Many of the problems with traditional reporting techniques stem from the difficulty of presenting several different lines of investigation in natural language documents. The different persepctives offered by human factors, software engineering, metallurgical, meteorological accounts make it different to form a coherent view of the events leading to major failure s. This, in turn, leads to significant ommissions and inconsistencies between the different strands of analysis within many reports.
In this talk, I will use a range of formal methods to identify errors in acciden t reports. A first order temporal logic will be used to construct a time-line for the events leading to a `failure'. This model will then be used to formally pr ove that the conclusions in an accident report are actually consistent with the events described in the report. If such a proof cannot be constructed then eit her there is an error in the formalisation OR there is an error in the accident repo rt itself.
If time allows, I will also show how formal refinement techniques can be used to identify political/organisational bias within accident reports.
I will begin by presenting a series of examples drawn from reports produced by the US Coast Guard, the Air Accident Investigation Branch and US Presidential Boards of Inquiry. These examples will then be used to illustrate a technique for formally identifying bias in documents. This technique not only identifies missing information between two accounts, it also provides means of reasoning about the inferences that are lost when critical information is ommitted.
Although the focus will be on accident reports, many of the same arguments could be made about requirements documents in both software engineering and interface design. Partial information and individual biases have a profound impact upon the inferences that a designer can make about a potential systems. By hiding information about design requirements, it is possible to bias key development decisions.
Finally, I will explain why Steve Draper and Phil Gray had an accident on the way to the GIST meeting...