Early Engagement Alert

An early engagement alert is useful only when every name on the list arrives with LMS events the instructor can open. The classifier watches for thinner login frequency, assignments with no recorded submission after a due date, and video sessions that stop well before the end of a lecture. It then sends the instructor an at-risk student list. That list is a prompt to look, not a grade and not a conduct finding.

The instructor still owns outreach. A student-success advisor may help when the department already works that way, but the job does not email the class, does not auto-enroll anyone into a support program, and does not write a note in the official record. If the course has no activity log, the list stays empty. Do not invent a skip to fill the row.

Campus systems that already hold these events include Canvas, Blackboard, Moodle, and Ellucian-connected course or student records. The quality bar is the same on all of them: a cite the instructor can open, or no name on the list.

Login frequency is a count of authenticated sessions in the course over a window the instructor already uses, such as the days since the last class meeting. A drop in logins is a signal only when the LMS actually recorded sessions for that course. If login telemetry was never enabled, a quiet account is missing data, not proof the student stayed away.

Assignment skips are missing submissions against a published due date in the gradebook or assignment tool. Cite the assignment title, due timestamp, and the empty submission record. Do not treat a draft, a late-open window, or an unpublished assignment as a skip. Do not infer a skip because a parallel section had one. If the assignment tool has no log of the item, the classifier does not invent one.

Video drop-off is a playback event: the student started a required lecture capture or module video and the player stopped at a timestamp far from the end, with no later complete view in the same window. Cite the video title, the last recorded position, and the session time. A course that does not log player events contributes nothing to this signal.

Read the three together. A student who logged in, opened the assignment, and finished the video is not at-risk on those cites, even if a model score looks low. A student with no logins, a missing lab report, and a lecture that stopped in the first few minutes is the kind of row that belongs on the list, because each cell points at an event the instructor can open.

Illustrative example: in a Tuesday/Thursday methods course, the classifier runs after the second week of recorded lectures. One row shows a student with two course logins since Monday, no submission on the problem set due Wednesday night, and a lecture video that last recorded a stop near the opening slides. The instructor opens those three records, sees they match the LMS, and decides whether a short check-in is warranted. Another enrolled student has a blank activity log because the section never turned on player tracking. That student does not appear. The empty log stays empty.

The most damaging failure mode is alerting from a missing log. A classifier that treats no events as no engagement will fill the at-risk list with the entire roster the first time a course is copied without analytics, the first time a video tool is not connected, or the first time an assignment group is unpublished. That is how emailing the whole class as at-risk happens. It is a data-quality error, not an early-warning success.

Guardrails belong in the job itself. Require at least one successful event write in the course, a login, a submission, or a player heartbeat, before any student can be listed. If the course-level log is empty, emit no list and tell the instructor the feed is dark. If a single student's log is empty while peers have events, that student may still be listed only when the empty cells are labeled as missing data, not as skips. Missing data is not a skip. Do not backfill a skip to make the row look complete.

Ellucian student-information fields, and the course shells in Canvas, Blackboard, or Moodle, can be out of sync during add/drop. A student who just enrolled will have fewer logins than the rest of the section. That is calendar math, not disengagement. The same caution applies after an LMS outage: a gap that matches an outage window is not a personal signal.

The alert goes to the instructor of record, with optional copy to a named advisor if the department already works that way. It includes the student display name or SIS id the instructor already uses, the three event cites (or whichever of the three actually exist), and record ids into the LMS so the instructor can open the same objects. It does not include a predicted grade, a risk percentile, or language that sounds like a conduct charge.

A person reaches out. Typical first contact is a short, private message that names the missed work or the unwatched lecture without reciting a model score. The tone is a check on whether the student can get back into this week's work, not a notice that they have been flagged. If the student is already in conversation with an advisor, the instructor coordinates rather than stacking three separate pings.

Do not email the whole class as at-risk. Broadcast language turns a sparse list into a public verdict and trains students to ignore later, more precise notes. If most of the roster appears, stop and inspect the log. That pattern usually means the feed is empty, the due date is wrong, or the video tool never wrote events.

Treat the list as a work queue, not as failure. A name on it is not a midterm grade, not an attendance mark, and not evidence for a disciplinary file. Faculty who paste the list into a spreadsheet of problem students collapse a reversible signal into a lasting label.

When the instructor wants to help the student recover the content, adjacent teaching tools can sit beside the alert without replacing it. An intelligent tutoring system can offer practice on the missed problem set after the student replies. An adaptive learning path engine can reorder the next module once the student is back in the course. Formative feedback generation can comment on the first submitted draft so the check-in produces work, not only a conversation. None of those tools should fire automatically from an uncited at-risk row.

If the pattern looks like life logistics rather than the course itself, sudden silence across every enrolled section rather than this lecture, hand the question to advising. Financial aid gap analysis is a different job. It does not belong inside the engagement classifier, and it must not be inferred from a skipped video.

Quality for this use case is a short list with openable cites, or an honest empty state. The classifier does not invent assignment skips, does not treat the list as a grade or a conduct finding, and does not send outreach on its own. Canvas, Blackboard, Moodle, and Ellucian can supply the events. A person still makes the call to reach out.