make sure you can run the code from the beginning to the end; this is the minimum requirement
the ideal case is you provide an R object, say, fit, that other people need to know nothing but simply run predict(fit, newdata) to get the prediction, otherwise you have to make it very clear on how to get the predictions; to test if you can get predictions without troubles, you may extract a subset from the original dataset and try to make predictions on it
please report your error rate to group members (on the test set and training set); it is very important to split the training/test sets randomly for multiple times to get valid estimates of the error rates (don't just split once -- it might be unstable)
fit, that other people need to know nothing but simply runpredict(fit, newdata)to get the prediction, otherwise you have to make it very clear on how to get the predictions; to test if you can get predictions without troubles, you may extract a subset from the original dataset and try to make predictions on it