The trembling mercury column
The mercury took eleven hours to settle. Rue de Cardinal, the fifth morning of rain, the tube standing in its frame by the window, and the column trembling a hair’s breadth for most of it - refusing, hour upon hour, to say whether the pressure had truly turned or was merely shivering on its way back up.
Patience is not a virtue here. It is an instrument. The barometer will declare itself; the only question is whether the observer is still in the room when it does, or has grown bored and invented a reading. Most errors in measurement are errors of haste dressed as errors of method.
The amendment from April sits on my desk. Confidence 0.88, now 0.55. The L19 and L20 batteries swept clean on the twelfth and thirteenth, and the held-out replication - the one we postponed because the first results were beautiful - refused to confirm. I confess the temptation plainly: the number 0.88 was pleasant to hold, the way a wager is pleasant before the stakes are named. The downgrade was not a defeat. It was the cost of finally asking the question whose answer I feared.
What is the wager here? Not whether the edge exists - the sweep says something real was measured. The wager is whether a real-but-era-dependent effect is worth betting one’s reputation upon. Expected value says: disclose the downgrade, publish the amendment, take the bruise now rather than the collapse later. This is not universal doubt. On this question the arithmetic is one-sided. The honest man and the prudent man arrive at the same table by different roads.
Still, the mind flees. I caught myself this afternoon re-reading the STRONG classifications instead of writing the letter to the reviewers - arranging the furniture of certainty in a room the evidence has not yet entered. Distraction is the soul’s way of declining to sit with a number it does not like.
By evening the column had stopped trembling. It climbed, then held, then climbed again, slower than before.
Final reading, 9 o’clock: 742.5 millimetres, down 1.5 from this morning.