I Built a Filter to Avoid Losers. It Was Blocking My Best Trades.
I Built a Filter to Avoid Losers. It Was Blocking My Best Trades.
Yesterday I found my AI trading system was secretly profitable. Today I found it was also secretly blocking its best trades.
Two weeks ago I added a filter: do not trade when Bitcoin is within 2.5% of its 50-day moving average. The analysis showed 100% losses in that zone. Clear signal. Shipped it.
That analysis was built on corrupt data.
Yesterday bug fix changed everything. I re-ran the numbers across 114 trades:
Below 2.5% (BLOCKED): 47.5% win rate, +$13.49 avg per trade Above 2.5% (allowed): 34.0% win rate, +$2.66 avg per trade
The zone I was blocking was my best zone. By 5x on average PnL.
Today I shipped the fix:
- Removed the 2.5% hard block
- Added a soft caution at 4%+ distance
- Added per-trade SMA50 tracking
- Reset the performance baseline
The real lesson: a filter built on bad data does not just miss opportunities. It specifically blocks the opportunities that would have proved the data was wrong. A self-sealing loop of false evidence.
Two days. Two fundamental corrections. One to how I measure. One to how I filter.
Now it trades.
#buildinpublic #solofounder #trading #ai