Finally, we give an alternative for the critique and modify steps that tries to recover from ``too many diagnoses''.
In analysing the failure of the verification step, the system uses the knowledge that if the number of observations four or less and there are too many diagnoses, then the repair action becomes ``ask to the user the relevant observables for the computed diagnosis''. This repair action changes the input problem (since additional observations are requested). In contrast, the previous repair actions only changed the method.
New observables need to be asked from the user. The relevant observables are those which are connected to a cause of the computed set of diagnoses, but that are not already part of the observed behaviour. Our set of observables for asking to the user is therefore based on the causes: short-circuit, battery-age, battery-water-lack and alternator-belt.
In our problem the following observables are asked:
fuses, distributorstatus, acceleratorresponse, fuel, batterypower.
The user gives only a value for distributor-status, namely wet. The new observation theory contains therefore:
The diagnostic problem has now been adapted by adding new information.
The knowledge verification still satisfies as before.
We perform diagnosis using the adapted problem and the method of term (11). This results in just one diagnosis:
We continue with the test of the dynamic goals, namely max-number-diagnose(2), which succeeds. The diagnosis problem is now solved. Notice that we end up with another diagnosis then in the previous scenario, where it was. short-circuit). This is because we recover from ``too many solutions'' by asking new observables, whereas in the first scenario we made the Selection component stronger.