About
Built because the research kept falling apart between the tools.
A universe that quietly differs from one analysis to the next. A backtest on data cleaned differently from the features. A Sharpe ratio quoted without the number of trials it took to find it. Quant Compose exists to make those things impossible to do by accident, and visible when they are done on purpose.
01Why this exists
The methods were never the problem. Nothing connected them.
Quantitative research tends to scatter across notebooks, spreadsheets and scripts. Each is good at its job. The trouble is the handoffs: every one is a manual copy, so the state that matters, which universe, which window, which cost assumption, which split, lives in someone's memory rather than in the work. When a result has to be defended by a person who did not produce it, that memory is not available.
The workspace started as an answer to one question: what would it take for a number in a report to carry, by construction, everything needed to check it? The answer turned out to be a project that owns its state, a catalogue of methods with typed parameters and cited sources, validation that is switched on before you arrive, and a record that cannot be quietly edited. Everything else on this site follows from that.
02Principles
Six rules the code is written to.
These are the workspace's own words, taken from its onboarding and its documentation, and they are enforced in the code rather than stated in a brochure.
- 01
It refuses rather than guesses
Too few observations, a missing column, no data: the workspace says what is missing instead of drawing a blank chart. An undefined statistic renders as a dash, never as zero, because zero reads as "the worst possible value" and a dash reads as "not available".
- 02
Every result carries how it was made
Method, assumptions, data hash and who started it stay with the number, so it can be traced back six months later. A saved Sharpe ratio with no record of the window, the cost assumption or the feature set that produced it is not evidence.
- 03
It counts your trials against you
A ledger of what you tried deflates the best result, including the securities you looked at and dropped. The Deflated Sharpe Ratio refuses to compute without it, because an estimate without the trial count is not conservative, it is a fabrication.
- 04
Workflows are repeatable
Build a pipeline once, version it, run it again on new data; each run is an immutable record of which steps ran, were skipped, were blocked or failed, with the parameters actually used.
- 05
Synthetic data is labelled
The sample project runs on generated series that are reproducible from a seed and marked as such wherever they appear. They demonstrate the workspace; they are never a basis for a claim about a real security.
- 06
Research, not execution
Nothing places or routes an order. That is not a gap waiting to be filled; it is the boundary that keeps the product honest about what a result means.
03How it is built
A research layer at the centre, a workspace around it, and a validation record for every method.
The workspace is built around a research layer that holds the 119-method catalogue: a self-contained, tested core with deliberately few dependencies, so a method behaves the same wherever it runs. The application around it handles projects, data, reports and accounts, and the interface is designed to render the same on every machine, online or offline.
Every change ships through the same automated contract: the whole suite passes, the application starts with live order routing disabled, and no file may become less clean than it was. A validation registry grades the evidence behind each method, and a method earns the label validated only from a reference case that did not come from the codebase itself.
04Where it is
Independent. In private beta. No institutional customers to reference, and this page will not pretend otherwise.
Quant Compose is proprietary software developed in a private repository in London, England, and the people building it read every application. Access opens by cohort so that each cohort is small enough to talk to. Enterprise and university deployment are described on this site as design intent, once, and then the pages get on with being specific. The roadmap separates what exists from what is planned and from what is possible, and only the first list is a claim.
If that sounds like how you work, bring a question.
Private beta is free. Applications are read by a person.