Qualifies· weakTRN-E019
Sinkkoinen 2026: open Bayesian reanalysis reports the association on occurrence, not count
- what the source states
- A public Bayesian hierarchical reanalysis (Stan models and R scripts, CC BY 4.0) of the nuclear-test/UAP/transient correlations models transient counts with latent random-walk temporal autocorrelation and excludes plate-edge artifacts. It reports nuclear tests at one-day lag p = 0.037, UAP reports at two-day lag p < 0.001, same-day UAP p = 0.048, and notes the effects appear to act on occurrence probability rather than count magnitude.
- what we infer
- Methodologically the most useful item in this batch and evidentially among the weakest: it reruns the same catalog under a model that handles temporal autocorrelation, which the original chi-square framing did not, and the association survives — but attenuated, and the strongest term is the UAP-report lag rather than the nuclear one. Its real contribution is that the code is public and rerunnable, which no published analysis on either side of this dispute currently offers.
- limitations
- Not peer reviewed and not a paper; a code repository with a README.
- Same underlying catalog as the analyses it extends — a check on the statistics, not independent evidence.
- Results not re-run for this import; reported figures are the author's own.