R. Durgesh, P. J. Pessi, E. E. O. Ishida, J. Peloton

The ExtragaLactic alErt Pipeline for Hostless AstroNomical Transients (ELEPHANT), has been developed as a framework for filtering hostless candidates, in real time alert systems, and implemented as a filter in the Fink broker. ELEPHANT works on stamps and requires minimal information, thus allowing for fast identification of extragalactic transient events. In this work we evaluate the performance of the ELEPHANT pipeline by systematically analyzing flagged hostless candidates identified between 1 September 2023 and 31 December 2025. Our goal is to quantifying its accuracy and identify dominant sources of contamination. For each flagged candidate we collected additional information from multiple catalogues and archival repositories. We further examined their light-curve evolution and astrometric consistency (coordinate dispersion over time) to refine source classification. Results. Out of 877 flagged events, 67 are confidently confirmed as genuinely hostless candidates, with no detectable host galaxy in either existing catalogues or archival imaging, representing a high-purity sample of intrinsically faint or absent hosts. Additional 51 events are linked to visually identifiable hosts that are entirely absent from both catalogues and ZTF stamps. For the confirmed hostless subset, the inferred upper limits on host-galaxy absolute magnitudes extend well below the luminosity range of typical dwarf galaxies. The pipeline showed an overall accuracy of 0.84, with the majority of the classified flagged events being Type Ia supernovae, and the second most detected class being Type I superluminous supernova. ELEPHANT has been adapted to deal with the Rubin alert stream and has been processing its alerts since February 2026.