Lighthouse

Lighthouse watches the internet for what AI systems do there and what their activity leaves behind, and publishes what it finds, with the evidence and its limits.

Studies LH010

Do pinned dependents move faster before the advisory to fixes whose notes announce a security flaw than to silent ones, when each class has enough libraries that no one or two carry the result?

The answer so far. The study does not settle the question. Before the advisory, pinned dependents moved to announced fixes at a higher pooled rate than to silent ones under every clock and every restriction tried, but with the library as the unit the brief's classes cannot be told apart under any of them (only the strict class under a clock chosen after the fact reaches p < 0.05; "How many comparisons" below), and under the brief's own clock the pooled lead's uncertainty spans no difference. The top 500 projects gave 43 libraries with a qualifying event, not the 40 or more the design needed: 22 announced and 21 silent, of which 18 announced and 12 silent had pinned dependents to follow (observation; brief.md amendment 1, data/summary_by_class.csv). Their 209 repository and event pairs gave 142 moves by 27 September. Version 0.3 reanalysed the retained evidence after every outcome was known (outcome-aware and post hoc, not preregistered; Findings, "Reanalysis of retained evidence, version 0.3"); the first three bullets are version 0.2's figures, unchanged, with one interval added.

Study, version 0.3, 27 September 2026 studies/LH010/LH010.md

From studies/LH010/ in the repository, last changed 9 November 2026.