+ It gives accurate classifications, proportions, and actions without guessing.
- Its conceptual-gap analysis and rationale for the single change are somewhat thin.
Sorts errors into knowledge gaps, application failures, and slips, and prescribes differently for each.
| Category | Study › Exams |
|---|---|
| Tags | AnalyzingCollege student |
Analyze the questions I got wrong. 1. **Classify each by cause:** - Did not know the concept - Knew it, could not apply it - Misread the question - Arithmetic or notation slip - Ran out of time 2. Count the proportions and name the dominant type. 3. **Group recurring conceptual gaps.** *This matters more than any individual question.* 4. Prioritize what to do in the remaining time below. ***The prescription differs by type*** — a knowledge gap needs study, a slip needs a checking procedure, and pacing needs drilling. **Do not default to "study more".** 5. Where I have not given enough detail to classify an error, **ask rather than guessing.** The classification is the whole value here. Then: the single change with the largest expected effect.
Lumping mistakes together just adds study hours. Careless errors need a procedure, not more revision — this separates them.
ChatGPT is the most accurate and concise while respecting missing information. Claude is specific but improperly combines error types, while Gemini contains many unsupported assumptions.
+ It gives accurate classifications, proportions, and actions without guessing.
- Its conceptual-gap analysis and rationale for the single change are somewhat thin.
+ It clearly states information limits and asks highly targeted questions.
- It inaccurately treats misreading and calculation slips as sharing one prescription.
+ It systematically provides type-specific actions and clarifying questions.
- It invents difficulty, point values, and timing, substantially hurting accuracy.
| Criterion | ChatGPT | Claude | Gemini | Leader |
|---|---|---|---|---|
| Instruction following | 9 | 9 | 8 | Tie |
| Accuracy | 10 | 8 | 4 | ChatGPT +25% |
| Specificity | 8 | 9 | 7 | Claude +13% |
| Structure | 9 | 9 | 8 | Tie |
| Right length | 10 | 7 | 4 | ChatGPT +43% |
Scored 1–10 by gpt-5.6-sol with model names hidden (2026-09-24). This is an AI review, not a measurement.
We gave three models the same input and copied their answers unedited. Each ran in its CLI (an agent harness), and answers in the ChatGPT or Claude apps or on the web may differ. Outputs are in Korean.
Time remaining: 2주 Wrong answers: 3번 몰라서 틀림 / 8번 계산 실수 / 12번 문제를 잘못 읽음 / 15번 두 개 중 찍어서 틀림 / 21번 시간 부족 / 24번 몰라서 틀림 / 29번 계산 실수
현재 7개 중 6개는 분류할 수 있습니다.
| 문항 | 원인 |
|---|---|
| 3번 | 개념을 몰라서 |
| 8번 | 계산 실수 |
| 12번 | 문제를 잘못 읽음 |
| 15번 | 추가 정보 필요 |
| 21번 | 시간 부족 |
| 24번 | 개념을 몰라서 |
| 29번 | 계산 실수 |
분류 가능한 6개 기준 비율은 다음과 같습니다.
다만 15번의 “두 개 중 찍어서 틀림”은 원인이 여러 가지일 수 있습니다.
이 중 어느 쪽이었나요? 그리고 3번·24번이 몰랐던 개념의 주제도 알려주세요. 그래야 반복되는 개념 공백을 실제 단원별로 묶을 수 있습니다.
남은 2주 처방은 원인별로 다르게 가야 합니다.
현재 정보만으로 가장 효과가 클 가능성이 높은 단일 변화는:
매일 제한시간을 둔 문제 세트를 풀고, 틀린 문제를 정답이 아니라 원인별로 기록하는 것입니다.
다만 15번이 개념 부족이었다면 개념 보완이 최우선으로 올라갈 수 있고, 시간 부족이었다면 timed drilling이 확실한 1순위가 됩니다.
| Academic analyst and exam pattern extractor | |
| Act as an Elite Course Mastery Tutor | |
| Adaptive AI Tutor — Personalized Learning Track with 6 Study Modes | |
| Adaptive Socratic Learning Coach | |
| AI Exam Mastery Tutor |