From V1 to V6: Building Our Automated Trading System

Our trading system did not begin as V6. It reached V6 through discarded ideas, weak forward results, architecture changes and a growing insistence on preserving evidence instead of polishing it away.

The development path

STAGEWHAT CHANGED

Early orderflow researchWe tested short-horizon orderflow and imbalance ideas. PAPER results showed that plausible microstructure logic was not enough.

V4A simpler adaptive engine became the control branch and a reference point for later systems.

V4.4A revised branch tested changes without rewriting V4 history.

V5.1We explored adaptive ML policies, learned from unstable behavior and preserved the losing research record.

V6We moved the ML experiment onto Freqtrade/FreqAI infrastructure and kept execution in PAPER mode.

The most important architecture change was methodological

Early development made it tempting to respond to every loss with another filter or parameter change. That creates a moving target. We eventually separated versions, kept a control branch and treated material changes as new experiments.

IDEA→FREEZE VERSION→PAPER→MEASURE→NEW VERSION

What survived every redesign

Three principles became more important than any indicator or model: use net outcomes, preserve failed versions, and separate historical simulation from forward PAPER evidence. V6 is not presented as the final answer; it is the current experiment.

Where the project is now

V4 remains a control, V4.4 remains a separate revision and V6 runs as a FreqAI PAPER experiment on BTC, ETH and SOL. Public metrics are sanitized before they reach this website.

Compare current systems Open V6 record →