ValemontInvest
A hand-drawn line chart sketched in ink on a page of a notebook, resting on a wooden desk beside a clear plastic ruler and a pen

The Valemont Invest origin story · Twelve years

From a single problem to a comprehensive system

Two mathematicians left Cambridge with one principle and no answer. Twelve years later they had a working system — and were still testing the answer.

Question taken up
2015
Company founded
Sep 2020
Renamed
Sep 2026

Some companies begin with a business plan. Valemont Invest Inc began with a rule about the order in which questions and answers should be trusted.

A principle, not an answer

Evan Valemont studied mathematics, probability and the structure of financial markets. His training left him with a specific habit of mind: when a judgment is offered, ask what has to be true underneath it, and ask under what circumstances it would fail. Ryan Mercer arrived from applied mathematics and computational finance and then moved towards data and engineering. His preoccupation was narrower but just as demanding — whether a theory that looks powerful on a page can be written into something workable, and then tested repeatedly without falling apart.

At Cambridge the two shared a tutor, and through that tutor a long-running argument about how a judgment should be earned. What they carried away from those discussions was not a technique. It was a principle, stated plainly and meant to be lived with: understand the problem first, then trust the answer.

The order is the whole point. The principle is not a claim that answers are worthless. It is a rule that an answer without an understood problem cannot be checked, and an uncheckable answer cannot be relied on — however elegant it looks, and however confidently it is stated. This is an uncomfortable rule to build a career on, because it means the first duty is always to characterise the problem more precisely, and that duty never really ends.

Students seated around a table in a library, leaning in towards a laptop to work through a problem together
A shared tutor, a shared set of questions. The discussions that began at Cambridge were the beginning of a collaboration that outlasted the degree.

Their paths then diverged, as paths do. But wherever each of them went, the same challenge kept surfacing, and it was a challenge about access rather than about intelligence.

The gap they kept returning to

Mature institutions can support a decision with four things working together: data, research, models and risk management. Each is ordinary on its own. Together they form a loop in which a view can be formed, challenged, constrained and reviewed before any money is committed to it.

Data is the raw record: prices, volumes, macroeconomic releases, market structure, flows. Research is the process that turns that record into a question worth asking. Models are the formal statements of a relationship that can be tested against history and against out-of-sample data. Risk management is the constraint layer — the part that decides how much of a view is survivable if the view is wrong.

An individual facing the same market rarely has any of that machinery. What they have instead is news, opinion and immediate reaction: a continuous stream of conclusions with the reasoning stripped out. The conclusions arrive fast and confidently, and there is almost no way to ask the one question the Cambridge principle demands — under what conditions would this be wrong?

Mature institutions can support a decision with data, research, models and risk management. Many individuals, facing the same market, can only find direction among news, opinions and immediate reactions.

Evan Valemont and Ryan Mercer wanted to know something specific, and it was not whether individuals could be given better tips. They wanted to know whether institutional research methods could be taken apart and reassembled into something clearer and more verifiable — a capability in which the reasoning survived the trip, rather than being compressed into a verdict.

That question is harder than it sounds. Institutional research is not a single object that can be copied; it is a set of habits embedded in teams, in process and in infrastructure. Deconstructing it and rebuilding it for a different setting means rebuilding the habits too, and the habits are the expensive part.

That is why the question they settled on was never really about tools. A tool can be borrowed; a habit of verification has to be built, and then maintained under pressure. For anyone who has tried to make sense of markets without an institution behind them, this is the whole difference between a system that can be examined and a stream of confident assertions that cannot be checked at all.

2015: deciding to answer it seriously

In 2015 they decided to answer the question properly. It is worth being precise about what that decision was, because it was not a product launch and it produced nothing immediately. It was a commitment to work on the problem until the method itself could be trusted — a decision about how they would spend their time rather than about what they would sell.

From there the two took up opposite ends of the same problem. Evan Valemont analysed market structure, risk boundaries and the logic of judgment: what a market is actually made of, where the limits of a position sit, and how a conclusion should be assembled from evidence. Ryan Mercer worked to embed that logic into data architecture and models: how the record is stored, how it is joined, how a relationship is expressed in a form a machine can evaluate.

The failures arrived quickly, and they were instructive in a way that successes are not. An assumption might hold up on paper and lose its persuasiveness the moment it met historical data. A model might perform well inside a sample and then expose a weakness in a different market environment. Neither outcome was treated as a verdict on the idea; both were treated as information about what had not yet been understood.

