Qualifies· weakTRN-E018
Doherty 2026b: linear alignments among ML-scored candidates on seven plates
- what the source states
- Using machine-learning scoring of transient candidates on POSS-I plates from 1949–1957, the paper reports seven plates carrying alignments of five to eight sources exceeding Monte Carlo expectations (p < 0.03), argues the sources are point-like rather than trailed, and reports angular rates it characterizes as inconsistent with known satellites of that era.
- what we infer
- Extends the aligned-multiple-transient line rather than replicating it: same archive, same candidate family, new statistic. Its weight is limited by exactly what makes it interesting — alignment searches over many plates with a post-hoc-chosen alignment criterion are the classic setting for look-elsewhere inflation, and the p-values reported are modest. Filed as qualifying rather than supporting because it adds a pattern claim without addressing whether the underlying marks are on the original exposures at all, which is where this case's crux sits.
- limitations
- Unreviewed preprint by an author now affiliated with the proponent group (see SRC-DOHERTY-2026 independence note).
- Alignment significance depends on the alignment criterion and the trials-factor accounting; neither is preregistered.
- Inherits the catalog's contested provenance — if candidates include copy-plate defects, alignments among them inherit that problem.