Running in production, executing every US trading day
One process, applied the same way every trading day.
Curson is the systematic platform we build and operate. It forms return forecasts across a universe of several hundred US large-cap equities, constructs a risk-controlled long and short portfolio from those forecasts, and implements it in the market. The process does not vary with the mood of the morning, and no one has to be at a desk for it to run.
Where it stands
- In production since
- June 2026 — the process has run unattended on every US trading day since.
- Trading a balanced book since
- July 2026 — long and short, rebalanced daily.
- Stage
- Pre-launch. Execution runs in simulation while we build the track record that precedes launch.
- Performance
- Not published. We show no returns, simulated or otherwise, on this site.
Investment philosophy
Why we think a systematic process is the right way to hold equities.
We believe that a disciplined process, applied consistently across a broad universe, is a sounder basis for investment decisions than judgement exercised under time pressure. Markets present a small number of persistent regularities — in price trends, in valuation, in how quickly information is absorbed — and we believe those regularities are better captured by measuring many names the same way than by forming an opinion about a few.
We aim for breadth rather than conviction. A process that takes a modest, well-sized position in several hundred names does not need to be right about any one of them, and it can be evaluated statistically in a way that a concentrated bet cannot. That is the trade we have chosen: give up the story, keep the evidence.
We also believe that most of the difficulty in systematic investing is not in the models. It is in doing the same thing every day without drift: the same universe, the same definitions, the same limits, the same decision when the data is late or wrong. Our process is built so that the boring outcome is the automatic one.
The investment process
Five stages, in order, once a day. Each hands a finished result to the next, and each can be inspected on its own.
Universe and data
We define the investable universe from liquidity and listing criteria rather than from a view, and refresh it on a schedule. Prices, corporate actions, dividends, fundamentals and macro context are collected daily and stored so that any past day can be reconstructed exactly as it was known then, which is what makes an honest evaluation possible at all.
Signals
Dozens of families of signals are computed across the whole universe every night, spanning price behaviour, valuation and quality, market sensitivity and macro conditions. Each definition is versioned, so a forecast can always be traced to the exact inputs that produced it.
Return forecasts
Signals are combined into a cross-sectional forecast for every name in the universe before the market opens. The models in production hold their place only for as long as no candidate outperforms them on data neither has seen; nothing is promoted because it looks reasonable.
Portfolio construction
Forecasts become a portfolio, not a list of ideas. Positions are sized within limits per name and in aggregate, long and short exposure is balanced with a market-neutrality objective, and the risk gates described below sit between the target portfolio and the market.
Implementation
We trade the difference, not the portfolio: the target is compared with holdings as the broker reports them, and only the orders that close the gap are sent. Implementation is treated as part of the return, so the process is deliberately unhurried about it.
Oversight
Every stage records what it did and what it decided. A stage that fails raises an alert within minutes, each day's exceptions are reviewed rather than filed away, and the founders see the state of the whole cycle without having to ask it.
Risk management
Risk control is part of the process, not a report written after it.
Diversification first. The primary control is the shape of the portfolio itself: many modest positions rather than a few large ones, balanced long and short with a market-neutrality objective, and limits on how much any single name or the book as a whole may carry. Those limits are inputs to construction, so a portfolio that would breach them is never built.
Gates between the portfolio and the market. Before anything is sent, a fixed sequence of checks has to pass: the forecasts have to be current, the account has to read as expected, exposure has to sit inside its limits, and the size and number of orders have to stay under their ceilings. One failing check stops the whole cycle and raises a single alert. There is no override.
Do nothing, deliberately. If the data behind a day is incomplete, the process holds cash rather than trading a partial view. Going to cash is visible and reversible; a half-built portfolio is an exposure nobody chose. We would rather explain a quiet day than an accidental one.
Reconciliation. Positions and account value are read from the broker before every cycle rather than remembered from the last one, and a discrepancy is investigated and re-anchored rather than compounded. What we report about the portfolio always traces back to what the account actually held.
Research process
Hypothesis, evidence, then implementation — and a willingness to publish the negative result.
Research runs as a loop rather than a project. A candidate signal or model starts as a stated hypothesis, is measured against the production process on data it has not seen, and is either promoted or discarded on that evidence. Because every historical day can be reconstructed as it was known at the time, a result can be attributed to the inputs that produced it instead of to hindsight.
An example, including the part that did not work. In July 2026 we built news sentiment end to end and tested two independent ways of reading the news, then measured a candidate model against the process already in production. Neither version improved on it. We removed the signal, recorded the finding, and left the groundwork in place in case a longer horizon changes the answer. The process trades on price behaviour, valuation and quality, market sensitivity and macro context.
What we are working on. Deeper evaluation of the production models, so their results can be attributed to a market regime and to a family of inputs; a further family of signals that needs more history before it can be judged fairly; and the length of record that every commercial decision on the business model page is gated on.
Technology
The reason two people can run a daily process at all.
We treat the platform as the firm's principal asset and invest in it accordingly. It runs on managed cloud infrastructure so that no part of the daily cycle depends on a machine anyone maintains by hand, and the whole environment — the process, its schedule, the limits it enforces — is defined as code, reviewed before it changes, and applied automatically. There is no manual path to production.
That discipline is what buys the capacity. The same cycle that runs today, on the same infrastructure and at essentially the same cost, works on a larger balance and a wider universe. Building the machine first is a deliberate choice about where a small firm's effort compounds.
What this is, and what it is not
Stated plainly, because the category invites assumptions.
It is a systematic process we run ourselves. Curson is technology built and operated by Curson Capital for its own account. It is not a fund, it is not a collective investment scheme, and it does not manage money for anyone else.
Nothing on this site is an offer or advice. This page describes a process and its current stage. It is not an offer to sell or a solicitation to buy any security or interest, it is not a recommendation, and it is not personalised investment, tax, legal or accounting advice. We hold no authorisation to provide those services and have not sought one.
No performance is shown. We publish no returns on this site, whether live, simulated or hypothetical. When there is a record worth discussing, it will be presented with the limitations of the method it came from stated alongside it — not as a headline number.