Block-level editing: make layout a drag, not a negotiation块级编辑：让排版变成一次拖拽，而不是一场拉锯

Every paragraph, list, image and card became a block with a handle. Dragging a block onto another creates columns; dragging into a heading folds it; dragging sideways indents. Block conversion switches a selection between styles in one move. The rule I held: a block operation must be doable with a fingertip, so the same model works on mobile where precise selection fails.每个段落、列表、图片和卡片都成为带手柄的块。拖到另一块旁边生成分栏，拖进标题即折叠，横向拖动即缩进。块转化一步切换选中内容的样式。我坚持的规则：块操作必须能用一根手指完成，这样同一套模型在精确选区失效的移动端也能成立。

• →Block drag: order, columns, fold, indent块拖拽：排序、分栏、折叠、缩进
• →Block convert: paragraph ↔ list ↔ quote ↔ card, without retyping块转化：段落 ↔ 列表 ↔ 引用 ↔ 卡片，不需要重新输入
• →Block link & comment: a block is an addressable unit, not just text块链接与块评论：块是可寻址的单位，而不只是一段文字

Keyboard-first: collapse mouse-heavy formatting into memorable commands全键盘操作：把靠鼠标的高频排版动作收敛为可记忆的指令

Markdown syntax for the fluent, a slash menu (快捷指令) for everyone else, and a shortcut system designed as a vocabulary rather than a list: the same modifier logic across headings, lists, cards and blocks so that learning one command teaches the next. The measure of success was simple: could a heavy user write and lay out an entire document without touching the mouse?熟手用 Markdown 语法，其他人用「/」快捷指令菜单，快捷键则设计成一套词汇而不是一张清单：标题、列表、卡片、块共用同一套修饰键逻辑，学会一个就等于学会下一个。成功标准很简单：重度用户能否全程不碰鼠标写完并排好一篇文档？

Feature cards: give structured content a first-class block特色卡片：给结构化内容一个正式的块

Callouts, collapsibles, columns, embedded files and links, image galleries, data tables, and small live apps such as polls, check-ins and calendars were unified as cards: one creation path (the plus panel or a slash command with a memorable pinyin shortcut such as /hb for whiteboard, /fl2 for two columns), consistent selection and drag behavior, one settings panel. A poll or check-in stays interactive for readers, so a document can carry a small application inside it. Users stopped pasting screenshots to get a layout and started composing with cards. This is also the "embed" stage of the model: front-loading pressure by pulling in what already exists.高亮块、折叠块、分栏、嵌入文件与链接、图册、数据表，以及投票、打卡、日历这类小应用被统一为卡片：一条创建路径（加号面板，或带拼音缩写的斜杠指令，如 /hb 画板、/fl2 两栏），一致的选中与拖拽行为，一个设置面板。投票和打卡在阅读端仍然可交互，一篇文档因此可以装下一个小应用。用户不再为了一个版式去贴截图，而是用卡片来组织内容。这也是模型里的「嵌入」阶段：把已有的东西引进来，把压力前置。

Knowledge network: from "a document" to "a body of knowledge"知识网络：从「一篇文档」到「一套知识」

Knowledge bases are trees. Human memory is a network. My concept paper argued that no user will dutifully walk the "note → document → knowledge base → permissions" path from the manual, so the product should let links form as a by-product of writing: paste a Yuque link and choose its form (plain, title, card, embed), @-mention a document, or pull in a fragment with a slash command. Published pages then show what they reference and what references them. I drove the integration of the editor and the knowledge base into one framework so that a team's knowledge stops being a folder of files and becomes a graph you can write into.知识库是树，人脑是网。我的构想文档里论证过：没有用户会老老实实按手册走完「小记 → 文档 → 知识库 → 权限」这条链路，所以产品应该让关联成为写作的副产物：粘贴一个语雀链接并选择它的形态（原链接、标题、卡片、嵌入），@ 提及一篇文档，或用斜杠引入一个片段。发布后的页面会显示它引用了什么、被谁引用。我推动编辑器与知识库合并为一套框架，让团队知识不再是一个文件夹，而是一张可以直接写进去的网。