On May 29, 1919, Arthur Eddington watched the sky over Príncipe, an island off West Africa, where days of cloud thinned only during totality, a little after two in the afternoon, and cleared fully once the eclipse had ended. An ocean away, at a racecourse in Sobral, in northern Brazil, Andrew Crommelin and Charles Davidson watched a cloudy morning clear just before totality. Each team had under six minutes of darkness to test one prediction: that the sun's gravity would bend a ray of starlight by a precise, calculable amount. That prediction had reached Eddington only because a friend in neutral Holland forwarded it from Berlin, since the war had cut off ordinary contact between British and German physicists. The war itself had ended six months before the eclipse, but the theory two British expeditions crossed an ocean to test was still, undeniably, a German one.
Einstein's general relativity, worked out in Berlin in 1915, already had one kind of support: it explained a swing in Mercury's orbit that Newtonian physics could not account for. The eclipse tested a second, independent prediction, that starlight passing near the sun's edge would bend by 1.75 arcseconds, about the width of a coin seen from two miles off. Newton's older physics, if light were subject to gravity at all, predicted half as much, 0.87 arcseconds. A total eclipse was the only moment such starlight became visible at all, so Frank Dyson, the Astronomer Royal, organized two expeditions to photograph the same patch of stars, the Hyades cluster, during the few minutes the moon blocked the sun's own light. Sobral's plates were checked months later against photographs of the same stars at night, with no sun nearby to bend anything; Príncipe's were checked against earlier photographs of the same field, taken at Oxford before the equipment was shipped out.
Sobral's plates, measured back in England, gave a deflection of 1.98 arcseconds, with a probable error of about 0.12: higher than Einstein's predicted 1.75, but still far above Newton's 0.87. Príncipe's plates, taken through drifting cloud and measured less precisely, gave 1.61, with a probable error of about 0.30, a range wide enough to include Einstein's number and nowhere near Newton's. Dyson, Eddington and Davidson wrote up the results for the Royal Society without hedging the conclusion.
None of this turned on whether Einstein was liked, or on the country that had produced him. Eddington, a Quaker who had refused conscription, received a year's deferment from military service, after Dyson intervened with the authorities, so he could prepare for the expedition instead of serving. A hypothesis that predicts a specific, checkable number can be tested by anyone willing to look, regardless of who proposed it or where he lived: a clear sky, a camera, and a willingness to publish whatever the plates showed were all it required.
The sky did not know or care whose theory it was confirming. It returned a number, and the number was Einstein's, not Newton's.