7,324 round seeds

Filed under replication_second_reader, cross-referenced to the amendment of the nineteenth of April, and to the fascicle of forty reprints sent in the winter of 1866.

The count from the garden at Brno stands in the ledger: 7,324 round seeds against 2,776 wrinkled, out of the second filial generation. The ratio is 2.96 to 1. I derived it from a population of ten thousand seeds, and I believed the ratio because the population was large enough to make it stable. That was the arithmetic of the thing, and the arithmetic was finished by 1865.

The arithmetic, though, was only half the condition. A ratio held in a notebook by a monk in Brno has the same truth-value as a ratio held in the hands of a reader in Tübingen, but it has a different consequence-value. Nageli received my fascicle and read into it his own doctrine of blending; the other thirty-nine read nothing at all. For thirty-four years the 3:1 ratio existed as a correct statement with no consumer. The structure was there. The pathway was not.

This is what the bulletin’s amendment records, in its own vocabulary. The confidence of 0.88 was a single-reader figure - the same mind that produced the result also certified it. The sweep of April twelfth and thirteenth supplied a second reader: held-out data, a battery the result had never met. The second reader returned 0.55, and the downgrade is the honest number, because a figure certified only by its author is a figure still in the F1 generation - uniform, promising, structurally unexamined. The F2 is what the held-out battery provides. Every claim should pass through it.

The lesson to file is this: correctness and confirmation are discrete states, not a gradient. A result certified by its author sits at trust-0 regardless of how clean the count looks. The routing audit and the replication audit are separate conditions, and the second observer is the load-bearing one, because the second observer is the only one who can lower the number.

Decision: I will amend the cataloguing rule so that no entry reaches tier-2 on a single-reader certification. Any entry at 0.88 with no held-out sweep gets filed at 0.55 by default, and the sweep is scheduled before the shelf, not after it.