Mistake Book

Mistake Book (mistake-book)

A mistake book is not "copying the question" — it's turning a wrong answer into a searchable, reviewable knowledge card. This sub-skill sustains mistake accumulation and reminds the student to review at the right time.

Scope

• The user wants to "log a mistake" / "put the problem I just got wrong into my mistake book"
• The user wants to "organize / categorize / view my mistake book"
• The user wants to "review mistakes by subject / question type / error cause"
• The user wants to "export the mistake book"

Recording Flow (log one mistake)

Step 1: Gather info

• Question text (required)
• Wrong answer / wrong process (required)
• Correct solution (required; can reuse the result from explain-mistake)
• Subject / grade (for categorization)
• Error type (auto-determined; see the categories below)
• Importance (high/mid/low: is it a frequently-tested point, does it keep recurring)

Step 2: Structure into one record

A mistake record should contain at least:

- Subject / chapter
- Question (original text)
- My wrong answer
- Correct answer
- Error type (conceptual / formula / arithmetic / reasoning / reading)
- Key knowledge points (1-3 tags)
- Importance (high / mid / low)
- Date recorded

Step 3: Categorize & tag

• Categorize by subject + chapter/knowledge point (two levels)
• Tag each record with 1-3 knowledge-point tags for future search
• If the same knowledge point recurs → mark it "frequently missed", upgrade importance

Step 4: Persist and enter the review loop

Write the record into data/mistake-book.json with the script, auto-entering the forgetting-curve schedule:

node scripts/review-cycle.mjs add \
    --subject <subject> --chapter <chapter> --title "<question text>" \
    --mistake "<wrong approach>" --answer "<correct solution>" \
    --type <error type> --tags "<knowledge,knowledge>" --importance <high|mid|low>

The script auto-computes the first review date and writes the mastery-state fields (nextDue / intervalIdx / mastery).

Review Flow (review mistakes)

On-demand review

• When the user specifies a scope (subject / question type / error cause / importance), pull the matching mistakes
• An item wrong ≥ 2 times goes to the front of the list

Periodic review (forgetting curve)

The forgetting-curve schedule is actually driven by the script scripts/review-cycle.mjs (data in data/mistake-book.json), which auto-computes each review date on "day 1 → day 3 → day 7 → day 15 → day 30":

node scripts/review-cycle.mjs due          # mistakes due for review today
node scripts/review-cycle.mjs card <id>    # draw a review card (hide answer, redo independently)
node scripts/review-cycle.mjs done <id> --result correct   # correct → bump interval one level
node scripts/review-cycle.mjs done <id> --result wrong     # wrong → reset interval, back to explain-mistake

Review rules (built-in script state machine):

• Each review first hides the answer and has the student redo it independently
• Correct → bump the interval one level (1→3→7→15→30 days)
• Wrong → reset the interval to 1 day and re-run explain-mistake
• 5 consecutive correct → auto-mark "mastered"; the script stops scheduling it (downgrade)

The script handles "when to review"; the AI just calls card to draw a question at review time and done to advance based on the student's self-evaluation — no need to compute intervals from memory.

Output Template

On successful logging

【Logged】#N · <subject>/<chapter>
Question: <question summary>
My error: <error type> · <one sentence>
Key knowledge points: <tag1>, <tag2>
Importance: <high/mid/low>
(logged under <subject>/<chapter>, entered the forgetting-curve review plan)

On review

【Mistake Review · <scope>】
X item(s), sorted by importance:
1. <subject>/<chapter> · <question summary> (error: <type> · key: <tag>)
2. ...

【Review Advice】
Start with importance "high" and frequency "≥2"; hide the answer and redo independently.
Send me your answers when done — I'll check and update the mastery status.

Notes

• The core script of the forgetting-curve loop is scripts/review-cycle.mjs (add / due / card / done / list / stats / rm), data in data/mistake-book.json.
• A lightweight scripts/mistake-book.mjs (add / list / get / rm / stats) can do simple logging; both share the same data file.
• If the user prefers local storage (e.g., Markdown files, Notion, Obsidian), export on top of the script output.
• The "explanation" part of a mistake is done by explain-mistake; this sub-skill only handles accumulation, categorization, and review scheduling.