Start with an input contract
Describe the data columns, units, timestamps and expected output. Tell the model which information is unavailable and require it to surface missing assumptions. Do not include credentials or private account data in a prompt.
A reusable specification outline
Use this outline: objective; eligible universe; information available at decision time; signal; order time; exit; position limit; costs; missing-data behavior; tests. Ask for unresolved questions before code, then review each field.
Check behavior independently
Construct examples with no signal, one signal, missing observations and insufficient history. Compare expected decisions with actual output. Keep preprocessing fitted only on the training portion when evaluating a learned model; chronological splits require attention to overlapping information.
Deliver an evidence record
Save the prompt, available model identifier, generated version, manual changes and test outcomes. This site supplies the workflow and reference tools; it does not host a model runtime or execute generated trading code.
Start the practical task
Use the research studio to document rules or inspect a suitable equity series. Review the input assumptions and export your work before leaving.
Open the workspace ↗Questions about this workflow?
For a content question or suggestion, describe your research objective, asset type, current stage and the step you need explained. Please do not send account credentials or trading keys.
info@aiquantbase.com ↗Sources & references
- scikit-learn — Cross-validation and time series evaluation ↗
- scikit-learn — Data leakage and common pitfalls ↗
References support specific definitions or methodological context. Examples and research checklists are explanatory; no performance result is claimed.