Withdrawn claims
Divergence under nominally identical repeats
Section titled “Divergence under nominally identical repeats”What was observed. In an early measurement, a set of repeats that were expected to be byte-identical were not. A small number of trials differed, and the differing trials fell entirely within one of the two conditions.
What it was taken to mean. That the instrument itself behaved differently across the two conditions — which, if true, would have undermined every comparison made with it.
What happened next. A rerun on the current version of the instrument did not reproduce it. Cells with replicates agreed exactly, including in the condition where the divergence had appeared.
What was ruled out since. Several explanations were tested and failed: non-determinism within a single session, reloading the model between runs, and the sampling temperature. Each was checked by running the relevant comparison rather than by argument, and none accounted for the original observation.
What is not known. Why the original measurement diverged. The runtime logs for the evening in question were not retained, which is itself a process failure and has since been corrected: the instrument now records a fingerprint of its execution environment alongside every run.
Current position. The claim is withdrawn. Nothing is asserted about asymmetry of the instrument between conditions. The asymmetry that was once read into the original numbers is adequately explained by one condition sitting closer to the ceiling than the other, which requires no additional mechanism.
Why this page exists at all
Section titled “Why this page exists at all”The original observation was striking, and both external descriptions of this project suggested publishing it. It would have been the most quotable thing the programme had produced.
It did not survive a rerun. Publishing it with a hedge — “preliminary”, “suggestive” — would have been the easy path, and the hedge would have been dropped by everyone who cited it.