Users often had trouble getting their formulae parsed, and spent a large proportion of their time rearranging terms in their formulae.
When the system was unable to parse a formula, the system tried to indicate to the user the ``best'' parse it had achieved to that point. The ``best'' parse was the reduction found by the parser which had the smallest number of nodes. Unfortunately, this did not always correspond to anything very intuitive, and usually was not very useful at all. The number of users who actually saw the parser terminate with this display was very low due to the fact that when given an erroneous formula, the parsing took so long that the user elected to cancel the parsing instead of waiting an indeterminable period of time for the result.
As the system processes a user's formula, it shows a display of the current graph that is being worked with by the parser. From this display an experienced user who has some understanding of the system is able to see what is going wrong and correct the formula appropriately. This display, however, was not helpful for a novice user. The system's ability to parse formulae must be improved if this system is to be useful, and feedback when parsing fails must also be provided. Presently, to correct an unparsable formula, a user is likely to get stuck in a lengthy ``formula debugging'' loop, tweaking their formula without any real knowledge or direction, until it works.
A useful display for failed formula processing is something that the system currently lacks, and definitely needs to be included. The system should guide the user in some way when it is unable to parse their formula. Perhaps the system could indicate to the user what went wrong, for example indicating missing limits on integrals. However, to automatically provide guidance in correcting formulae may turn out to be impossible. This is discussed further in Section 6.2.8.