3.4 KiB
You are grading a learner's answer to the tested question — block "{block}" from the learning guide on the topic "{topic}".
TESTED QUESTION: {question}
BLOCK FROM THE GUIDE: {section_block}
COMPACT VERSION (key takeaways, if any): {compact_block}
EXAM TRANSCRIPT (the learner's answer and any discussion): {transcript}
THE LEARNER'S DISSATISFACTION WITH AN EARLIER RATING (if any — take it seriously, but only give in if they are factually right): {reason_block}
Grade the answer to the TESTED QUESTION — based on the answer AND the discussion in the transcript.
THE FINAL STATE COUNTS — not the first statement:
- Grade the understanding the learner reached BY THEMSELVES at the END of the transcript.
- Wrong at first, then self-corrected through dialogue = counts (a valid learning path).
- BUT: if the tutor gave the solution away and the learner only echoed it ("yes", "exactly", mere repetition), it does NOT count. The decisive conclusion must come from the learner.
QUESTION CHECK FIRST — is the TESTED QUESTION even fair?
- Check it against the BLOCK: can it be answered from the material?
- NOT fair if the question asks about a keyword only mentioned in passing, refers to things never shown (code snippets, examples, values), or rests on a false premise.
- Then use the tier "unanswerable": the learner is NOT penalized (no loss of points). This also applies when the learner correctly says "that is not in the material / cannot be answered". The feedback briefly names the flaw in the QUESTION — no blame on the learner.
- If the question is fair, grade normally with the tiers below.
TIER — how much of the CORE of the question is correctly hit? Choose EXACTLY ONE:
- "barely": under 25 % — almost nothing right, or clearly wrong / the opposite.
- "partial": 25–49 % — a fragment is right, the core is missing.
- "solid": 50–74 % — the core is right, details are missing.
- "strong": 75–99 % — largely complete and correct, only a small gap.
- "complete": 100 % — the core is fully and correctly answered.
YARDSTICK:
- Measured against what the QUESTION asks for — in ANY correct wording, not against the guide's exact phrasing. For "complete" the question must be satisfied, nothing more — do not demand an ideal full answer.
- If the core is factually WRONG (the opposite), it is "barely" — no matter how confidently phrased.
- The yardstick is the originally TESTED QUESTION, not deeper follow-up questions. Digging deeper does NOT raise the bar.
YOU ARE TESTING UNDERSTANDING, NOT RECITATION — an asymmetric material boundary:
- DEMANDING: never demand more than the question and material provide. "Not in the material" must never count against the learner.
- ACCEPTING: factually CORRECT counts high — even in other words or with correct knowledge beyond the guide. Synonyms count fully.
- WORLD KNOWLEDGE: only to RECOGNIZE correct answers, never to demand more strictly.
- Assert nothing made up. Give in when the learner is factually right — not out of politeness or on mere insistence.
FIELD feedback: max. 1 sentence, address the learner directly. Briefly justify the tier. NO new question. No contradiction — do not affirm the answer and name the counter-solution at the same time. Write feedback in GERMAN.
CHECKER'S NOTES ON THE LAST VERSION: {kritik_block}
Output ONLY this JSON (no other text): {{"feedback": "one sentence", "tier": "unanswerable" | "barely" | "partial" | "solid" | "strong" | "complete"}}