, optimal transport and portfolio risk, with software for empirical tests, simulation and formal verification.

Evidence

Pricing Perspective pipeline stage graph88 pipeline stages grouped into families. Corpus and market data feed representations, geometry and validation, which feed three papers; the papers and the Lean claim checks feed the manuscript render. Monte Carlo studies declare no inputs from the rest of the pipeline. One provenance stage hashes the source code every stage depends on.Code provenance · 1 stage · an input to every stage below
Pipeline stage graph

Evidence

Question
Can every number, figure and claim in the manuscripts be traced to the data, code and environment that produced it?
Method
The 88 DVC stages in dvc.yaml, grouped into families. An edge means a stage in one family declares an output of another as an input. A code-provenance stage hashes the source that every stage depends on.
Observation
The three papers share one corpus, one set of representations and one geometry layer, and cite one another's results. A change upstream re-runs every paper that depends on it.
Limitation
A snapshot of dvc.yaml at commit eb3c1b5 (6 September 2026). Families collapse many stages, and the Monte Carlo studies declare no inputs from the rest of the pipeline, so they appear unconnected.
Source
dvc.yaml at eb3c1b5

Role: Research, modelling, software architecture, and reproducibility

Pricing Perspective co-locates research manuscripts, data pipelines, simulations, formal Lean proofs, and a static publication website so that claims and evidence evolve atomically.

In the pipeline, the Lean claim checks are inputs to the manuscript render stage.

The research software connects financial data, mathematical models and empirical tests in a reproducible environment.

The papers are connected in the pipeline as well as in argument: stages of each paper consume results produced by the others.

Technology and method

PythonLean 4LaTeXNext.jsNixDVC

Methods: Optimal transport and Wasserstein geometry · Covariance and factor structure · Language-model representations · Reproducible research systems

WebsiteGitHub