Smart Import

Upload raw materials (files, screenshots, or pasted text); the system recognizes questions in the background, and you review them before they are saved: New Import → background recognition → review → save. All imports live on the Import page in the sidebar, and refreshing or closing the page never loses progress.

There are two parsing methods:

• AI Recognition: AI automatically recognizes question structure; suited to documents and images without a fixed format.
• Tag-Based Parsing: parses precisely by tags such as 【题目】【选项】【答案】, without AI — instant and fully controllable.

The default is Auto: documents containing tags such as 【题目】 are parsed by tags (fast, and uses no AI quota); everything else uses AI Recognition.

Supported Formats

Type	Extensions	AI Recognition	Tag-Based Parsing
Word	.docx	✅	✅
Markdown / plain text	.md .txt / pasted text	✅	✅
Markdown archive	.zip (Markdown plus referenced local images; the whole archive is one import)	✅	✅
Image	.png .jpg .jpeg .webp, etc., or a pasted screenshot	✅ (vision model OCR)	—

Prerequisites

AI Recognition relies on AI models: text content uses the text model, and images use the vision model. Configure and activate the corresponding models in AI Providers & Models first. Tag-Based Parsing requires no AI configuration.

Tag-Based Parsing

If you already have an electronic draft and are willing to format it with fixed tags, this method is recommended — zero cost, instant, and 100% controllable.

Tag Syntax

Use 【题目】 to separate questions: a question starts at 【题目】 and ends at the next 【题目】 (or the end of the file). Standalone questions only need 【题目】; material-based question groups additionally need structural tags to declare the material and the group boundaries.

Tag	Field	Notes
【题目】	Stem	Required; also serves as the separator between questions
【题目材料】	Material	Required for groups; the name is optional — give one when reusing the same material
【题组】	Group start	Required for groups; if the name is omitted, it refers to the most recently declared material
【题组结束】	Group end	Optional; questions after it become standalone again
【题型】	Question type	Write 单选/多选/判断/填空/解答 (single choice / multiple choice / true-false / fill in the blank / free response); inferred automatically if omitted
【选项】	Options	Write A. … B. … C. … inside the block, or use 【选项A】【选项B】 one by one
【答案】	Answer	Letters for choice questions (e.g. AB for multiple choice); 对/错 (true/false) for true/false questions
【解析】	Explanation	Can span multiple lines
【思路】	Approach	Can span multiple lines
【难度】	Difficulty	1–5, or 简单/中等/困难 (easy/medium/hard)
【小结】	Summary	Can span multiple lines

Example

【题目】下列关于集合的说法正确的是
【选项】A. 空集是任何集合的子集  B. 0 是空集  C. {0} 是空集  D. 以上都不对
【答案】A
【解析】空集是任何集合的子集。

【题目】判断：0 是自然数
【答案】对

【题目】1 + 1 = ____
【答案】2

Mixing Material Groups and Standalone Questions

Sub-questions in a group are saved in the order they appear. Questions after 【题组结束】 become standalone again; the current group also ends automatically when a new group starts, a new material is declared, a question-type section heading is encountered, or the end of the file is reached.

【题目】卷首独立题
【答案】略

【题目材料】材料一
阅读下面的文字，完成后续题目。

这里可以包含多段 Markdown 内容。

【题组】材料一
1.【题目】请概括题目材料的主要内容
【题型】解答
【答案】略

2.【题目】下列理解正确的是
【题型】单选
【选项】A. 选项一  B. 选项二
【答案】A
【题组结束】

【题目】题组后的独立题
【答案】略

Material names are only used within the current file; the system generates internal temporary IDs automatically. The same material can be reused by multiple groups: just write 【题组】材料一 again. A 【题目】 inside a group is only a sub-question of that group and does not produce a duplicate standalone question in the composition.

Structural tags must be on a line of their own. Duplicate material names, references to unknown materials, empty materials, or empty groups block the import during the review pre-check, so the structure of the original file is never silently changed.

Unified Answer Section (Optional)

Some papers put all answers and explanations together at the end instead of following each question with 【答案】【解析】. In that case, add an 【答案区】 tag after the questions (aliases: 【答案表】/【统一答案】/【答案速查】). Content after the tag is parsed in the following two structures:

【答案区】
| 题号 | 1 | 2 |
|:---:|:---:|:---:|
| 答案 | B | C |

1．B因为……
2．C因为……

• Answers: taken by question number from the 题号 | 答案 (number | answer) table.

• Explanations: each numbered paragraph (1．…… 2．……) is used in full as the explanation for that question number. If a paragraph begins by repeating the answer (e.g. 10．ACD对于……), that repeated prefix is removed automatically.

• If a question's answer is not in the table (common for fill-in-the-blank and free-response questions, where the answer is written at the start of the explanation and is hard to separate), you can add 【答案】/【解析】 tags under that question number. These take precedence over the table and numbered paragraphs:

  12．【答案】20/21【解析】此人患色盲的概率……

• 【答案区】 may be preceded by other heading text (e.g. "参考答案【答案区】"); the tag does not need to be at the start of the line.

