A crucial feature of ARCADE is its ability to predict from an early stage which students are having difficulty, or are being lazy, and which are performing to a suitable standard. The approach used here is to compute an overall mark for each module for each student, right from the beginning of the semester or year. Then as the time goes by, the mark gets more and more accurate and by the end it is the final mark. This approximation is based on two dimensions: which pieces of future work will the student complete, and what the marks for them will be. No attempt is made to predict whether future work will be flagged late or not, as the only point of the extrapolation is to determine which students are in danger of failing and late flags cannot make a student fail.
Predicting whether or not the work which is not yet completed will get missed, is done as follows. If the work is already flagged late, we cautiously assume it will be missed. Likewise if it is excused, but the next session has passed (i.e. it is late, but not flagged as such). Otherwise, we predict that it will be missed if and only if the previous two pieces of work are either missed or predicted to be missed.
Mark prediction is even simpler: if the student has had no pieces of work marked for that module, the mark for each piece of work completed (but not yet marked) or predicted to be completed is assumed to be 70% of the maximum available. If the student has had one piece marked, the prediction is taken as the average of 70% and that mark. Otherwise, the prediction is taken as the student's average for those pieces that are marked for that module.
Although clearly very crude, this extrapolation of marks is remarkably effective in automatically spotting those students in danger of failing. All students start off with a predicted result of 70% (before scaling). Those, for example, who miss the first two deadlines suddenly get a predicted mark of zero. Those who miss one, and do badly in the next one may fall under the pass level, and so on.