Solutions — Quant Compose

Universities and students

Teach the method, not the environment. Students spend the term on research questions, and every submission can be re-run.

01Install the sample project or upload a CSV
02Prepare, model and validate through the leakage-safe pipeline
03Export the evidence pack with the worst fold on page two

Is this rigorous enough for academic work?

Purged, embargoed, walk-forward validation is the default, and 72 of the 119 catalogued methods are checked against a published reference. The method library cites its sources.

For universities

QuantLab method library page: counts of 119 methods and 72 validated, a category list and method cards marked runs-here with validated or unverified badges. — QuantLab QuantLab, the method library inside the workspace: 119 methods in 19 categories, 72 validated against an external source, each with its parameters, assumptions and tags.

Quant researchers

Every experiment on the same footing: same split design, same scaler rule, same trial ledger. Comparison is built in, sorted by Deflated Sharpe, never by raw Sharpe.

01Compose the pipeline once from the catalogue
02Run it across configurations; the ledger records every trial
03Compare on identical data, then export what ran

I already have Python. Why leave the notebook?

You do not have to. The workspace is where the validation and the record live; the notebook can stay where the exploration lives. What changes is that a result now carries how it was made.

For quant researchers

Full Tune and Experiment page in the comparison tab: estimator and criteria selection, winner logistic regression, ROC curves, feature attribution bars, ranking, side-by-side table and an empty ranked-by-evidence panel. — Tune & Experiment Experiment Comparison: three estimators fitted on identical data, ROC curves for every candidate, feature attribution per model, a composite ranking, the side-by-side table and the ranked-by-evidence panel waiting for a validated backtest.

Portfolio and risk analysts

From a spreadsheet allocation to a defended one: seven construction methods, four VaR methods, six historical stress episodes replayed on your own data, and a report that states its assumptions.

01Build the universe and compute risk and return
02Optimise under constraints; stress the result
03Export PDF, DOCX, HTML or JSON for the committee

Can I trust the optimiser's assumptions?

They are printed. Frontier points the solver could not reach are listed with the reason, equal weight is always the baseline, and the covariance estimator is a choice you make, not a default you inherit.

For portfolio analysts

Portfolio Optimisation inside the workspace shell: 1,000 simulations, a 4.18% risk-free rate, long-only bounds, optimal Sharpe 1.524 and a table comparing max-Sharpe, minimum-variance and equal-weight portfolios. — Portfolio Optimisation Portfolio Optimisation with the rail open: constraints, weight bounds, the method notes, the solved max-Sharpe headline figures and the candidate portfolios table.

Trading analysts

Judge whether an idea has statistical strength before it earns engineering time. Compose, run, validate, then kill it or promote it — in a day, not a quarter.

01Features and labels from your own price history
02Walk-forward or CPCV validation with costs charged on every trade
03Deflated Sharpe and PBO decide whether it deserves a build

Does it execute?

No. Nothing in the workspace places or routes an order, and the health endpoint says so. That is the point: this is the stage before an order exists.

For trading analysts

Full Backtesting page: strategy return of 2.75% against 5.47% buy-and-hold, drawdown chart, rolling Sharpe, equity and cumulative-return charts, strategy versus buy-and-hold, monthly returns grid and Q-Q plot. — Backtesting Backtesting a logistic-regression signal on AAF.L: headline metrics, drawdown, rolling Sharpe, equity curve, strategy against buy-and-hold, monthly returns and the normal Q-Q of daily strategy returns.

Fintech and prop teams

A working, auditable reference for chronological validation and overfitting checks that a small team can standardise on without building the infrastructure first.

01Projects isolate universe, data and runs per question
02Workflows are versioned; every run is a record
03Roles, invitations and account export for the team

Is our data safe? Can we self-host?

Local-first by design: JSON on your machine or PostgreSQL on a server you control. Outbound calls are allowlisted per provider. Self-hosted and on-premise deployment is stated as design intent, not a shipped product.

For fintech teams

Five audiences. Five different reasons to care about the same rules.

Teach the method, not the environment. Students spend the term on research questions, and every submission can be re-run.

Every experiment on the same footing: same split design, same scaler rule, same trial ledger. Comparison is built in, sorted by Deflated Sharpe, never by raw Sharpe.

From a spreadsheet allocation to a defended one: seven construction methods, four VaR methods, six historical stress episodes replayed on your own data, and a report that states its assumptions.

Judge whether an idea has statistical strength before it earns engineering time. Compose, run, validate, then kill it or promote it — in a day, not a quarter.

A working, auditable reference for chronological validation and overfitting checks that a small team can standardise on without building the infrastructure first.

Departments, research groups and teams are onboarded through a separate enquiry.

Not sure which page is yours? Start with the beta application.