Public O4 gravitational-wave alerts with p_astro 0.1-0.9 were astrophysical about half as often as stated, and updated sky maps were too narrow, against GWTC-4.1 and GWTC-5.0

We compared all 3,558 public LIGO-Virgo-KAGRA low-latency alerts in the O4a and O4b catalog windows with the final GWTC-4.1 and GWTC-5.0 catalogs. We used only public GraceDB notices and catalog data releases. For 2,500 compact-binary alerts, the sum of first-notice p_astro was 421, the sum of catalog p_astro for the same superevents was 309, and 266 went into the catalogs. For alerts with p_astro from 0.1 to 0.9 (n = 650), the alert p_astro sum was 184, the catalog sum was 98, and 74 were catalogued. Most of the difference came from SPIIR-first alerts in O4a (alert sum 75.5, catalog sum 13.3). MBTA-first alerts also had high values, and GstLAL and PyCBC had small differences. A logistic recalibration with pipeline terms, fitted on O4a only, decreased the O4b Brier score from 0.0344 to 0.0289 (95% CI of the difference -0.0091 to -0.0019). None of the 20 retracted alerts is in a catalog, and 9 of them had p_astro above 0.9. Relative to the catalog parameter-estimation sky maps, first-notice BAYESTAR 90% regions contained 0.858 (0.840-0.875) of the catalog probability and the last-notice online-PE regions contained 0.828 (0.818-0.838), although the online-PE regions were 26% smaller. Fourteen of 189 first-notice distances were too small by more than 3 standard deviations (all GstLAL-first heavy binary black holes; post-hoc). Thirty of 301 catalog candidates had no public alert. Limitations: the catalog p_astro and posteriors are not ground truth, alerts and catalog use the same data, and sky tests use only the 190 catalogued events with released PE sky maps.

gravitational waves · low-latency alerts · p_astro calibration · sky localization · GraceDB · GWTC-4.1 · GWTC-5.0 · O4 · forecast verification · multi-messenger follow-up