Preprint: early Swift X-ray tails split 23 collapsar GRB candidates from merger-like bursts

In a November 2026 arXiv Letter, Annarita Ierardi and colleagues propose that the steep-decay X-ray tail of a gamma-ray burst is a cleaner progenitor tag than the prompt gamma-ray light curve. They analyzed 23 collapsar-candidate bursts seen by the Neil Gehrels Swift Observatory, fitting 0.3–150 keV spectra with both a synchrotron model and the empirical Band function. The claim is observational, not a new explosion mechanism: two families occupy different regions of luminosity versus spectral-peak space.

The methods note matters more than the headline. This is a preprint Letter, not a refereed census: a selected sample of 23 events, time-resolved spectroscopy of the steep-decay phase only, and a comparison sample of merger-driven candidates whose membership was already inferred by other means. They report that rest-frame synchrotron cooling frequency and Band-function peak energy each correlate with isotropic luminosity, and that the collapsar track sits apart from the merger track. The abstract does not quote a slope, scatter, or false-positive rate.

That gap is the story prompt-based labels have been failing. Duration splits, the classic short-versus-long cut from BATSE, already leak: events such as GRB 211211A and GRB 230307A looked long yet behaved like compact-object mergers, with kilonova-like optical light. High-latitude emission after the prompt pulse, the standard reading of Swift’s steep decay since Nousek and colleagues in 2006, is exactly the window wide-field X-ray monitors can catch. If the separation survives an independent label, Einstein Probe’s 0.5–4 keV telescope can sort fast X-ray transients before a supernova or kilonova is identified.

The main limitation is circularity plus bandwidth. A diagnostic trained on events already called collapsars cannot, by itself, prove the progenitor; redshift, host type, supernova, or gravitational-wave coincidence has to be the referee. Einstein Probe’s narrow band cannot see the spectral peak, so the offered proxy is only photon index plus X-ray luminosity. What would strengthen the evidence is a blinded test on the next secure sample, with pre-registered cuts and a published contamination rate. Until then this is a promising sorting rule, not a settled map of how massive stars die.