In the crypto industry, every serious project begins with a whitepaper. In keeping with the tradition... This is the whiteboard.
The whiteboard was drawn at Wyden, a crypto infrastructure provider in Zurich, in marker ink, by Patrick Mehrhoff. It contains the complete intellectual logic of the IADT, the UCT, and the ISP, which was originally already applied since 2018 working at Crypto Finance. Three formal theories that did not yet have names. The first public large language model arrived four years later. The published work is a one-to-one mapping of what was on this board. The names came later. The logic was already complete.
Six columns. Marker ink. 2019. Untouched since. The first column lists market trends. The second pairs each trend with its cause. The third names the target groups the trends will move. The fourth lists the problems. The fifth, in red, lists the solutions. The sixth, in green, scores eleven jurisdictions side by side. A note in green ink at the top of the country column reads: Target Groups and Solutions will differ in countries depending on points. The marker colours are incidental. The logic is not.
Three theories sit on this board, none of them yet named. The arrows between columns are the UCT cascade. The country comparison is the ISP cross-scale identity. The columns themselves are the four-force IADT. The published work has spent years articulating what was already drawn here.
WHITEBOARD TO THEORY
The mapping is one-to-one. Seven structural elements identified on the 2019 whiteboard, seven formal-theory equivalents in the published work. The first five rows correspond to the four-force IADT architecture and its action output. The sixth row is the UCT cascade. The seventh is the ISP cross-scale identity.
The whiteboard date is a fact. The whiteboard photograph carries a 2019 timestamp in marker ink. The first publicly available large language model, ChatGPT, arrived four years later, in November 2022. The intervening period sat entirely inside one person's head, sketched once on a wall in Zurich, with no chatbot to summon, no model to query, no AI tool to consult. The date is provenance. It is the work's first verifiable fact. Every claim on this page rests on it.
Three facts, demonstrated simultaneously by the timestamp.
2019, four years before ChatGPT (November 2022) and before any publicly available large language model. The chronology is incompatible with an AI-generation account of the work.
All three theories (IADT, UCT, ISP) are visible in embryonic form on a single whiteboard. This is not a partial sketch later expanded. It is the complete conceptual architecture, present at the moment of inception.
The whiteboard was drawn to solve a business problem: why identical institutional clients in different countries make divergent decisions about the same product. The theories emerged from observation, not from academic literature.
AI language models generate outputs consistent with their training distribution.
The IADT claims the training distribution, additive behavioural models, is structurally wrong.
The 2019 whiteboard predates the public availability of any AI language model by four years.
Patrick Mehrhoff joined Crypto Finance (a Deutsche Börse company) in 2018, the Zurich firm founded in 2017 as the first FINMA-regulated provider of trading, custody, and investment services for digital assets. His responsibility was to understand why institutional clients in different jurisdictions, with similar product needs and similar regulatory environments, made dramatically different adoption decisions about the same set of services. The logic that became the IADT, the UCT, and the ISP was first applied here, in practice, on real marketing and sales processes.
The question was operational, not academic. Why does an identical product, presented to identical client profiles, succeed in Switzerland and fail in Singapore? Additive models, the standard analytical approach at the time, could not explain the divergence. A country with strong scores on most factors but zero on a single factor still failed to adopt. A country with moderate scores across the board sometimes succeeded. The pattern was not linear, and it was not random.
The 2019 whiteboard, drawn at crypto infrastructure provider Wyden in Zurich the following year, was where the pattern became visible to itself. Drawing the five columns left to right, with the cascade arrows and the country comparison stretched across the right edge, made the structural logic of the problem inescapable. The same shape, drawn at every magnification, described every case the data covered.
The formalisation of the three theories took several more years. The empirical demonstrations across nine domains took longer still. But the work that the studio publishes today is, in its essentials, an articulation of what the 2019 whiteboard had already captured. The names came later. The logic was already complete.
Decision outcomes are governed by structural forces, not by stable attributes. The same multiplicative logic recurs at three scales and across domains.
Action under complexity is determined by the dynamic interaction of four forces. The interaction itself is the mechanism. Additive models systematically under-perform because they cannot represent it.
How individual decisions aggregate into institutional outcomes through three sequential phases. Ignition. Polarisation. Institutionalisation.
A criterion for when a structural pattern recurs across scales and domains. The pattern transfers only where structural identity holds.
Patrick Mehrhoff is a German entrepreneur, independent researcher, AI engineer, and author. He is the author of the three original theories that constitute the Mehrhoff Research Program. He built the work the studio carries.
