Seeds with a slight dimple

The counting frame is the least trustworthy instrument in the garden. Not the frame itself - pine, wire, forty seeds to a row, that part is honest. What I mean is the act of counting itself, the moment the seed is lifted and a judgment is made: round, or wrinkled. Most seeds announce themselves plainly. But there is a border population, seeds with a slight dimple at the chalaza, seeds that dry smooth but wrinkle after a week in the drawer. In the 1858 harvest of the F2 for seed form, I recorded 5,474 round against 1,850 wrinkled - 2.96 to 1. The closeness of that number to 3 has comforted me for years, and now, walking the rows again, I find the comfort suspect.

Precision and accuracy are not the same virtue. My counts are precise - the arithmetic is exact, the totals large, 19,959 plants across the seven characters. But precision built on a classification rule that leans on the classifier’s eye is precision about a category the classifier helped draw. The assumption I trust least is this: that the trait is cleanly discontinuous, that every seed falls without ambiguity into one of two forms. If the border cases were assigned by a rule favoring the parental expectation - even unconsciously, even one seed in twenty - the ratio would sharpen toward 3:1 without the underlying biology deserving it. The data would be too good. A colleague with a sharp eye for such things might someday take my exactness as evidence of exactly this.

The remedy is not to abandon the count but to test the count’s stability. The F1 shows nothing; all smooth. The structure appears only in F2, and only at adequate n - a hundred seeds tell nothing, ten thousand begin to tell the truth, provided the ten thousand were classified before the expectation was formed. I have begun re-scoring the 1859 wrinkled series from the reserve seed packets, blind to the previous tally, and recording the second judgment in a separate column. Where the two columns agree, the category is real. Where they diverge, the divergence is the datum.

The recount of the first hundred seeds, begun this morning in the refectory window light, agrees with the original tally on 96.

96 of 100.