Repository navigation
0.1.0-candidate: The recomputed split likelihood-ratio interval is not a confidence sequence
Pre-release
Unrefereed research candidate. In one sentence: a statistics method that is valid if you look once silently fails if you keep checking it as data arrive.
Pace, Salvan and Sartori (Canadian Journal of Statistics, 2023, Section 3) conjectured that universal-inference split likelihood-ratio intervals, recomputed after every new pair of observations with the same threshold, keep asymptotic persistence at least 1 − ε. This release proves the conjecture false (asymptotic persistence 0 in any model), gives the exact iterated-logarithm constants ((√5 − 1)/2; c(r); 1/K under cross-fitting), a sharp Gaussian repair and a model-free checkpoint repair, and quantifies finite-horizon severity: the error first exceeds nominal at about 100× the first look at 80%, about 10⁴× at 95% and about 10⁹× at 99%.
- DOI: https://doi.org/10.5281/zenodo.23856728 (concept DOI 10.5281/zenodo.23856727)
- Release page: https://evidencepress.org/releases/split-likelihood-persistence/
- Assets:
split-likelihood-persistence-0.1.0-candidate.zip(complete package),split-likelihood-persistence-0.1.0-candidate.pdf(manuscript, 15 pages),RELEASE-SHA256SUMS.
Assurance: written proofs; six certificate scripts with negative controls, fully replayed under Python -O (outputs identical); internal AI-assisted review only, not peer review, independent reproduction or formal verification. Author: Anonymous. Text, data and diagrams CC0-1.0; code MIT.