diff --git a/docs/subsystems/conversation.i18n.yaml b/docs/subsystems/conversation.i18n.yaml
index 0cb0498dff..64ca728d7e 100644
--- a/docs/subsystems/conversation.i18n.yaml
+++ b/docs/subsystems/conversation.i18n.yaml
@@ -18,8 +18,8 @@
   en: c216e2df128b8cb9
   zh: c039cf84b0d86ad8
 /conversation-assembly/definition-and-typed-chat-payload:
-  en: 266afa4ecc7a99bd
-  zh: 0a423eece69bebfd
+  en: d8f5e4773704c718
+  zh: f3db8c0d01080dba
 /conversation-assembly/predecessor-reads:
   en: da849b3b5da81da8
   zh: 37a647a439bc24b1
diff --git a/docs/subsystems/conversation.md b/docs/subsystems/conversation.md
index 87d9193ba4..ef6f6ffbdb 100644
--- a/docs/subsystems/conversation.md
+++ b/docs/subsystems/conversation.md
@@ -253,6 +253,18 @@ A match result with `exclusive: true` claims that event on the Definition's targ
 
 `target` and `buildViewNode(context)` declare one target-owned rendering contribution and must appear together. Preserve `context.key` as the React-facing identity, choose `anchorSeq` from the matched event ordering, and return only renderer-ready data. Once a target Node has been published, keep returning the same key; use `visibility: 'hidden'` when it must temporarily leave the visible flow rather than withdrawing it with `null`.
 
+A Chat Node also chooses where its row stands through `groupTree`. Positions are ordered from the outermost inward, and a Node at a position also has the effect of every inner one. Chat process grouping, whole-Turn folding, and Turn-input detection read only this position; [group-tree.ts](../../packages/client/ui-chat/src/client/contract/group-tree.ts) holds the complete contract.
+
+| The Node is | Declare |
+|---|---|
+| A person's input claimed in Step 1; its Turn opens an exchange | `exchange` |
+| Another input of its Turn, such as later human input or a trigger notice; before the process it is the opening input, after it the whole Turn stays open | `input` |
+| A Turn notice that stays outside whole-Turn folding, such as an error or the completed-turn footer | `turn` |
+| An independent row that whole-Turn folding still hides, such as a model retry | `process` |
+| Ordinary process content that joins the surrounding step run | omit `groupTree` |
+
+The [Chat business rules](../../packages/client/ui-chat/src/client/conversation-nodes/README.md#process-grouping) list the positions the built-in Nodes declare. When a generic Definition of the same target also matches the event your Node presents, claim the event with `exclusive: true` as described above.
+
 ## Predecessor reads
 
 Some Definitions need the latest earlier State of another business kind. `start` receives a `ConversationContextReader`; call `reader.previous<State>(kind)` there instead of accepting a Context collection or scanning events. The reader returns the nearest started Context before the current start `seq` as read-only data.
diff --git a/docs/subsystems/conversation.zh.md b/docs/subsystems/conversation.zh.md
index ce2d099dba..a3ef583db2 100644
--- a/docs/subsystems/conversation.zh.md
+++ b/docs/subsystems/conversation.zh.md
@@ -255,6 +255,18 @@ match 结果带 `exclusive: true` 时，即在该 Definition 的 target 上认
 
 `target` 与 `buildViewNode(context)` 必须同时声明一项由 target 拥有的渲染贡献。把 `context.key` 保留为 React 侧身份，根据持久排序证据选择 `anchorSeq`，并且只返回 renderer 可以直接使用的数据。某个 target Node 一旦发布，就要继续返回同一个 key；需要暂时离开可见流时使用 `visibility: 'hidden'`，不要改为返回 `null` 撤回它。
 
+Chat Node 还通过 `groupTree` 选择所在行的位置。位置由外到内排列，站在某个位置的 Node 同时具有所有内层位置的效果。Chat 的过程分组、整轮折叠和轮次输入判断只读取这个位置；完整契约见 [group-tree.ts](../../packages/client/ui-chat/src/client/contract/group-tree.ts)。
+
+| Node 是 | 声明 |
+|---|---|
+| Step 1 认领的人的输入，所在轮次开启一个交互段 | `exchange` |
+| 所在轮次的其他输入，如之后的人的输入或触发通知；位于过程之前时是开场输入，之后出现时整个轮次保持展开 | `input` |
+| 留在整轮折叠之外的轮次通知，如错误或已完成轮次页脚 | `turn` |
+| 整轮折叠仍会隐藏的独立行，如模型重试 | `process` |
+| 并入相邻步骤组的普通过程内容 | 不声明 `groupTree` |
+
+各内置 Node 声明的位置见 [Chat 业务规则](../../packages/client/ui-chat/src/client/conversation-nodes/README.zh.md#process-grouping)。如果同一 target 的通用 Definition 也会匹配你的 Node 所呈现的事件，按上文用 `exclusive: true` 认领该事件。
+
 ## Predecessor read
 
 有些 Definition 需要另一个业务 kind 在当前位置之前的最新 State。`start` 会收到 `ConversationContextReader`；应在这里调用 `reader.previous<State>(kind)`，不要接收 Context 集合或扫描事件。Reader 返回当前 start `seq` 之前最近一个已启动 Context 的只读数据。