Most practitioners spend a career inside one domain. He worked as a practitioner at six of the frontiers that redefined how society organises information, money, and value over the last two decades. The consumer internet, Swiss fintech, European monetary policy, Southeast Asian financial publishing, institutional crypto assets, and the first wave of applied AI.
The same structural observation appeared at every frontier. By 2019, on a whiteboard at crypto infrastructure provider Wyden in Zurich, the pattern had become visible to itself. It did not yet have a name. Four years before any AI language model became publicly available, the complete logic of the Three Laws of Decision-Making was already on the whiteboard.
The studio was built from Bangkok, without institutional affiliation and without external funding. The Rubens workshop model defines how it operates. The founder sets the intellectual direction at every decisive step. The tools and collaborators execute within it.
Tested against nine foundational scholars, three of them Nobel laureates, and across fifteen structural dimensions. Every cell is independently challengeable.
Every dimension independently challengeable.
Most frequently cited as predecessors.
Among the nine scholars tested.
Confirmed across all nine scholars.
The three laws describe the same multiplicative structure that governs decisions across scales. What each audience recognises in them is different. Each doorway below opens into the same set of laws, framed for the way you are most likely to find your own work inside them.
The literature in your field has been reporting tail-failure for decades. Outliers, noise, model misspecification. The three laws name a different culprit. The structural form of the model itself. Read the laws and find the place where the assumption you never questioned has been doing the work all along.
Every senior decision-maker has watched a model work on average and break where the decision mattered. The three laws name what is happening. The breakage is not an edge case the model failed to anticipate. It is the multiplicative architecture finding its zero. Read the laws and the pattern beneath every model break of your career becomes legible at once.
Each major institutional cascade of the past fifteen years was investigated as if it were unique. The reports document different operational failures in different sectors. The three laws read across the reports and recognise one structural pattern. Read the laws and the recurrence stops looking like coincidence.
Every philosophical tradition that has tried to describe the world has eventually arrived at the same conclusion. The three laws turn the conclusion into a working structural explanation. Decisions form. They cascade. Their patterns transfer across scales. Read the laws the way our studio reads them.
Most foundational research is finished before a student arrives. The three laws sit early enough in their life that the field around them is still to being created. A graduate student who reads them now enters a body of work inside its formative window. The reading path is curated for entry from any quantitative or social-science background.
Universities train within additive frameworks. Journals review within them. Regulators write rules under them. Funders evaluate against them. The three laws name the structural assumption beneath the entire foundational layer. It has been invisible because no curriculum, no review, no framework has ever required it to be examined. Read the corpus and the link your institution has been operating without becomes legible.
The first program states the architecture and demonstrates the laws that govern it. The second tests those laws against the empirical record of human decision-making. Together they establish the studio's canonical core and its first empirical substantiation. Further programs are in active preparation.
The three laws at the micro, macro, and meta scales, derived and stated formally. Plus the empirical evidence that the multiplicative architecture holds across biological, artificial, and hybrid substrates. Every paper in every other program traces its structural claims back to this canonical core.
The laws tested against the empirical record from every sector, industry, and discipline where humans make decisions. From financial risk to clinical judgement to educational selection to electoral behaviour, the same multiplicative architecture governs the outcome. Every test in the program substantiates the structural finding the foundational program advanced.
Mehrhoff Research operates outside academia and outside industry. An independent science and research studio with no external funding, no board, and no university partnership. Founded 2019 in Zurich, domiciled in Tallinn, run from Bangkok.
The work is published open access. Every paper passes adversarial peer review before deposit. Every claim names the empirical conditions under which it would be falsified. Every citation traces to a primary source. The standard is not the institution. The standard is the work.
Deposited on Zenodo, permanently archived.
Data points analysed across all empirical validations.
Foundational and human decision-making.
No paywall, no journal embargo.
A horizon decades away. The daily practice of moving toward it. Seven principles the studio holds itself to, with or without anyone looking. Five codes the studio answers to when no external authority requires them. Each statement enforced internally. Each verifiable from outside.
Vision and mission, stated directly. What the studio is working toward, and what it does today.
Seven standards held against every paper, every conversation, every relationship the studio enters. Decision-making, rigour, independence, transparency, originality, curiosity, and courage.
Self-regulated science. Published standards, open methods, full results regardless of what they show.
Press, academic peers, conference organisers, policy bodies, industry practitioners, and readers reach the studio here.
Six formats, from public keynote to closed doctoral seminar. The full range is welcome; the substance of the questions decides the invitation.
Four positions, from visiting researcher to fellowship, with two scheduled per calendar year. A residency produces sharper work than a single talk because the conversation extends.
Press kit with bios, headshots, and topic briefs is ready to download. Lectures, learning materials, and the architecture itself are licensable.