How to interpret failed proofs in Event-B

In formal reasoning, modelling and proving activities are closely related. Models give rise to different proof obligations and information about failed proofs gives indications on how models should be improved. This document is an attempt to address the latter issue: to understand how to deal with unprovable obligations. We consider here proof obligations related to invariant preservation of an Event-B model: firstly, to understand the meaning of the proof obligations; secondly, to analyse various ways to fix the model accordingly. Our analysis is based on the concept of reachable states and inductive invariants.