Currently Pursuit's confidence values are the internal scores used for the algorithm this presents two problems:
Initial confidence values start out very low (e.g. around 2%) as this score is also used to determine if something should enter the official lexicon rather than a more competitive learning stage. A more better representation of the internal confidence of known patterns would be the ratio between the specific pattern's score and the maximum score of all potential patterns for that concept.
We don't currently have an easy-to-extend method to determine which of the exact patterns was the one matched, I only easily have access to the global concept. For now the return value will simply be the maximum internal score for the concept.
Currently Pursuit's confidence values are the internal scores used for the algorithm this presents two problems: