During a 2-week period in July 2012, JET produced over 120 almost identical H-mode pulses, providing two orders of magnitude more data for a single pulse type than has previously been available. A systematic statistical analysis of this data revealed the unexpected observation of a series of maxima and minima in the probability density function (PDF) for the waiting times between ELMs. A search for the cause has produced an unprecedentedly detailed picture of the plasma's behaviour. Between ELMs the plasma cross-section calculated by EFIT moves slowly both radially outwards and vertically downwards towards the divertor, before an ELM restores its position. An oscillation with a period of approximately 8 milliseconds (125Hz frequency) is also found, that narrows the plasma as moves it downwards. The oscillations calculated by EFIT coincide with enhanced (or reduced) Berylium II (527nm) light emissions, and with an enhanced (or reduced) likelihood of an ELM dependent on the phase of the oscillation. ELMs tend to occur at a maximum in both the plasma's downward velocity and position. Previously no evidence was found for the vertical control system (VCS) being the primary cause of the observations. Here we find that the plasma velocity signal used by the VCS shows no strong indication of the vertical plasma oscillations seen in EFIT reconstructions, that evidently coincide with observations of Berylium II light emissions and ELM occurrence. The cause of the oscillations could therefore be plasma instability, or simply a small vertical oscillation that the VCS is not reacting to prevent. Whatever their cause, the observations suggest new opportunities for improved ELM mitigation and plasma control, whether by manipulating the naturally occurring oscillations to achieve ELM pacing, or for improving plasma stability.