Course Placement Recommendation

The quality bar is a placement recommendation that cites the high school transcript lines and the placement test result it used, names a catalog band, and leaves any missing score field empty. Advising still owns the placement. The model does not enroll anyone.

A registrar or advising lead should open the artifact and see, without a ticket to IT, which sending courses were read, which test instrument and date were read, and whether the model is silent because a required input is not on file. If those three things are not visible, the output is not ready for a busy advising week.

The band is the language your catalog already uses: developmental, college-ready, or accelerated in a subject. It is not a CRN, not a seat, and not a waitlist position. Credit already posted through transfer credit evaluation can change what an advisor will accept in the same term. Placement and transfer credit are related decisions. They are not the same decision, and they must not overwrite each other silently.

Do not attach a success story about lowering mis-placement. This page does not claim a reduction rate. The operational win is fewer students scheduled from an invented score or from an unconfirmed model output.

Start with the official high school record, not a self-reported GPA widget. Pull course titles, level markers (honors, AP, IB, dual enrollment), grades, and the year taken. Each cite should be specific enough that an advisor can find the same line on the scan: sending school, course name, local course code if one exists, term, and grade. If the transcript uses a district code instead of a common title, keep that code. Do not translate it into Algebra II unless your articulation table already says so.

Then read placement scores the institution actually administers or accepts. Cite the instrument, form if stored, test date, and the score exactly as recorded. Do not convert a blank into a typical cut. Do not substitute SAT math, ACT math, or a sibling's score for an exam the student has not sat. Do not treat a classroom diagnostic as the official placement test unless written policy says it replaces that test.

Student-information and admissions suites such as Ellucian, Workday, Slate, and Technolutions often already hold the transcript image or parsed courses and the test row. Canvas may hold quizzes that look like placement because they are math or writing diagnostics inside a course shell. Treat LMS activity as classroom evidence unless policy maps it to the official instrument. Mixing an unproctored Canvas quiz with a transcript-cited placement test, without labeling the source, is how a band gets justified by the wrong artifact.

When a title is ambiguous (Integrated Math 3, Math 4, Senior Math), stop at the cite. The model should not assume the course is Algebra II. An advisor who knows the sending district can interpret. Course-to-outcome meaning belongs closer to a knowledge graph from syllabus than to a letter grade. Use that mapping only if your process already maintains it. Do not invent a receiving-course equivalent from the title string alone.

One walk-through: Maya is admitted for fall. Her high school record lists Algebra II in spring of senior year, grade B, and no precalculus. The math placement test row is blank. English shows AP Language with a score of 3, and an English placement test is on file with an instrument name, date, and score. The math recommendation must cite Algebra II, B, and must leave the test-cited math score empty. It may tell advising that a math band is not ready until she sits the exam, or it may attach a transcript-only provisional note if policy allows that note, labeled as provisional. It must not write a numeric cut that was never recorded. The English side can recommend a band because both the AP line and the placement test are present and cited.

A band is a bucket the catalog and advising handbook already share. Name that bucket. Attach the cites. Then stop. Do not pick the section. Do not reserve a seat. Do not write the student into a waitlist. The quality outcome is the recommendation plus cites, not a schedule.

If policy requires both a transcript threshold and a test threshold, the band is only as complete as the weaker input. A strong AP line does not fill a missing placement test when the department requires both. Fail closed on the test-dependent portion. Empty stays empty.

Write the recommendation into the advising workspace, not into registration. The advisor confirms, edits, or rejects the band and records a reason. Until that confirmation exists, the SIS should not treat the model output as an enrollment instruction. After confirmation, later-term sequencing may feed an adaptive learning path engine. That engine should consume the advisor-confirmed band, not the model's first guess.

Treating the recommendation as registration is a failure mode with a clear signature: the student appears in a section before an advisor note exists, and the only approval is a model timestamp. Reverse that path. The recommendation sits where counselors already work. The advisor writes the placement. Registration follows the advisor write, the same way it follows a manual placement today.

Auto-enroll from a model band puts students in the wrong room with no conversation, and it collides with holds, co-requisites, time conflicts, and cohort restrictions that only the registration office sees. Keep enrollment on the SIS registration path after human confirmation. Ellucian or Workday (or whichever SIS you run) remains the system of record for the seat. Slate or Technolutions may have collected the application and some test files. None of those products should silently mint a placement score or a registration.

Department override rules sit above the model. If mathematics requires a minimum test score for MATH 201 regardless of high school calculus, a transcript-only recommendation into MATH 201 is invalid even when the high school record looks strong. Ignoring that override because the model is confident is a process defect, not a model feature. Show the rule on the artifact: identifier, department, and that the recommendation was blocked or downgraded because the rule could not be satisfied. An advisor can still take the human override path faculty already approved. The model does not invent that path.

Three failure modes belong in training, QA, and the runbook, not in a footnote after go-live.

Placing from a missing score: if the test cell is empty, do not impute. Do not use last year's cohort average. Do not copy a high school GPA into a test field. Do not round a nearby SAT into the placement instrument's scale. The recommendation either omits the test-dependent band or marks that portion empty and routes the student to sit the exam. Quality here is restraint. Inventing a test score so the workflow can complete is a quality failure.

Treating the rec as registration: covered above, and it is worth repeating in staff training because it is the fastest way to create an irreversible first-day problem. A recommendation is evidence for a person. It is not a registration API payload.

Ignoring a department override rule: catalog footnotes, directed self-placement policies, and restricted sequences (nursing math, engineering calculus) are not suggestions. Encode the rule. Cite it. When the rule cannot be evaluated because a score is missing, fail closed. Do not recommend the restricted course on transcript strength alone.

A wrong first-term band shows up later as a graduation requirement gap alerts problem: the student is off the map for a major that assumed college-ready math in term one. Fixing placement quality at enroll is cheaper than chasing gaps in year three. The alert is still not a reason to auto-correct placement without advising.

Keep the vendor set boring on purpose. Read what those systems already stored. Cite it. Recommend a band. Leave empty fields empty until a person and a real test fill them.