Research limits

Research reports observations from a specified record. It does not establish a profitable strategy, a causal mechanism or a future outcome.

Setup-time predicates and forward outcomes

In a setup-first query, predicates select conditions observable at the anchor. Forward outcome fields cannot be smuggled into those predicates. The query grammar owns exact validation rules and errors.

Outcome-first exploration is a separate retrospective question: start with moves that already happened and examine their lead-up. Its selection must stay explicit. Likewise, changing how outcome rows are presented does not change the population that was counted.

Absence and zero

A missing or insufficiently warmed measurement is not zero. Read the eligible denominator and exclusions. Zero matches in a valid, covered scan is a different result from having no eligible data or a failed request.

A market outcome is not an account return

A forward price return does not include your entry, exit, fees, slippage, funding, leverage or liquidation rules. MFE and MAE are excursions over a stated window, not executable fills or realized profit.

Reproducibility does not remove selection bias

A pinned artifact lets someone inspect the same evidence. It does not prevent choosing dates, markets or thresholds after seeing outcomes, repeated testing, or publishing only favorable results. Keep a research log and test hypotheses on data not used to select them.

Read selection bias, inspect the data coverage, and retain each result’s receipt when comparing revisions.