Blog

Notes from building it, written to be checked.

Short pieces on decisions the workspace makes and why. Each claim traces to the source. An RSS feed is available.

RSS feed

  1. How private beta access opens

    Quant Compose is in private beta. Access opens in stages by cohort, invitations arrive by email, and there is no self-serve sign-up during this period.

  2. Research, not execution: where Quant Compose stops

    The workspace is a research product. It does not place trades, route orders or give investment advice, and every generated report says so.

  3. The worst fold is the headline

    A mean ROC AUC of 0.62 can hide a fold that barely beats a coin flip. The number that should lead a validation report is the worst one, not the average.

  4. What '119 methods' means, and what it doesn't

    The catalogue spans 119 research methods across the whole workflow. Only 17 describe predictive-model families, and none is a count of selectable models.

  5. Why the workspace counts your trials against you

    A backtest chosen from sixty tried configurations is a weaker claim than the same number tried once. How the workspace tracks that difference.

  6. Bring your own data: the CSV contract and the provider allowlist

    What the workspace requires from an uploaded CSV, what happens when a column is missing, and which data providers it will and will not fetch from.

Read the method, then run it.

QuantLab teaches the vocabulary; the workspace implements it under the same names.