Learning

Competence diagnosis (competence check)

対象学生教員

What can the learner already do, what is this assessment based on, what is still missing — and how does the course help in a targeted way? With tasks from practice, evidence per criterion and reassessment.

目的#

Slides that have been read are not competence. The diagnosis checks in conversation whether someone can also apply knowledge, finds the specific gap, suggests a targeted learning step and then checks again with a new task.

使い方#

クリック手順CourseCourse overviewYour competences

1Open the course overview. If the competence check is switched on for the course, the invitation “What can you already do? Let's find out together where you stand.” appears at the top — “Start check” or “Skip”.
2Below it, under “Your competences”, three tiles: “My competence check”, “My competence profile” and “My next learning step”.
3In the check: for each competence (“Competence 1 of 5”) first a comprehension question, a follow-up question if needed, then a task from practice. Type answers or “Speak answer”, check the transcript and “Submit answer”.
4After every answer: “That was good” and “You can work on this”. At the end “Check completed” and the route “To the competence profile”.
5In the profile read, for each competence, the level, the justification, what is still missing and the learning need per criterion; “Continue learning” or “Reassess”.
6“My next learning step” takes you with one click to the right place in the course — afterwards “Reassess” with a new task.

設定とスイッチ#

テナントごとに切替可能既定：オフモジュール：Kompetenzcheck

Per tenant via the feature “Competence check (CIE)” (default off) or the bookable module “Competence check” — and additionally switchable per course.
Introduced as a project-based extension: the competences of a course (learning outcome, rubric, tasks) are set up together with the customer and approved by those responsible for the course — only then is there something to check. This does not mean immediate across-the-board availability for every course; for the pilot course “AI competence” five competences are ready.
Off means unchanged: no tiles, the competence pages answer with “not found”, player and learning progress stay as before.
The assessment uses the same text model that the tenant has enabled for the workshop; every call is booked in the usage.

What a competence is here#

A competence is an observable ability — not a chapter and not a lesson that has been read. Not “knows the risks of generative AI”, but “can identify relevant risks in a specific business application, justify their impact and select appropriate countermeasures”.

Every competence has a learning outcome (“What this means in concrete terms”), an expected level, assigned course content, suitable tasks and an assessment rubric with criteria. Definition and rubric carry a version.

Level 1 — Understanding
Level 2 — Applying
Level 3 — Analysing and deciding
Level 4 — Independent problem solving

The levels are an internal didactic order, not qualification levels. Consumed content never makes a competence “proven”.

The competence check step by step#

Header: the badge of the mode, “Competence n of m”, the current competence and the figure of the course. “Read question aloud” speaks the question with the course voice.
Comprehension question — a short starting question.
Follow-up question — adaptive: it builds on your own answer and targets the weakest criterion. It never counts as evidence of its own.
Task from practice — a case study, a decision with justification, an error analysis, a transfer to a changed situation or an open problem. At least one per competence.
Speak answer: the transcript appears in the field with the note to check and correct it before submitting. Without a microphone you type.
“Pause” and later “Resume”: the check picks up again at the task that was set.
An answer that cannot be evaluated “does not count against you”.

Learning mode and assessment mode#

The check starts in assessment mode: the answer is assessed; help is only available for understanding the question (“Question about the task”), not for the solution.

“Switch to learning mode” allows explanations, tips and examples. The current task then counts as practice, not as evidence — even if you switch “Back to assessment mode” afterwards.

The mode is always visible as a badge above the task. The switch is logged.

How assessment works#

0 to 3 points per criterion: 0 not met or serious error, 1 partially, 2 largely, 3 completely — each criterion with a justification, not by overall impression.
The rubric is fixed before answering and is never adjusted afterwards.
High scores need a verbatim quote from the answer; without it the score is reduced to 1 — visibly.
“Demonstrated” needs at least two independent pieces of evidence that were met, one of them from practice.
If pieces of evidence of the same kind clearly contradict each other, nothing is averaged: the level is “Still unclear”, and a further task clarifies it.
Eloquence, length, self-description or the model's confidence are not evidence.

