Applications
Always ask about sample size before drawing conclusions. A pattern in ten observations means less than the same pattern in ten thousand. Demand adequate data before making important decisions. When sample sizes are small, acknowledge huge uncertainty rather than pretending to knowledge you don't have. Provisional conclusions from small samples are fine; confident action is not.
Separate local patterns from global probabilities. Just because you understand the immediate dynamics doesn't mean you can predict long-run outcomes. Market traders can understand daily price movements yet still can't beat long-term returns. Poker players can play individual hands optimally yet variance still dominates short-term results. Design strategies that work across both timeframes.
Build adequate sample sizes before testing. Don't launch products to ten users and declare victory or failure. Don't make hiring decisions based on one interview. Don't change strategies after one bad quarter. Set thresholds for evidence that account for normal variation. The more important the decision, the more data you need before committing.
Use base rates as starting points, then update with specific evidence. Bayes' theorem provides the mathematical framework: start with prior probabilities, update based on new information, arrive at posterior probabilities. This prevents both ignoring patterns (too little updating) and chasing noise (too much updating). The base rate is your anchor; evidence determines how far you move from it.