> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tqx.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Versions and reproducibility

> Use strategy versions to compare code, parameters, and backtest evidence.

Qube creates a new strategy version when a meaningful code or saved-parameter change is recorded. A version identifies the code snapshot and its current backtest parameter baseline. Each backtest run also stores its own effective parameters.

## What a version records

* Version number and optional label.
* Strategy source code.
* Current backtest parameter baseline.
* Creation time and origin, such as manual edit, AI edit, or revert.
* Code and parameter change summary.
* Backtests and optimization tasks linked to the version.

## Compare versions

1. Open the **Versions** tab on the strategy detail page.
2. Select a historical version to preview its code and parameters.
3. Choose **Compare to current** to review code and parameter differences.
4. Label or star versions that represent a meaningful research milestone.

The current editor content is the latest version, also called `HEAD`. A historical preview does not silently replace `HEAD`.

Running a backtest can update the saved parameters used for the next run without creating a code version. Use the run record as the authoritative record of the parameters used by that simulation.

## Edit or revert

When you edit an older version, Qube bases the change on the latest version and saves a new version after confirmation. Reverting also creates a new version; it does not overwrite the historical record.

<Warning>
  A backtest is tied to the version selected when it starts. Editing the strategy later does not change that run. Always record the version and parameters when comparing results.
</Warning>

## Reproducibility checklist

* Record the strategy version and label.
* Keep the market and symbols unchanged.
* Keep the backtest period, capital, frequency, and cost settings from the run record.
* Compare runs with the same data assumptions.
* Treat a changed code or parameter as a new experiment.