Why strategy changes need version history

Strategy edits break the continuity of a backtest record, so presenting all results as one uninterrupted history is misleading. Traceable research stages, auditability of iterative testing, and metrics like the deflated Sharpe ratio are designed for multiple-testing contexts. The defensible conclusion is process-based: version history is essential to preserve provenance and prevent revised strategies from inheriting performance records that belonged to earlier versions.

A trading strategy that changes over time is not a single immutable object. If its rules, filters, parameters, or execution logic are edited, then presenting one continuous record of results as though nothing changed creates a false impression of continuity. The record may look like one unbroken strategy history, but it actually mixes multiple versions. Versioning is what keeps that record honest.

Sonar’s research-to-publishing framework is built around this distinction. In that workflow, research moves through explicit stages, including documented testing and publication, rather than being treated as a single opaque artifact. The point of that process is traceability: what was tested, when it was tested, and under which assumptions. That structure matters because once a strategy is edited, the evidence attached to the prior version no longer fully describes the current one. Treating all results as one uninterrupted series obscures that change rather than documenting it clearly.[1]

This is closely related to backtest overfitting. Sonar’s Backtest Overfitting Audit describes overfitting as the risk that repeated testing, tuning, and selection can make a strategy appear stronger in historical data than is justified. The tool is designed to audit the research process around that risk, rather than simply accept a polished equity curve at face value. That is exactly why version history matters: if multiple revisions are folded into a single performance record, the audit trail of what was changed and when becomes harder to inspect. A record with no visible joins can hide the extent of iteration that produced it.[2]

Suppose a researcher tests an initial strategy specification, then adjusts parameters or filters after reviewing historical results, then keeps the old backtest history alongside the newer specification as if it all belonged to one stable strategy. The resulting record would suggest a continuity of evidence that does not exist. The earlier segment reflects one version; the later segment reflects another. Without explicit version boundaries, the combined history implies that the final strategy had always generated the full record, which is not true.

The benefit of versioning here is qualitative, and it is worth stating it that way. Audit methods exist to expose how iterative testing and selection can distort the interpretation of backtest evidence. Versioning improves auditability and removes the temptation to present iterative edits as one uninterrupted backtest history; how much any given record was distorted is a question each audit answers for itself.[2]

The deflated Sharpe ratio offers another reason to separate strategy versions. Sonar’s glossary describes the deflated Sharpe ratio as a Sharpe-ratio adjustment intended to account for multiple testing and selection effects, helping distinguish apparently strong results from those that may have arisen through repeated trials.[3]

In other words, it is explicitly concerned with the problem created when many ideas are tried and the most attractive result is highlighted afterward.[3]

That framing is important for version history. If performance from materially different strategy versions is aggregated into one smooth record, the presentation can understate how much testing and revision occurred behind the scenes. Segmenting the performance by version does not magically solve overfitting, but it does preserve the information needed to interpret the record in light of iterative research. It shows where one specification ended and another began, making it easier to evaluate whether later results belong to a refined successor rather than the original design. This is aligned with the purpose of the deflated Sharpe ratio, which is to interpret performance in the context of multiple trials rather than ignore them.[3]

For quantitative traders, researchers, and risk analysts, that distinction is foundational. A strategy record should describe the strategy that actually existed at the time the record was produced. When the strategy changes, the record should branch, not pretend continuity. An edited strategy with one unbroken history of results is a fiction. Version history is the mechanism that keeps the historical record tied to reality.[1][2][3]

Drafted with AI assistance from cited sources. Reviewed and approved by Sonar Sciences Quant & Research Team.