Working together
Work with me
Verification-first analysis and tooling for finance and life sciences, delivered with the method published. In regulated work the question is rarely whether artificial intelligence was used. It is whether the result can be checked. Everything on this site was built to answer that, and the same standard applies to work done for others.
What that looks like
Analysis and models in which every figure carries its confidence — what is drawn directly from primary sources, what is modelled on top of them, and how far each can be trusted. Interactive tools built with the same discipline as the instruments here: an independent reference before anything ships, a self-test on the page, and a reviewer instructed to break it. Where an engagement allows, the method is published, so that a result can be held against it later rather than taken on trust.
Who it is for
People who have to decide something on the strength of a number and will be held to the decision — in funds, on boards, in policy teams, in the clinical and commercial sides of drug development. Places where a wrong number is expensive and an unverifiable one is worse.
The evidence
Equity Research →
The method above carried out in public, on a listed security: the
Varseta-M briefing prices one asset to a risk-adjusted net present value,
every assumption exposed as a control you can move, an independent Python
twin held to agree with the engine to one part in a million. Research,
not advice, and the position behind it is disclosed on the page.
The Margin Squeeze →
A conditional diagnostic of Luxembourg's growth model to 2040 — a working
paper and an interactive model, every figure confidence-tagged, sources
cited and not implied.
Pipeline Atlas →
A map of clinical development built from a 25-year public registry on one
rule that outranks every feature: never present an absence as a finding.
The instruments →
The standard applied to eight instruments, seven of them
known mathematics, each checked against an independent reference and most
carrying a self-test you can run — and to each of the pieces above.
Positions
I publish research on listed securities, and when I do I disclose what I own on the page: the Varseta-M briefing says I am long CTMX, the VK2735 briefing that I hold no position in VKTX, the Monte Rosa briefing that I hold no position in GLUE, the Annexon briefing that I hold no position in ANNX, and the Immunic briefing that I am long IMUX, a position I entered after writing it and disclosed there on 9 October 2026. So the question has a standing answer rather than an awkward one.
How to start
Write with what you are trying to decide and what a wrong answer would cost. That is enough to tell us both whether the method fits the problem.