// the find
buckyroberts/Ella
Self-improving decision organism
Ella is a toy trading bot that generates random condition-action pairs (neurons) from dataset statistics and fires them row by row. The author freely admits it was written in a morning while waiting for a hockey game. It is for beginners curious about rule-based systems, not anyone doing real algorithmic trading.
The neuron/brain abstraction is clean and easy to follow — Action, Condition, Neuron, Brain each live in their own file with obvious responsibilities. The example output is honest about what the system actually does, which is more than most weekend projects manage. The Poloniex crypto dataset included means you can run it immediately without scraping your own data.
The 'self-improving' label is false advertising — conditions are generated randomly once and never updated based on outcomes, so nothing actually learns. There is no fitness evaluation, backtest loop, or any mechanism to prefer rules that work over rules that don't. Last commit was November 2023 and the author explicitly said it probably won't be maintained. The trading simulation doesn't account for fees, slippage, or position sizing, so the $1008.65 result is meaningless as a performance number.