In this communication, we start with an overview of the field of learning from incident and accident (LFIA). First, there are classic analyses of disasters (Turner, Vaughan, Hopkins) which have been turned by some of these authors into studies of failure to learn. Second, there are reviews of LFIA with synthetic ambitions and third, empirical studies of how LFIA happens in practice. The increase of research on LFIA in recent years is commented, and some reasons hypothesised. The presence of a normative approach is emphasised, discussed and questioned. We follow this overview of the literature with a research strategy that complement LFIA studies by further contextualising and embedding empirical research of LFIA during daily operations. The purpose is to explain why in our view LFIA is an ideal that in practice remains imperfect because of the operational realities of high-risk systems. The characterisation and acceptable extent of such imperfection remain a thorny issue for safety research.