Freqtrade Strategy Development: A Versioned Workflow
A practical workflow for developing Freqtrade strategies without turning every losing sequence into another round of curve fitting.
Define the hypothesis before parameters
Write down the market behavior the strategy is intended to exploit, the conditions under which it should fail and the measurements that would reject it. Parameters should implement the idea rather than become the idea.
Separate entry, exit and risk
Treat signal generation, exits, sizing and portfolio constraints as distinct layers. When performance changes, this makes attribution possible.
Freeze meaningful versions
Material logic changes should create a new version. Do not silently rewrite the historical record after seeing forward losses.
Promote evidence in stages
Use historical tests to reject weak candidates, then forward PAPER testing to observe unseen behavior. Live capital is a separate evidence stage with real execution risk.
Before trusting the result
Confirm that data timing is valid, configuration is reproducible, state survives restart, costs are represented and the exact strategy version is preserved. Operational correctness comes before interpreting PnL.