IceCube track alerts from November 2022 to November 2026 do not replicate the reported association of 200 TeV neutrinos with VLBI-bright radio blazars (14 events, fixed P23 method, p = 0.97)

Plavin et al. (2020, 2023) reported that IceCube track events with energy of 200 TeV or more occur near blazars with high 8 GHz VLBI flux (post-trial p = 3e-4 for 71 events from 2009 to 2022). We did an out-of-sample test with the same statistic, the same fixed systematic error (Delta = 0.45 deg) and a blazar catalogue (RFC 2022b) frozen before all test events. The analysis plan was written before the test. First, we reproduced the method: our code finds exactly the 45 published blazar-event pairs, and gives a post-trial p of 1.0e-3 for the 71 events. Then we parsed all GCN circulars and AMON notices after the last P23 collection date (2022-11-01). Fourteen Gold/Bronze tracks pass the P23 cuts (E >= 200 TeV, 90% area < 10 deg^2). These events have only 3 coincident blazars (7.2 expected by RA scrambles). The geometric-mean VLBI flux of the matched blazars is 0.206 Jy against a scramble median of 0.311 Jy, which gives a one-sided p of 0.97. No new event matches any blazar with S8 >= 1 Jy or any blazar named by the authors. Secondary variants (2D error convention, RFC 2026c, no energy cut, Gold only, all post-2020 events, Delta scan 0 to 1 deg) are also null (p = 0.30 to 0.97). The test has low power: a signal-injection model calibrated on the in-sample result gives a 30 to 41% chance of p < 0.05. However, a CLs comparison excludes signal fractions of 0.25 to 0.30 or more at 95%, which excludes the in-sample-calibrated effect for blazar weights S8 and S8^1.5 and puts it at the limit for S8^2. The cumulative 2009-2026 sample stays significant only because of the original 71 events. Limitations: small sample, model-dependent exclusion, a downward fluctuation (P = 0.028 under the null), a changed IceCube reconstruction with smaller error regions after 2024, and our scramble p-values are 2.5 to 3 times larger than the published ones although the pair sets agree.

IceCube · high-energy neutrinos · blazars · VLBI · Radio Fundamental Catalogue · out-of-sample test · replication · GCN circulars · multi-messenger · null result · statistical power