Before Sputnik · source record · preprint

Independent Recovery of Vanishing Sources on POSS-I Photographic Plates Using Automated Source Detection and Cross-Epoch Matching

Hayes, Z. · 2026

AI-verified citationSRC-HAYES-2026

identifier / locator: arXiv: 2604.04810 (April 2026)
url: https://arxiv.org/abs/2604.04810
verification note: Full text read via arXiv 2604.04810 (2026-08-22 batch).
reliability notes: The only fully independent source-extraction replication: new pipeline, no reference to existing transient catalogs, benchmarked with synthetic injections.
Independent researcher; preprint. The full-sky sweep used preliminary (looser) detection parameters — the author flags this limitation explicitly.

Evidence drawn from this source

Supports· strongTRN-E008

what the source states: A from-scratch detection pipeline (sigma-clipped detection, PSF filtering, local astrometric refinement, cross-epoch matching against POSS-I Blue and POSS-II Red, Pan-STARRS DR1 rejection) recovers 8 of 9 April 1950 sources and 3 of 3 July 1952 sources, with ~0.2 false positives per 10-arcmin control field; a full-footprint sweep yields 2.85 million candidates, of which 3,450 of Solano's 5,399 published candidates (63.9%) are matched at a median separation of 0.94 arcsec. None of the unrecovered Solano sources within the footprint has a Pan-STARRS DR1 counterpart within 3 arcsec.
what we infer: The canonical detections do not depend on the VASCO team's private software. The 500-fold gap between his 2.85-million-candidate sweep and the vetted 5,399 also demonstrates how completely the "population" depends on extraction and vetting choices.
limitations: Independent extraction, not independent material: the pipeline reads the same digitized DSS descendants of the same photographic copies.
The full-sky sweep ran with preliminary (looser) parameters; the author flags that a calibrated reprocessing should yield a substantially smaller catalog.

Undermines· strongTRN-E009

what the source states: Applying nuclear-test temporal windows to the independently extracted catalog over 368 POSS-I observation nights gives a post-test calendar-day relative risk of 1.35 ([+1] window) that is not statistically significant (95% CI 0.91–2.00, Fisher p = .17) and is schedule-sensitive; a negative binomial model of nightly counts with nightly patch coverage as exposure is null (IRR = 1.03, 95% CI 0.89–1.18, p = .71), as are all asymmetric windows tested.
what we infer: The single most probative negative result for the nuclear correlation, because it varies the disputed step (source extraction) and models the observing schedule — the two main confound channels. A correlation that replicates on the original rows but not in an independently built catalog looks like a property of the catalog.
limitations: The independent catalog was built with preliminary parameters and averages ~7,250 candidates per night — mostly noise — so a real but weak effect could be diluted below detectability.
Preprint; single author.