The photograph of the eclipse

In 1919 two expeditions photographed stars near the sun during a total eclipse. The point was to measure whether starlight passing close to the sun was bent, and by how much.

Newton's physics predicted a small deflection. Einstein's predicted roughly twice as much. The difference was tiny, close to the limits of what the equipment could measure, and the result was announced as a confirmation of Einstein.

Historians of science have examined the plates since, and the episode is messier than the story. One set of photographs was discarded as unreliable. The error bars were large enough that a careful statistician at the time could reasonably have called the result inconclusive. The announcement was made with more confidence than the data supported, and it was right.

This makes the case uncomfortable to teach, because it resists both of the easy lessons. It is not a story about science working perfectly, and it is not a story about a conclusion being wrong. The measurement was repeated with better instruments over the following decades, and the prediction held each time.

What the episode actually shows is how science handles an underpowered result: not by trusting it, and not by discarding it, but by treating it as a reason to measure again. The 1919 photographs did not establish the theory. They made it worth the expense of trying to refute it properly, which is a smaller claim and a more accurate description of what evidence usually does.