Why We Keep Failed Trading Bots in Our Research Record
Why preserving failed automated-trading versions improves research quality and prevents hindsight from rewriting the development history.
Failure is part of the dataset
A discarded strategy tells us which hypothesis, implementation or operating condition failed. Deleting it removes information that can prevent the same mistake from returning under a new name.
Survivorship bias starts inside projects
If only the latest successful-looking branch remains visible, the development record becomes artificially clean. The same distortion that affects strategy databases can affect a single research project.
Version history improves attribution
Preserved versions let us ask what changed between systems rather than comparing a current result with a memory of an older implementation.
Public failures create useful constraints
Publishing weak PAPER outcomes makes it harder to cherry-pick the narrative later. Our orderflow and V5.1 records are retained for exactly that reason.
How we use this lesson now
We keep experiment versions separate, publish unfavorable PAPER outcomes, record net closed-trade metrics and avoid converting an early result into a profitability claim. The purpose of the record is comparison and diagnosis.