Why We Moved Our V6 Experiment to FreqAI
Why we stopped extending a custom adaptive ML trading engine and moved V6 onto Freqtrade/FreqAI while keeping the experiment in PAPER mode.
The custom engine taught us what was hard
V5.1 made policy selection, reward design, portfolio exposure, execution and learning behavior our responsibility at the same time. That produced useful research, but too many moving parts complicated attribution.
Infrastructure should not be the hypothesis
We wanted V6 to test ML/trading behavior rather than our ability to reinvent every part of a trading framework. Freqtrade and FreqAI provided a more structured base for execution, persistence and model integration.
The move did not create an edge
Changing frameworks does not make a strategy profitable. V6 still has to earn evidence through unseen forward PAPER trades, net outcomes and stability across time.
What stayed the same
We kept version history, PAPER-first evaluation, BTC/ETH/SOL scope and public evidence. Failed earlier branches remain visible rather than being rewritten as steps that were always successful.
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.