Define the observation
Specify the instrument, venue, time window and collection method. A displayed quote, an executed trade and a reconstructed event stream answer different questions.
Preserve provenance
Record the source and version of every input, collection time, missing observations and asynchronous fields. Distinguish an interface screenshot from an API response or a full event history.
Match test and claim
Descriptive observations explain a sample. Predictive claims require a target and out-of-sample testing. Causal claims require separate identification. A small historical sample alone cannot establish a robust trading strategy.
Include operating conditions
Execution research must consider information delay, order size, fees, queue position and unfilled orders. A decision may use only information available at the time.
Label status and revisions
Preprints, preliminary market observations and synthetic experiments are different forms of evidence. If inputs or conclusions change materially, update the status and version.
Send a question or correction