Building a research question

Choose observable conditions, confirm their exact definitions, and inspect the scope before running.

Start with a market condition that can be stated using the versioned Reading library. It is the canonical inventory of measurements, calculations, units and limitations; this guide explains how to use them in a question.

Choose the measurement

A question might concern top-of-book imbalance, signed taker flow, funding, or a descending resistance state. Read the specific field before choosing a threshold. Similar live-terminal names can have different scopes or parameters.

Define the scope

In the Research Workbench, inspect the selected markets, venue, dates and available coverage. Use the visible controls and confirmed query as the authority for this run; do not infer defaults from an example in a guide.

Confirm the exact question

Plain-language input proposes a versioned document. Review its fields, thresholds, windows and any inferred values before execution. A proposal is not a result. The API grammar owns supported operators, windows and sequences.

Setup predicates describe information at the anchor. Forward outcome columns describe what followed. Outcome-first exploration selects moves retrospectively and must retain that conditioning; it does not turn the resulting setup into an unbiased forecast.

Read the answer

Inspect occurrence counts, the eligible denominator, missing-data exclusions and explicit outcome horizons. Compare with the appropriate baseline. Open both supportive and contradictory examples in replay; replay access depends on the event and account entitlement.

Save useful work through the workspace, or inspect published reports before designing your own question. See Research limits and selection bias before interpreting the result.