California, and seemed to be an ordinary supernova already well under way. It was infrequently observed over the next few months, and was eventually classified 4 as a common type of supernova called type II-P in January 2015. Its subsequent failure to decline led Arcavi and colleagues to flag it for special attention.Not only did iPTF14hls not fade, but as months stretched into years, the authors found that its brightness varied by as much as 50% on an irregular timescale, as if it were exploding again and again. Although not among the most luminous supernovae ever seen, iPTF14hls was brighter than an ordinary type II-P explosion and, by lasting so long, it radiated much more energy.Just as the light from iPTF14hls refused to fade, the spectrum of the light refused to evolve. Usually, as a supernova expands, it reveals deeper, more-slowly moving material, and the lines in its spectrum get narrower. But in the case of iPTF14hls, Arcavi et al. found that the light-emitting region maintained the same speed throughout the supernova's lifetime.This result might make sense if the authors had observed the emission from a single shell of ejected matter. However, the radius of such a shell would have increased over time. Radiation theory tells us that if something gets bigger but maintains its luminosity, it should also get cooler. Confounding expectations again, iPTF14hls remained at the same temperature (about 6,000 kelvin), suggesting that the radius of the light-emitting region was roughly constant. What was going on?Supernovae shine for one of four reasons 5 : radioactive decay; radiation released by the shock-heated envelope of a massive star as it expands and cools (ordinary type II-P supernovae); colliding shells that convert kinetic energy into light (type IIn supernovae); or radiation from a central, compact stellar object such as a magnetar. In the case of iPTF14hls, radioactivity can be ruled out because the isotopes that have the correct lifetime to explain the emission are not sufficiently abundant in the explosion. Likewise, radiation from a shock-heated envelope would require an envelope mass and an explosion energy that are incompatible with our understanding of stellar evolution.It therefore falls to a magnetar or colliding shells to explain iPTF14hls. Arcavi and colleagues explore both possibilities and rule out the simplest models. Using standard formulae, which might not be adequate for this event, they conclude that the initial luminosity of a magnetar would be too high to explain their observations. Similarly, the colliding shells that are typical of type IIn supernovae would produce X-ray and radio emission that was not seen for iPTF14hls, and narrower spectral lines than were observed.Having ruled out all of the standard theoretical models, Arcavi et al. tie their hopes to an alternative scenario: a pulsational pairinstability supernova 6 . In this model, during the death of an extremely massive star, violent thermonuclear instabilities in the final stages of nuclear fusion lead to repeated su...