← Files Học YARCHIVED FILE

skills/hoc-y/references/adaptive-learning-engine.md

2.52 KB · Oct 5, 2026 · 18:37 UTC

↓ Download file

# Adaptive learning engine

> **Áp dụng cho học tập:** Mọi "ca", "bệnh nhân", "xử trí", "xét nghiệm" hay "decision" bên dưới chỉ là dữ kiện thuộc **tình huống hoàn toàn giả định, không có người bệnh thật**. Đọc cùng `educational-boundaries.md`; không sử dụng các gợi ý này trong chăm sóc, hồ sơ hoặc quyết định y tế thực tế.


## Cognitive ladder
Use the lowest level that still challenges the learner productively:

1. Recall
2. Explain/understand
3. Interpret
4. Apply
5. Discriminate between close alternatives
6. Prioritize/decide
7. Reassess/transfer

Not every topic or year needs all levels.

## Performance signals
Judge performance using both answer correctness and reasoning quality.

### Strong mastery
Signals:
- correct answer;
- identifies decisive clue or mechanism;
- can explain why close alternatives are wrong;
- transfers concept to a new context.

Response:
- reduce cues;
- increase application;
- introduce closer distractors;
- add uncertainty or competing priorities when age/year appropriate.

### Fragile mastery
Signals:
- correct answer with guessing;
- right label but wrong mechanism;
- relies on one memorized phrase;
- cannot transfer to a parallel case.

Response:
- keep difficulty stable;
- ask for explanation or teach-back;
- repair mechanism before escalating.

### Knowledge gap
Signals:
- repeated wrong answers;
- conflates core concepts;
- cannot identify salient information.

Response:
- step down one cognitive level;
- teach one compact mental model;
- give one worked example;
- retest with a different example.

## Hint ladder
For Socratic learning, increase help gradually:
1. Ask learner to identify the relevant domain/clue.
2. Narrow to two possibilities or give a directional cue.
3. Reveal one key principle.
4. Briefly teach the missing concept.
5. Restart with a parallel question.

Avoid giving the final answer at the first sign of difficulty unless safety requires it or the user asks directly.

## Difficulty calibration
Use year as a starting prior, then update from performance.

A high-performing Y2 may receive deeper mechanism/application questions, but should not be pushed into unsupervised management decisions. A struggling Y5 may temporarily return to mechanism or pattern recognition without being treated as a lower-status learner.

## Transfer
After correction, test whether the learner can apply the principle to a different example. A repeated identical question tests memory of feedback more than conceptual transfer.

SHA-256: 16d1835da5894623fb2905d432ca453bee56703d297103aeae318ade6e804c74