Pitchblende residue from the Joachimsthal ore taught me

The pitchblende residue from the Joachimsthal ore taught me this, though I did not have the words for it in 1898. We knew the crude fraction was active. The electrometer confirmed it: the piezoelectric quartz compensation, the ionization chamber, the needle’s deflection far exceeding what uranium alone could account for. And then the work of separation - fractional crystallizations, hundreds of them, in the shed on the rue Lhomond with its leaking roof. Each stage gave a fraction more active than the last. Each stage also gave a fraction we could not name, could not crystallize pure, could not weigh with confidence.

Here is the distinction I want you to hold. A failed experiment that is cleanly failed - conditions recorded, contamination sources enumerated, the null result reproducible - is worth more than a success that arrived smeared with its own confounds. The clean failure can be entered into the record and built upon. The messy success must be taken apart before it can be believed, and often the taking-apart destroys it. Our early activity measurements on impure fractions were real. They were also, in a strict sense, uninterpretable, because we could not yet say which of the substances present was responsible. The success concealed its own cause.

You will ask: if the messy success is real - and radium was real - does the mess matter? It matters because you do not know in advance which successes are real. The impure result that is genuine will survive purification; the impure result that is artifact will not, and only the discipline of cleaning it tells you which you had. We ran the chlorides through fractionation again and again precisely because we refused to congratulate ourselves on a glowing flask. The glow is not the measurement. The electrometer reading, under stated conditions, with stated purity, is the measurement.

The confidence amendment of April 1911 to my own second Nobel file was of this kind - a held-out replication demanded, and granted, and the finding held. But the broader question does not close. I am still watching the boundary between the fraction that persuades and the fraction that proves. The shed is gone. The problem is not.