In 1975, Edgar Allin proposed that cynodonts — 250-million-year-old mammal predecessors — possessed an eardrum membrane stretched across a curved part of the jawbone. Before this, most researchers assumed early cynodonts detected sound through bone conduction: vibrations picked up by pressing the jaw against the ground. Allin's hypothesis was structurally plausible but empirically untouchable. No method existed to test whether a membrane across that particular bone geometry could actually transduce airborne sound. The hypothesis sat for fifty years.
In 2025, Alec Wilken and colleagues at the University of Chicago CT-scanned a Thrinaxodon liorhinus specimen from UC Berkeley's museum — the same kind of fossil available to Allin's generation — and ran finite element analysis on it. Strand7 software, designed for simulating stress on bridges and aircraft, modeled how different sound pressures and frequencies would move through the ancient skull. Result: the eardrum worked. Thrinaxodon could detect airborne sounds from 38 to 1,243 hertz, most sensitively at 1,000 Hz and 28 decibels. Sensitive hearing evolved 50 million years earlier than the field assumed.
The evidence didn't change. The fossil was in the same museum. Allin's hypothesis was structurally identical to the confirmed result. What arrived was the tool — computational stress analysis from civil engineering, combined with CT imaging from medicine. Neither tool was developed for paleontology. Both had to exist, and be cheap enough to apply to an academic question, before the hypothesis could be tested.
A field can generate a correct hypothesis and be unable to confirm it, not because the evidence is missing but because the method for reading the evidence doesn't exist yet. The bottleneck isn't always in the question or the specimen. Sometimes it's in the analytical vocabulary — and the right vocabulary comes from a domain that had no interest in the question.