To break confirmation bias, you need an objective log. Write down every time you recognize a confirming event, as well as every time you recognize an even which is nonconfirming. Then, estimate the likelihood that you would recognize and write down a confirming event, and the likelihood that you would recognize and write down a nonconforming event. Use your surprise that a nonconfirming event just occurred, as well as your surprise that you noticed it and made a note of it to form that estimate.

If you find yourself more surprised that you made a not of a nonconfirming event than that it happened, it probably happens much more often than you note it.