This contribution to the TADEUS series focuses on the features with which a diagnosis algorithm must comply in order to be applicable as part of a monitoring system running on a power plant, where robustness, ergonomics for the final user, and calculation speed play an important role. Some third-party approaches reported to be applied in industry are reviewed. These approaches rely mainly on the use of a simulator. Taking into account the drawbacks of these methods, a more advantageous algorithm is proposed and illustrated through its application to the TADEUS problem.The proposed algorithm starts from the two plant data sets to be compared, but is reworked to a coherent representation of the plant graph. As diagnosis can be understood informally but abstractly as "sharing the difference in an indicator among the various differences in the degrees of freedom", both known in advance, the algorithm makes use of all the implicit constraints of the graph to obtain a relationship between dependent and independent variables that in turn yield the sensitivity of the indicator to the degrees of freedom.In short, compared to conventional methods, the need for a fine-tuned model is removed, the solution is achieved without the need of time-expensive iterative processes, and it is exhaustive in the sense that every malfunction is taken into account.