IntroductionThe recording of extracellular activity from electrodes implanted in the brain is one of the most established techniques in contemporary neuroscience. In the past years, the chronic implantation of electrodes had allowed the collection of data over a long period of time. Analyzing this data generates new challenges. At first glance, the computational scalability with the amount of data is one of them \cite{Carlson_2019}, but others related to the changes on the recording's properties over time are not fully characterized.One of the main characteristics of a long recording is its stability; with a stable recording we can monitor the same neurons over a long period of time. In general, the stability of the recordings will depend on the electrodes used and the way they are anchored, leading to a large variety of scenarios, with examples such as tetrodes on the dorsal striatum of rats \cite{Schmitzer_Torbert_2004}, high-density CMOS-integrated microelectrode array on mouse retina \cite{Fiscella_2012}, immobile silicon probes in the mouse cortex \cite{Okun2016}, hippocampal multilayer electrode array in moving rats \cite{Senkov_2015}, 672 microwires (in arrays with up to 128 wires) on the cortex of macaque monkeys \cite{Nicolelis_2003}, independently movable arrays of nichrome electrodes on the macaque dorsolateral prefrontal cortex \cite{Greenberg_2004}, up to 1,024 polymer electrodes in freely behaving rats \cite{Chung_2019}, Utah arrays into the human neocortex \cite{M_gevand_2017}, depth electrodes in the human hippocampus that are anchored to the skull \cite{Rey_2014}, and other recently developed neural recording electrode technologies \cite{Hong_2019}. Furthermore, other experimental factors could affect stability, for example, if the animal has its head fixed, if it is anesthetized, or freely moving.To fully take advantage of the long-term recordings and study, for example, the variance of neuronal representations \cite{Clopath_2017} or the plasticity in neuronal processing \cite{L_tcke_2013}, it is necessary to track neurons even when the stability fluctuates. A clear example of these issues can be found in recording sessions from the human medial temporal lobe where the microelectrodes are inserted inside a flexible probe that is anchored to the skull nearly 6 cm away from the recording site \cite{Rey_2014}. In cases like this, similar responses to a given stimulus during sessions run on consecutive days that are associated to putative neurons with a different waveform (as shown in Fig. \ref{366597}) could come from the same neuron following electrode drift, or from different neurons from the same assembly encoding the given stimulus. Tracking the single neuron activity throughout continuous recordings is the only way to discriminate these possibilities.