Human compensations for undependable systems

Randell\u27s (2000) dependability fault-error-failure model was originally designed with the objective of describing the propagation of faults in technical systems. Conversely, Reason\u27s (1990) swiss cheese model was intended to describe the organisational facet of systems\u27 failures. However useful these two views have been, there has not been a lot of effort devoted to highlighting their common features. Moreover, these two models say little about the positive human contribution to the delivery of an acceptable service with undependable systems. The investigation of these two aspects forms the main focus of this paper. Our first objective will therefore be to integrate the two models. In doing so, we will also provide an answer to the problem of scale in the description of events in complex settings: organisational factors and pure technical causes could be integrated in the same descriptive picture. Our second objective will be to show that the dependability of the service of socio-technical systems is often a matter of human compensations for poorly designed systems. This claim will be supported by three concrete examples where human compensations have permitted a partly-automated system to deliver an acceptable service