The level of a competence#

Not assessed — no answers yet. That is never a gap.
Learning need identified — for example “Something essential was still missing in the practice task.”
Developing — partially shown, or proof is still missing.
Demonstrated — “Shown in several independent tasks, including in practice.”
Still unclear — contradictory evidence.

If a piece of evidence is waiting for review by a teacher, this is indicated. Percentages are deliberately not given — no spurious precision.

My competence profile#

For each competence: name, level, a short justification, “What this means in concrete terms”, “Still missing” (e.g. a practice task, further proof, a clarifying task, review by a teacher) and the learning need per criterion with the justification of the assessment.

“Download evidence (JSON)” gives a structured, verifiable file of your own evidence — not a certificate.

In the dashboard's learning progress the competence profile appears as a separate section next to mastery.

My next learning step#

Three blocks: “This is what you can already do.”, “This is where you still need to learn.”, “This is what we recommend next.” A click on “Start now” opens the course at the right slide with the right method; above the slide it then says what is being practised specifically. “Or rather:” offers alternatives.

The recommendation only comes when there is a reason — not automatically after every module. Competences already demonstrated are not taught again.

Basics missing → view slide, view explanation or example
Misconception → Ask me (Socratic) focused on the gap, otherwise explanation or slide
Application uncertain → view example, then guided exercise
Justification weak → Explain it back to me or Socratic
Problem solving missing → guided exercise (multi-step decision)
Unclear evidence → do not teach, reassess instead

Reassess#

After a learning step, “Afterwards: reassess” leads to a new, equivalent task for the same competence in a different context — never to the same task. The level results only from the new evidence; progress is only shown when it is supported. Original performance, learning step and new evidence are compared and recorded.

For those responsible for courses: maintaining competences#

Workshop course: in the phase “Content” the tab “Competences” (only when the feature is switched on). At the top the switch “Competence check in this course”.
“Suggest competences from the slides” — the suggestions appear with the status “Suggestion” and the origin “AI suggestion”.
For each competence: criteria of the rubric (type: knowledge, application, justification, analysis, problem solving; key criteria marked), assigned content (a missing slide is marked) and sample tasks.
“Edit rubric” → “Save as new version”: the changed rubric needs a new approval; previous assessments remain unchanged.
“Approve” makes a competence effective, “Withdraw” takes it out.
Lecturers switch file courses in the lecturer area under “Competences”; the review list of disputed assessments is there too.

Limits — honestly#

A tool for individual learning support. It claims no diagnostic validity: rubric points are not measurements, the level is a comprehensible assessment from evidence, not a certificate.
No certificate, no QR verification, no ranking, no use for personnel decisions, no passing on to superiors. The competency matrix of people development is deliberately not connected.
The assessment per criterion is made by a language model. The form is safeguarded (exactly the criteria, 0–3, quote, repetition on errors, “not assessable” instead of guessing); wrong judgements on content within a valid form are possible — hence two pieces of evidence, review list and correction with a log.
Very short, correct answers can be scored lower because of the quote rule; this is shown visibly.
Attempts to give the assessment instructions are detected by a narrow pattern recognition (German and English) plus the model; they only lead to review by a person, never to penalty points.
The content of the pilot course is German; answers in other languages are assessed.
A draft in the answer field is not saved — only submitted answers.
The invitation speaks of a few minutes; a check across all five competences of the pilot rather takes 15 to 25 minutes.
Accounts without an institution get no competence check — all data hangs on a tenant. Several runs of the same course share the competence level.

The pilot course “AI competence”#

For each competence there is at least one comprehension task and one independent application task. The pilot is a small course of its own; existing courses remain unchanged.

Basic understanding of generative AI
Assessing AI-generated information
Safe use at work
Selecting suitable use cases
Critical review of AI-supported decisions

ヒント#

No permanently visible buttons are added to the player. The competence diagnosis lives on the course overview; only a learning step leads into the player.
AI may suggest competences from the slides; they only take effect after approval by those responsible for the course.
A single competence can be checked specifically — from the profile with “Check now” or “Reassess”.
Thresholds and the number of required pieces of evidence are stored as versioned defaults (every piece of evidence names the version it was assessed by); deviations for an institution are set in the introduction project.