Automatic Question Type Inference

When 【题型】 is omitted, the type is determined as follows: options + a single answer letter → single choice; options + multiple letters → multiple choice; no options + answer is 对/错/√/× → true/false; no options + stem contains ____ → fill in the blank; everything else → free response.

Fault Tolerance and Warnings

The parser does its best and does not stop because of formatting problems in individual questions. Missing answers, out-of-range answers, empty stems, stray text at the beginning, and similar issues are flagged with a yellow warning on the affected question in the review page, so you can fix them on the spot.

Import Flow

1. New Import

On the Import page, click New Import, drag files onto the page, or press Ctrl/⌘ + V to paste a screenshot or text. You can pick several files at once; each file becomes its own import.

Imports go into the current subject (switch it at the top of the sidebar); recognized questions and the images inside the document belong to it.

• Parsing Method: Auto (default) / AI Recognition / Tag-Based Parsing.
• Answer: Use Answers from Document (default), or Let AI Fill In Answers and Explanations. This only applies to AI Recognition and is hidden for Tag-Based Parsing; tag-parsed documents always use the answers in the document.
• Save Automatically After Recognition: skips review; suited to well-formatted documents from trusted sources (this replaces the former batch import).

The parsing method and the automatic-save setting are remembered for your next import; Answer always defaults to Use Answers from Document.

2. Background Recognition

Imports are queued and recognized in the background. The list and detail view show the current stage (reading document → recognizing questions → matching knowledge points) and how many imports are ahead. You can leave the page or close the browser while it runs, and you can Cancel Import at any time. The desktop app shows a system notification when an import is ready, fails, or is paused, and the sidebar Import item shows how many imports are awaiting review.

3. Review

When recognition finishes, the import moves to Awaiting Review. Open it to view and edit each question, with fields including:

• Stem, question type, options, answer, explanation, approach, difficulty
• AI-suggested tags (year, source, grade, semester, exam type, feature, etc.)

You can edit, delete, or adjust each question before confirming, or Discard Import if you don't need it.

4. Save

After you confirm, the questions are written to the question bank in bulk and linked to this import, so you can later filter and manage them by batch; you can also save them as an editable composition.

Duplicate Check

On the review page, every question is automatically compared with the question bank and with the other questions in this import (same rules as Question Bank · Duplicate Check):

• A summary at the top shows how many questions are duplicates or highly similar; each question is marked "Duplicate / Similar xx%" and can be expanded to compare with existing questions.
• Standalone questions that duplicate an existing question, or an earlier question in this import, are unselected by default; highly similar questions stay selected with a warning.
• Keeping a question selected confirms importing it. Sub-questions of a material can't be unselected individually and are only marked.
• The server checks again on submit; duplicates added by others during your review are skipped unless confirmed, and the result explains why.

Import Status

Status	Meaning
Queued pending	Waiting for background recognition
Recognizing processing	Reading the document, recognizing questions, or matching knowledge points
Awaiting Review review_ready	Recognized and waiting for you to review and save
Paused paused	The AI service is temporarily unavailable; see below
Completed completed	Questions saved
Failed failed	The document couldn't be read, no questions were found, or the AI couldn't process the document; can be recognized again
Cancelled cancelled	Cancelled or discarded; can be recognized again

When the AI Fails

AI errors are handled by cause, so one failure never stalls the whole queue:

• Temporary problems (rate limits, server errors, timeouts, network drops): retried automatically a few times; if they keep failing, the import is paused as "service unreachable".
• Service or configuration problems (quota exhausted, invalid API key, model not found, no model configured): this import is paused, together with the queued AI imports that use the same provider. Tag-based imports keep running, and imports that use another provider (such as a personal key) are not affected.
• Problems with the document itself (too long, truncated output, unparsable model output): only this import fails; split the document and try again.

How paused imports resume:

Cause	Resume
Service unreachable	Retried automatically every 5 minutes
Invalid API key / no usable model	Resume automatically after you change the AI settings
Quota exhausted	Top up, then click Resume All on the Import page

You can always Resume a single import or click Resume All at the top of the page. As soon as a call to a provider succeeds, every import paused because of that provider resumes automatically.

Import List

• The list is grouped into Awaiting Review / In Progress / Paused / Failed / History, with imports awaiting review first.
• Permissions: regular users only see their own imports; Super Administrators can use the "⋯" menu to show everyone's imports or a specific user's, and to reset tasks stuck in recognition.
• Possible duplicates: imports saved automatically after recognition run unattended, so possible duplicates are saved as usual and flagged as "Possible duplicates" in the result. Open it to compare each question and delete the new ones you don't need.

FAQ

• Inaccurate recognition results? Adjust the Prompt Templates or switch to a stronger model.
• Image recognition fails? Make sure a vision model is activated and the image is clear.
• Imports stay paused? Check the notice at the top of the page: top up and click Resume All when the quota is exhausted; key or model problems resume automatically once you fix the AI settings.
• Stuck in recognition? (Server edition) Check the Worker logs; administrators can reset stuck tasks from the "⋯" menu. The desktop app puts interrupted imports back in the queue after a restart.