I must say I am pretty dissapointed by the fact that TA4 still has a lot of very basic, obvious bugs and suboptimalities which I would have thought were clearly apparent in TA 3 and could surely have been corrected. For example:
(1) $a=0; $b=$a; sets $b = -0
(2) $c=frac(-2,-1) sets $c = --2
(3) The LaTeX to TA converted still does not recognise the Maple-graded question type (including the example in the manual) and chokes on all sorts of legitimate Maple in the code section. (For instance one can not directly use the less than sign.)
(4) It is not possible (except in some circumstances by a hack) to get a single Maple command to set more than one algoritmic variable. I suspect the resulting multiple calls to Maple slow the system down considerably. Correcting this would mean a complicated question would involve a single Maple call rather than (say) 10.
(5) The "equation previewer" produces 3((a+b)+c) if you preview 3*(a+b+c)
(6) Included Maple-generated graphics are very poor - far worse than jpg-exports from Maple itself. (Doubtless this can be fixed by subscribing also to Maplenet but TA is already pretty expensive and ought just to work well out of the box.)
Finally, why is TA so painfully slow? Perhaps there are good reasons for this, but with our setup on a reasonably good server being hardly used it takes about 1-2 seconds per question to import a qustion bank and the same sort of time to generate an instance of a question. I mean, what is it doing in all this time? Your average shopping website runs a whole load of PHP or similar with loads of database accesses and the page appears in almost no time.
I was really hoping that TA4 would be a substantial rewrite with this sort of thing fixed and some enhancement of the rather grungy programming capabilities. In fact, the only solid imprrovment I have spotted is the more flexible import of rosters. (At most a few hours coding work, I would have thought.)