Validate patterns with three filters: sample size, base rate, and two alternative explanations

Why This Is a Rule

Your brain is a pattern-detection machine that finds patterns even in random noise — pareidolia for behavior. A perceived pattern that hasn't survived validation is as likely to be apophenia (seeing meaning in randomness) as genuine signal. Three independent filters catch the most common sources of false patterns:

Filter 1 — Sample size: How many occurrences vs. how many opportunities? "This happens every time" based on 3 occurrences out of 50 opportunities is a 6% rate, not "every time." Occurrences must be compared to the denominator of opportunities.

Filter 2 — Base rate: Is the frequency actually higher than background? "I always get sick when I travel" — but you also get sick when you don't travel, at roughly the same rate. The pattern feels real because you notice sickness-while-traveling more than sickness-while-home. Compare your pattern frequency to the base rate of the phenomenon.

Filter 3 — Alternative explanations: Can at least two other explanations account for the same data? Generating alternatives forces you to consider that your preferred explanation isn't the only one, weakening premature commitment to a single causal story.

A pattern that survives all three filters is genuinely meaningful. A pattern that fails any filter needs more data or a revised interpretation.

When This Fires

• After identifying a pattern and before acting on it
• When a perceived pattern is driving a decision (schedule change, investment, behavior modification)
• During any self-analysis where "I always..." or "this never..." claims emerge
• Before recommending a personal pattern to others

Common Failure Mode

Running one filter and skipping the others. You check sample size (enough occurrences) but skip base rate (the pattern occurs at background frequency) and alternatives (your preferred explanation isn't tested). Each filter catches a different type of error. All three must pass.

The Protocol

Before concluding a pattern is meaningful: (1) Sample size: list occurrences AND opportunities. Calculate the actual rate. Is it high enough to be non-random? (2) Base rate: what is the background frequency of this phenomenon without the suspected cause? Is your pattern significantly above baseline? (3) Alternative explanations: generate at least two alternative explanations for the same data. Can you rule them out? (4) Only patterns surviving all three filters → act on. Patterns failing any filter → gather more data or revise interpretation.