How we turn a claim into evidence.

This page describes the default research process used across TRADE EVIDENCE. Individual studies may add stricter rules, but they should not silently weaken these controls.

Core rule: define the event and outcome before inspecting which version looks best.

1. Research question

Each study begins with a narrow question that can be answered by an observable dataset. “Does price respect FVGs?” is too vague. “What percentage of mechanically defined BTC FVGs receive a First Touch within a predefined forward window?” is testable.

2. Operational definition

The concept is translated into explicit conditions, timestamps, boundaries, exclusions, and tie-breaking rules. Exact production configurations may remain private, but the published protocol must explain enough to understand the study and its limitations.

3. Information timing

Every feature and event is timestamped at the first moment it could actually have been known. Pivots that need future bars are not treated as known at the pivot bar. Higher-timeframe values are not leaked into earlier lower-timeframe rows.

4. Data lineage and quality

venue, symbol, product type, and contract specification
canonical UTC timestamp and source availability time
unit definitions for price, volume, OI, Funding, and notional values
duplicate, missing, impossible, and structurally changed records
download date, source version, and exclusion counts

5. Event and outcome separation

Layer	Purpose	Example
Feature	Describe the state available at the timestamp	session, volatility, OI change
Event	Identify the occurrence being studied	FVG formed, level swept
Outcome	Measure what happened afterward	First Touch, forward return, MFE, MAE

6. Costs, overlap, and execution assumptions

Descriptive event studies are not automatically tradable strategies. A strategy backtest must separately specify entry availability, fill assumptions, fees, spread, slippage, Funding costs, position sizing, overlapping positions, and liquidation mechanics where relevant.

7. Reporting

show sample size and missing-data counts
report distributions, not only averages
retain null and contradictory results
label exploratory analysis separately from validation
separate primary rules from sensitivity tests
state the period, market, costs, and known limitations

8. Versioning and corrections

Material changes to definitions, data treatment, or reporting rules should create a new protocol version. Errors are corrected transparently rather than silently reframed as the original method.

Public methodology, protected edge

Published material explains the research question, event logic, data requirements, outcome measures, results, and limitations. Private deployment code, API credentials, production filters, optimized configurations, and execution infrastructure are not part of the public methodology.

Last updated: August 26, 2026.