What they learned was not how to make every test successful. It was how to find the parts they had not yet understood when a test failed.

That distinction is easy to say and difficult to hold to over years. The natural reaction to a failed test is to adjust the model until the test passes. The discipline they were building pointed the other way: to leave the failure in view, and to go looking for the part of the problem that had been misdescribed.

Two disciplines, deliberately held in tension

When people describe a quantitative research effort, they tend to describe one skill. In practice two are required, and they pull in opposite directions.

Evan Valemont supplies the research framework and the risk constraints. He defines what counts as an acceptable question, which relationships are worth testing, and what a position is allowed to look like once it exists. This work is essentially financial: it decides whether a result means anything in market terms, and it refuses results that cannot be given an economic interpretation.

Ryan Mercer supplies the data architecture and the engineering system: how information enters, how models are deployed, how results are produced reproducibly. This work is essentially structural. It decides whether an idea can be executed reliably, at a scale and a cadence that a research process can actually sustain.

Two colleagues sitting at a desk with open laptops, handwritten working notes and a pen between them
Research framework on one side, data architecture on the other. The two requirements are kept in contact so that neither is allowed to declare victory alone.

These two capabilities do not align easily. Some market phenomena resist being coded neatly — the relationship is real, but it will not sit still inside a clean formulation. In the other direction, some technically impressive results cannot withstand scrutiny from financial logic: they are accurate on the data and meaningless in the market.

It is precisely this mutual requirement that does the work. Because each discipline is obliged to answer to the other, neither can quietly mistake works for proven. A result that satisfies the engineer but not the risk framework, or the risk framework but not the data, is not allowed to graduate.

Read plainly, this is a description of a working relationship rather than a technical claim. Evan Valemont and Ryan Mercer did not set out to make two skills compatible; they set out to make each one answerable to the other. The friction is the point, and the friction is what makes the resulting method defensible to someone outside the room.

The chronology

Twelve years, marked

The work did not proceed as a single arc. It proceeded as two research and application phases of roughly six years each, separated by the decision to become a company.

  1. 2015

    The question is taken up seriously

    Evan Valemont and Ryan Mercer decide to answer the question they carried out of Cambridge: whether institutional research methods can be rebuilt into something clearer and more verifiable. No product is announced and no immediate result is produced. Evan works on market structure, risk boundaries and judgment logic; Ryan tries to embed that logic into data architecture and models.

  2. September 2020

    Wintermute AI is founded, in the fifth year of research

    Five years into the work, the two make a second decision and incorporate their first company. The moment matters less as a beginning than as a change of register: from here, research cannot stay inside notes and code. Someone has to be accountable for how data enters the system, how models are deployed, who checks for risk, and how results are reviewed.

  3. 2021

    CortexQuant moves towards practical application

    Roughly six years after the first question, the initial problems begin to take shape as a framework that can exist in real-world environments. To outside observers this looks like an achievement. Inside the work, the harder phase has just started: markets change, data gaps appear, and relationships that were valid stop being valid.

  4. 2022 – 2026

    Five years of absorbing change

    The system is deliberately exposed to data changes, strategy adjustments, risk testing and engineering iterations across different market phases. When results diverge from expectations, the same two questions are asked — has the market structure changed, or has the structure been misunderstood? Each deviation is traced back to its source: input data, model assumption, or the operating system itself.

  5. September 2026

    Renamed Valemont Invest Inc

    The company takes the founders' name. The renaming marks the arrival of the research at the point where its authors and its method are one and the same thing. The verification ethic that the earlier name was meant to protect does not disappear with it; it moves into the working method itself.

  6. 2027

    Global launch of CortexQuant planned

    Valemont has stated that it expects to launch and sell the CortexQuant application globally in 2027. No specific release date has been announced. Until then, the system continues to be described as a research and technology framework rather than a finished commercial product.

Why the rename matters

A company name is usually a marketing decision. In this case it carried some weight, because the original name was doing a specific job.

Wintermute AI, the name the company took in September 2020, commemorated the tutor Evan Valemont and Ryan Mercer shared at Cambridge. It was not a sentimental gesture so much as a reminder bolted to the front door: do not lose the initial thirst for verification. The name kept the standard visible at a point in the company's life when it would have been very easy to stop testing and start asserting.

