Competency Mastery Inference

Competency mastery inference reads a student's full assessment history and produces a per-competency mastery map for advisor use and, separately, for transcript review. Each filled cell names a competency the catalog already names, states an inferred mastery level, and cites the assessments that support that inference. Cells stay empty when the history is too thin. The model does not invent a competency the catalog does not name. The registrar still owns what posts to the transcript.

Treat the map as a quality artifact. It is not a posted grade. The most common failure is to copy an inferred mastery level onto the transcript, into a competency credit, or into a course grade without a human posting rule. Inference can summarize evidence. It cannot confer credit.

The LMS and SIS already hold most of the raw record. Canvas, Blackboard, and Moodle typically store attempts, item scores, and rubric rows. Anthology and Ellucian typically store the catalog, enrollment, and posting rules. The work is to join those sources without filling a missing quiz with a guessed score.

A related downstream product is competency transcript generation, which turns a reviewed mastery record into a document the institution is willing to issue. Inference stops earlier: it cites evidence and leaves gaps visible.

Assemble the evidence the inference is allowed to see before any model runs. Pull every scored attempt that maps to a catalog competency for that student in that program: quizzes, exams, rubric-scored performances, lab checkoffs, practicum observations, and any formal prior-learning artifacts already accepted into the record.

Scope the pull to the student's program of record and the catalog edition that governs that program. A competency that exists in a later catalog version, or in a different major, is out of scope even if the wording looks similar.

Include metadata with every attempt: assessment identity, attempt date, score or rubric dimension, whether the attempt was excused or incomplete, and the competency codes already attached in the LMS or syllabus map. If an assessment has no competency mapping, it does not enter the inference. Do not infer a mapping from the assessment title.

Incomplete and excused work stay incomplete and excused. Filling a missing quiz with a guessed score, an imputed mean, or a "likely mastery" placeholder contaminates every cell that uses that quiz. The map should show that the attempt is absent, not that the student probably would have passed.

Item-level scores help when the LMS stored them. Course totals hide mixed competency coverage. A single exam can touch three competencies; averaging the exam into one number and applying it to all three overstates evidence for the weakly sampled ones.

Rubric-based essay scoring can feed dimension scores into this pull when those dimensions already map to named competencies. Adaptive item sets from adaptive assessment generation belong in the history only as the items actually administered, not as a bank the student never saw.

The catalog is the allow-list. Every cell on the mastery map must use a competency identifier and label the institution already publishes. If the model proposes a skill the catalog does not name, drop that cell. Do not merge near-synonyms into a new label.

Mapping is a join, not a rewrite. Use the competency codes already attached to assessments in the LMS, the official syllabus map, or the program assessment plan. Where those sources conflict, stop and flag the conflict. Do not silently prefer the source that produces a fuller map.

Mastery levels should use the scale the program already uses (for example, beginning, developing, proficient, or the numeric rubric the faculty approved). Do not introduce a proprietary scale the transcript office has never adopted.

Cite the assessments used per competency. A filled cell that cannot list its sources is not ready for advisor or registrar review. Citations should identify the assessment, the date or term, and the score or rubric dimension used. They should not bury a thin history under a long list of weakly related tasks.

One illustrative pass: a teacher-preparation student has three short quizzes and a rubric-scored lesson plan aligned to "Classroom Assessment Literacy," plus a midterm the student missed and that was recorded as incomplete. The quizzes and the lesson-plan rubric dimensions that name that competency enter the map. The incomplete midterm does not become a guessed score. If those remaining artifacts are enough under the program's evidence rule, the cell fills and lists those sources. If they are not, the cell stays empty even though the student is otherwise in good standing in the course.

Empty is the correct output when evidence is too thin. Thin means too few independent assessments, a single low-stakes quiz carrying the whole claim, coverage that only touches a corner of the competency, or a history that is mostly incomplete, excused, or withdrawn.

Do not backfill thin cells from neighboring competencies, from the course grade, or from the student's overall GPA. A high course grade can coexist with an empty competency cell when the course never assessed that competency well.

State the evidence rule in the same place the map is produced. Advisors should see why a cell is empty: "fewer than the required independent assessments," "only incomplete attempts on file," "assessment not mapped to this competency." Silence reads as a system error. An explicit empty cell reads as quality control.

Prior-learning artifacts belong in the history only after they have been accepted into the record under the institution's PLA process. Prior learning assessment evaluation is a separate judgment. Do not use an unevaluated portfolio to fill a mastery cell.

Advisors use the map to see where evidence is strong, where it is missing, and where to send the student next: retake, additional performance, or a conversation about whether the competency is even in the current term's assessments. The map is a briefing, not a verdict.

Registrars use the same object differently. They decide whether any cell is eligible to post, under which catalog rule, and in which transcript field. Nothing inferred posts automatically. A filled cell that cites assessments is still a recommendation until a posting rule and a human (or an explicitly authorized registrar workflow) accept it.

Keep inferred mastery out of the grade roster. If a faculty member wants the inference as context while assigning a course grade, show it as context. Do not write it into the grade book as if it were an earned score.

When the institution does issue a competency-bearing transcript line, generate that line from the posted record, not from the live inference. That is the handoff to competency transcript generation.

A mastery map answers whether the record supports a claim about a named competency. A degree audit answers whether the student has satisfied catalog requirements for a credential. Mixing the two produces false clearance: a student looks "proficient" on several competencies and someone treats that as degree complete.

Automated degree audit consumes posted courses, credits, substitutions, and residency rules. It should not consume inferred mastery cells. If a program wants competency completion to count toward a requirement, that path has to be a posted, cataloged rule, not an ML output.

Other failure modes sit next to this one. Posting inferred mastery as a grade writes a model output into the official record. Filling a missing quiz with a guessed score manufactures evidence. Treating the map as a degree audit skips the catalog. Each failure looks like a complete dashboard. Each one is a quality defect.

Run inference on a schedule the registrar and assessment office agree on (end of term, after grade submit, after PLA posting). Freeze the input snapshot so two reviewers see the same citations. If later assessment arrives, re-run and show the new citations rather than editing an old cell in place without a trail.

The usable output is a mastery map that cites its assessments, leaves thin history blank, names only catalog competencies, and waits for the registrar before anything reaches the transcript.