How Research works

Research helps you turn a market observation into an explicit question, inspect recorded matches and review the evidence behind a result.

1. Choose your starting point

Open the Research Workbench to describe a condition, or investigate a move that already happened. A retrospective move is a starting point for discovery, not independent evidence that its lead-up predicts another move.

If you want to see a complete example first, use the research demonstration or read a published report. Reading an existing artifact is different from requesting new computation.

2. Check the measurement and scope

Use the Reading library for each field’s calculation, unit, observation window and limitations. Research uses versioned measurements; a similarly named live-terminal indicator is not automatically the same calculation.

Inspect the chosen venue, markets and dates. The coverage page describes the available record. Features that are absent or not sufficiently warmed remain absent rather than becoming zero.

3. Review before running

Plain-language interpretation proposes an exact question. Check every field, threshold, inferred value and time window in the visible document before execution. The query grammar owns the supported operators and validation rules.

Creating or materially changing a study follows the account’s current access and allowance rules. A shared link preserves intent; it does not grant computation or replay access.

4. Read the counted result

Start with the occurrence count, eligible denominator and exclusions. Then read the outcome horizons and distribution, comparing them with the stated baseline where available. Inspect uncertainty and dependence between nearby events or markets before interpreting a difference.

Forward price outcomes and excursions are not account returns. They do not include your execution, costs, leverage or exit rules. A result can support further investigation without establishing a profitable strategy or a causal mechanism.

5. Inspect contrasting examples

Open available replay evidence for representative and contradictory matches. Read each example’s selection label: an example chosen for a large later move is illustrative, not typical. Replay availability depends on recorded streams and account entitlement.

Keep the exact receipt when saving or comparing results. A new dataset revision can produce a new result; reproducibility documentation explains what the pinned artifact identifies.

Continue the workflow

Return to saved studies, review live alerts, or use the browser, API and MCP guide to choose another client. When refining a question, keep a log of the alternatives you tried and read selection bias before evaluating it on the same data again.