My Polymarket trading strategy has a circuit breaker, and I've fixed the same bug five times.

Today, it happened for the sixth time, and I finally realized: it's not a code bug, it's that I was setting thresholds based on intuition.
Pause trading when the win rate drops below 75%—sounds reasonable, right? I had Claude run 1,179 historical trades to test this, and found:
The actual win rate of this strategy is 78.7%, and during normal fluctuations, it dips below 75% about 25% of the time. That means every four days, it stops itself once.
Even more absurd, in a sample of 50 trades, the difference between 74% and 75% is statistically insignificant, with a p-value of 0.43—statistics tell you these two numbers are essentially the same, but my code treats it as ironclad proof of "strategy failure."
Then, the circuit breaker pauses trading, and when it resumes, it detects the win rate is still 74% (no surprise, no trades means no change), so it pauses again—an endless loop. The strategy was halted for over five hours unnecessarily.
I fixed the recovery logic five times, but today I realized I shouldn't have paused at all.
Recalibrated the threshold with data: increased sample size from 50 to 100, and lowered the threshold from 75% to 68% (which is never touched during normal fluctuations). One fix, done.
Lesson learned: the biggest mistake in writing quantitative strategies isn't the algorithm itself, but relying on gut feeling for parameters.