In MDL-65829 we attempted to discard enrolments whose start date is after the course start so that the new enrolments added halfway through the course would not get incorrect predictions. The analytics API will identify a new sample and will analyse it the first time the cron runs. They may get false negatives (incorrect at-risk predictions) because the indicators will be calculated since the start of the course (for example) and they would not have much activity logged up to that point in time. I suggest to require user enrolments to span through most of the analysis interval (e.g. a 70% of the interval).

Example of what would happen with this issue's patch in in the following scenario: Course starts on the 1st June, we analyse the course on the 10th June with an analysis interval from the 1st June to the 10th June.

1. New enrolments 5 mins after the course start -> All the enrolments will be analysed
2. New enrolment on the 2nd June -> It will be analysed
3. New enrolment on the 7th June -> It will not be analysed

• For analysis intervals with fixed start and end dates like "Past quarter", where the start and end dates are fixed and depend on the course start and end date, the student in case #3 will not get a prediction until the predictions for the next quarter of the course are generated.
• For analysis intervals with variable start and end dates like the ones added in MDL-65588, where the end date is now, the student in case #3 will get a prediction on the 24th June, that is 17 days after the enrolment (17/24 = 0.7083, which is higher than the 70% in the example above)