Two noisy sensors both print messy 0s and 1s, but one tends to repeat its previous bit after warming up, so what should a randomness assessor worry about first?

Answer: Hidden dependency patterns

Whether heat adds entropy — Heat can be involved in physical noise, but warmer is not automatically more random. If each bit partly copies the previous one after warming up, the extra heat has created a dependency problem rather than a better source. Entropy assessment asks how hard the next value is to guess, especially in the worst case.

Hidden dependency patterns ✓ — Hidden dependencies are exactly the kind of trap entropy tests look for. NIST SP 800-90B treats real entropy sources as often non-IID, meaning the bits may not be independent and identically distributed. Predictor and Markov-style checks matter because a messy stream can still leak patterns that improve an attacker's guesses.

Messier-looking plots — A messy plot can be reassuring to the eye and still miss the real issue. NIST separates output testing from entropy-source assessment because a source can have memory or bias that lowers worst-case unpredictability. The important question is whether the process has dependencies that help prediction, not whether the dots look chaotic.