We propose two alternatives for recovering from the impossibility to handle correct behaviour.
Critique
An alternative for the critique step is to better tune the Obs-mapping component to the Vocabulary component initial-fault-nodes. In general, the combination of abd-mapping and initial-fault-nodes is not an obvious choice, because using the initial-fault-nodes assumes that only abnormal behaviour is observed. However, the given problem contains also normal behaviour light(yes).
A more obvious choice of Obs-mapping component can be determined by
checking whether we have observed both normal and abnormal
behaviour. This is a case specific repair action, because we use the
current observed behaviour in the choice
of Obs-mapping. The abnormality or normality of the observed
behaviour is checked using the abnormality-mapping Obsmap component.
If execution of abnormality-mapping results in a non-empty set of
then we use the knowledge that the combination of
initial-fault-nodes and abd-mapping is a bad combination, and
abnormality-mapping is probably a better one.
Modify
The previous critique step results in the
same method as the ``
Modify'' step, namely term (11).
Critique
The other alternative for ``
Critique & Modify'' is
to adapt the Vocabulary component.
If the Obs-mapping component abnormality-mapping does not
result in an empty 
set, then a better choice of Vocabulary
is possibly all-initial-nodes. This vocabulary
contains all initial causes (including correct states) and the incompleteness
assumptions, and is therefore better tuned to Obs-mapping=abd-mapping .
Modify
We would now come to another method then before, namely:
Knowledge-verification
The knowledge verification still satisfies, as before.
Simulation-verification
Performing diagnosis results in the following diagnosis part
This yields 3 x 5 = 15 diagnoses, so we have too many possible diagnoses. We end up with these other diagnoses because the critique and modify steps are based on the observation that the vocabulary was too small, whereas before the devision of the observations was considered as wrong. After verification we establish that ``too many solutions'' are computed. A repair action for solving ``too many diagnoses'' is needed. We do not describe this trace further.
