The SPHERE (spectro-photometric exoplanet research) extreme-AO planet hunter saw first light at the VLT observatory on Mount Paranal in May 2014 after ten years of development. Great efforts were put into modelling its performance, particularly in terms of achievable contrast, and to budgeting instrumental features such as wave front errors and optical transmission to each of the instrument's three focal planes, the near infrared dual imaging camera IRDIS, the near infrared integral field spectrograph IFS and the visible polarimetric camera ZIMPOL. In this paper we aim at comparing predicted performance with measured performance. In addition to comparing on-sky contrast curves and calibrated transmission measurements, we also compare the PSD-based wave front error budget with in-situ wave front maps obtained thanks to a Zernike phase mask, ZELDA, implemented in the infrared coronagraph wheel. One of the most critical elements of the SPHERE system is its high-order deformable mirror, a prototype 40x40 actuator piezo stack design developed in parallel with the instrument itself. The development was a success, as witnessed by the instrument performance, in spite of some bad surprises discovered on the way. The devastating effects of operating without taking properly into account the loss of several actuators and the thermally and temporally induced variations in the DM shape will be analysed, and the actions taken to mitigate these defects through the introduction of specially designed Lyot stops and activation of one of the mirrors in the optical train will be described.