Renaming the company Valemont Invest Inc in September 2026 changed what the name points at without changing the standard behind it. By then the research had a working form and the two founders had a methodology they shared; the company could be named after the people who would have to keep defending it. The verification ethic did not need a separate monument any more, because it had been absorbed into how the system is built.

It also marks a genuine shift in kind. Before September 2020 the work was a research project with two participants. Afterwards it was an organization with accountable functions: ingestion, deployment, risk review, results review. The name change in 2026 confirms that the organization, not the project, is the enduring thing.

The long validation, and the lesson inside it

The validation phase transformed what CortexQuant is. It began as a research proposition and ended as a technological framework connecting data, market understanding, quantitative models, risk assessment, investment decisions and feedback. That chain is the actual output of the twelve years: not a single clever relationship, but a loop that runs continuously and can be inspected at every stage.

Two commitments held the loop together. The first was a refusal to substitute one success for the next test — a good result did not earn a pass on the following question. The second was a refusal to use technical complexity to mask unanswered questions. Complexity is persuasive; a system with many parts is easy to describe and hard to challenge. That is exactly why it is a tempting place to hide, and exactly why hiding there is fatal to the method.

CortexQuant found its home inside Valemont Invest rather than beside it. It is not a separate company, and it does not define itself by a single strategy. It is the core quantitative research and technology system of the firm — the place where Evan Valemont's judgment about market structure and risk, and Ryan Mercer's ability to build data and systems, stop being two capabilities and become one working methodology. Its architecture and its engine families are described in more detail on the CortexQuant architecture page and in the CortexQuant Lab notes on its engines.

No shortcuts to a worthwhile problem

Counting from the question posed in 2015, the path ran through two research and application phases of about six years each. Even after it finally became a system the path did not become easy, and the record deliberately does not pretend otherwise.

The lesson the work carries is simple to state. There are usually no shortcuts to solving a problem that is worth solving. A hypothesis that has been refuted has to be verified again rather than set aside, and a model that seems to work has to have its boundaries explored rather than trusted. Only that kind of patience turns an idea into a project that can withstand the test of time.

What Evan Valemont and Ryan Mercer took from Cambridge was never a ready-made answer. It was a way of dealing with complexity. Today that way of working is the foundation of Valemont Invest, and it can be summarised in two lines: structure brings clarity, and technology keeps a rigorous method running. The same conviction is carried into the firm's continuing education work, where the Valemont Q4 2026 course extends the research discipline to a wider audience, and into the planned 2027 global release of the CortexQuant application.

Questions

When did Valemont Invest Inc begin?

The research began in 2015, when Evan Valemont and Ryan Mercer decided to answer a question they had carried since Cambridge. The company itself was founded in the fifth year of that research, in September 2020, and was later renamed Valemont Invest Inc in September 2026.

What was Valemont Invest Inc originally called?

The company was incorporated in September 2020 under the name Wintermute AI and was officially renamed Valemont Invest Inc in September 2026. The renaming did not change the research; it aligned the company name with the two founders and the system the research had produced.

Why was the company first named Wintermute AI?

The name Wintermute AI commemorated the tutor Evan Valemont and Ryan Mercer shared at Cambridge and served as a standing reminder not to lose their initial thirst for verification. It kept the obligation to test an idea, rather than merely to have one, visibly attached to the company.

What does each founder do at Valemont Invest?

Evan Valemont works on market structure, financial mathematics and risk frameworks, and continuously defines the research framework and its risk constraints. Ryan Mercer translates those requirements into data architecture and a working technical system, covering model deployment and engineering. The two capabilities are deliberately kept in tension.

What is CortexQuant, and is it a separate company?

CortexQuant is not a separate company. It is the core quantitative research and technology system inside Valemont Invest, and it does not define itself by a single strategy. The research behind it began in 2015 and moved towards practical application from around 2021; a global launch is planned for 2027, with no specific date announced.

What principle did the founders take from Cambridge?

Their shared tutor left them one principle: understand the problem first, then trust the answer. The order is the point. It is not a claim that answers are worthless, but a rule that an answer without an understood problem cannot be checked and cannot be safely relied on.

The philosophy

Structure brings clarity. Technology makes it scalable.

Valemont Invest Inc was founded in September 2020 as Wintermute AI and renamed in September 2026. Everything the research produced in between rests on one habit inherited from a Cambridge tutor: understand the problem first, then trust the answer.