29 questions, answered briefly and honestly.

No. It is a research and analytics workspace for the stage before an order exists. Nothing in it places trades, routes orders or connects to a broker, and its health endpoint reports live_order_routing: false. Strategy Monitor tracks strategies you have built and backtested; it is a record, not a connection to a market.

No. Every stage is configured rather than scripted: you choose methods, set parameters and run. The method library explains what each method assumes and refuses. A custom-script space, Quant Code, is on the roadmap and labelled as future wherever it appears.

The research layer's method catalogue: 119 named methods across 19 categories, from data quality checks and statistical tests to portfolio construction, option pricing, stress testing and reporting. 72 of them have been checked against a published worked example, a closed-form identity or an independent library. It is not a count of machine learning models; the ML Models page offers 43 estimators, and the site never adds the two together.

14 classifiers (Logistic Regression through XGBoost Classifier, SVC, KNN, LDA, QDA, AdaBoost, MLP), 16 regressors (Linear, Ridge, Lasso, Elastic Net, tree ensembles, XGBoost Regressor, SVR, Bayesian Ridge, Huber), 6 clustering methods, 4 dimensionality reducers and 3 anomaly detectors. All 30 supervised models train through one chronological pipeline. Four lag-based methods serve forecasting.

Yes. A project owns its universe, benchmark, date window, data lineage, features, runs, strategies, saved results and reports, and nothing from one project appears in another. Switching persists the outgoing project first.

A snapshot of one panel, chart, table or metric block together with the parameters that produced it, filed into folders that mirror the workspace. Saved results are immutable in substance: they can be renamed, moved and annotated, never edited. They do not restore application state.

A way to write a research pipeline as an ordered, validated, versioned list of steps drawn from the method catalogue. A workflow is a specification; running it produces a record of which steps ran, were skipped, were blocked or failed, with the parameters actually used. Five templates ship as starting points.

Chronological split, scaler fitted on the training block only, no shuffling, and a purge gap floored at the label horizon. Walk-forward validation reports mean, standard deviation, minimum and maximum for every metric, so the worst fold is always visible. Switch any safeguard off and the page states what leaks.

The forecast that tomorrow's close equals today's. Every forecasting run computes it on the same held-out window and reports a skill score of 1 − RMSE ÷ baseline RMSE, positive or not. A model that cannot beat that baseline is not adding information, and the page says so.

A Sharpe ratio adjusted for the number of configurations tried and for non-normal returns. It requires knowing how many trials there were, which cannot be reconstructed afterwards, so the workspace records trials in an append-only ledger during a search. Without a trial population the statistic is reported as not available rather than estimated.

No, and the site does not say it is. Each experiment record carries a SHA-256 hash of its inputs so identical configurations hash identically and a divergent re-run is caught. Validation verdicts are write-once, deleting a validated record needs explicit confirmation, and the trial ledger is append-only by construction. That is a reproducibility mechanism, not a cryptographic chain.

A method is validated when at least one reference case of an external kind exists for it: a textbook worked example, a value from a paper, a closed-form identity, hand-worked arithmetic, an independent library or a regulator's example. A test whose expected value came from the codebase itself earns nothing, because it cannot catch a formula that is wrong and consistent.

Two ways. A historical replay applies the returns your securities actually printed during a named episode — six episodes from the 2008 crisis to the 2020 crash — and refuses if your data does not cover the dates. A hypothetical shock propagates a market move you specify through each asset's beta and rescales the covariance matrix. Neither is a forecast.

Yes, that is the intended workflow. Upload a CSV of daily prices with date, ticker, open, high, low, close and volume columns, or connect a market-data provider such as Yahoo Finance, FRED, Ken French, Alpha Vantage or Polygon, or a custom REST endpoint. Fifteen bundled index constituent lists give you a universe to start from, and every provider states its status in Settings.

Locally by default: JSON documents under a data directory on your machine, or PostgreSQL on a server you control when configured. Exports go to a local directory or an S3-compatible bucket. The hosted beta runs on infrastructure the company operates; see the security page for what leaves the machine.

Only calls to a market-data provider you have chosen, over an allowlist of approved hosts, HTTPS only, with every redirect re-checked and private addresses refused. The frontend makes no request to any origin other than its own; there is no analytics, telemetry or CDN in the workspace.

Stored server-side. The API can report that a credential is configured and where it came from, but never returns the value, and every secret is redacted from logs, error messages and provenance records before they are written.

No, and it is labelled as such everywhere it appears. The sample project generates deterministic series from a seed with no network call, so the workspace can be demonstrated and tested identically on every machine. It is never a basis for a claim about a real security.

The workspace runs from a Python environment or a Docker image on a machine you control today, and that is how development happens. Supported on-premise or VPC deployment for institutions is stated as design intent, not a shipped product.

Access opens in stages, by cohort, and each cohort is small enough to talk to. Apply on the beta page; an invitation arrives by email when your cohort opens. There is no self-serve sign-up during private beta.

Nothing during private beta. The planned plans — Researcher, Analyst, Desk and Institution — are published on the pricing page with what each includes and its limits, labelled as planned. Pricing is announced before billing goes live; a beta account is never charged.

The planned Researcher plan is free and is designed with students in mind. Cohort and department access for universities is handled through the institutions enquiry and priced case by case, as design intent until it ships.

The workspace models owner, admin, quant researcher, analyst and viewer roles, invitations with seat limits, account export and an audit log. Team onboarding during beta is arranged through the institutions enquiry.

The validation defaults follow published methodology (chronological splits, purging and embargoing after López de Prado, walk-forward folds, deflated and probabilistic Sharpe ratios after Bailey and López de Prado), and the method library cites the reference each validated method was checked against. What a marker will see is a report that states its split design, its trial count and its worst fold.

A project carries the universe, the data lineage, the parameters, the seed and the engine version, and a workflow is a versioned specification. Re-running the same workflow on the same data produces the same record. The sample project is deterministic by construction.

No. It teaches the vocabulary the workspace uses and the method behind each stage, with worked examples and sources. Each article links to the platform page that implements the idea.

Current versions of Chrome, Edge, Firefox and Safari. Fonts are bundled so the interface renders identically offline.

Yes. The workspace exposes a documented JSON API, and API keys can be created under Settings with an optional enforcement gate for programmatic callers. The evidence pack is available as JSON so a downstream service can consume it without parsing a PDF.

It refuses and says why. Too few observations, a missing column, a portfolio bound the solver cannot satisfy, a stress episode outside the loaded dates — each produces a named reason rather than an empty chart or a zero, and an undefined statistic renders as a dash, never as 0.