From 11368813cc1897f2ba03821eb4d15ac8d8b61118 Mon Sep 17 00:00:00 2001
From: yuetsh <517252939@qq.com>
Date: Mon, 28 Sep 2026 18:19:16 +0800
Subject: [PATCH] =?UTF-8?q?feat(=E6=B5=81=E7=A8=8B=E5=9B=BE):=20=E3=80=8C?=
 =?UTF-8?q?=E5=86=99=E4=BB=A3=E7=A0=81=20/=20=E7=94=BB=E6=B5=81=E7=A8=8B?=
 =?UTF-8?q?=E5=9B=BE=E3=80=8D=E5=8F=98=E6=88=90=E5=B7=A5=E5=85=B7=E6=A0=8F?=
 =?UTF-8?q?=E4=B8=8A=E7=9A=84=E5=88=87=E6=8D=A2=EF=BC=8CAI=20=E7=9A=84?=
 =?UTF-8?q?=E8=AF=84=E5=88=86=E5=92=8C=E5=8E=86=E6=AC=A1=E5=88=86=E6=95=B0?=
 =?UTF-8?q?=E6=8C=AA=E8=BF=9B=E7=BB=93=E6=9E=9C=E9=A1=B5=E7=AD=BE=EF=BC=8C?=
 =?UTF-8?q?=E9=A2=98=E5=8D=95=E9=87=8C=E4=B9=9F=E8=83=BD=E7=94=BB?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- 流程图题的工具栏最左边是「写代码 / 画流程图」。原来流程图藏在语言下拉里，叫一个名叫
  Flowchart 的「语言」；画完一道题之后它还被记成语言偏好，下一道能画的题一打开就是画布
- 「交给 AI 点评」交完切到「结果」页签：评的时候写着「AI 正在看你的图」、画布照样能改；
  评完是分数、评语、改进建议、四项得分，顶上一排历次分数，点哪次看哪次。
  原来是提交按钮旁一个 1000px 的弹框，盖住画布，翻历次评分要一页一页点
- 画到 A / S：给「照着它写代码 →」，题单里 1.5 秒后回题单（和代码通过一样），
  课堂条当场标成做完；只在这一页上刚评完的那次才算，进页面读到以前的 A 不跳
- 「把这一版载回画布」只在画布和那一版不一样时出现；载回、进页面读回草稿后都缩放到装下整张图
- 题单入口能画流程图了（原来只有题库入口能画），题单里的代码草稿、流程图草稿各存一份，
  加入题单之前写的、画的不会一进题单就摆在编辑器里
- 手机上不给「画流程图」（节点库是拖放，手机上拖不动），题面里以前画好的那张照样看得到

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 apps/web/CLAUDE.md                            |   2 +-
 apps/web/src/oj/api.ts                        |  14 +-
 .../src/oj/problem/components/ContextBar.vue  |  24 +-
 .../oj/problem/components/EditorToolbar.vue   |  24 +-
 .../oj/problem/components/FlowchartResult.vue | 388 ++++++++++++++++++
 .../oj/problem/components/ProblemContent.vue  |  16 +-
 .../oj/problem/components/ProblemEditor.vue   |  36 +-
 .../src/oj/problem/components/ResultPane.vue  |   9 +-
 .../oj/problem/components/ResultTabLabel.vue  |  18 +
 .../problem/components/SubmissionEffects.vue  |  14 +
 .../oj/problem/components/SubmitFlowchart.vue | 338 ++-------------
 apps/web/src/oj/store/code.ts                 |  23 +-
 apps/web/src/oj/store/flowchart.ts            | 147 +++++--
 apps/web/src/oj/store/problem.ts              |  42 +-
 .../components/FlowchartEditor/index.vue      |  38 +-
 ...2026-09-28-problem-page-redesign-design.md |   2 +-
 16 files changed, 747 insertions(+), 388 deletions(-)
 create mode 100644 apps/web/src/oj/problem/components/FlowchartResult.vue

diff --git a/apps/web/CLAUDE.md b/apps/web/CLAUDE.md
index 42ba7e1..cd67c69 100644
--- a/apps/web/CLAUDE.md
+++ b/apps/web/CLAUDE.md
@@ -58,7 +58,7 @@ API 调用不按模块分：学生端全在 `oj/api.ts`、后台全在 `admin/ap
 跨端的（登录、资料、标签、验证码）在 `shared/api.ts`。
 
 Shared logic lives in `shared/`:
-- `store/` — Pinia stores: `user` (auth/roles), `config` (site-wide settings), `authModal` (login/signup form state), `loginSummary` (AI activity summary), `collab` (help-request queue + collab room), `achievement` (解锁弹窗队列), `myFlowchart` (流程图弹窗的 mermaid 源码)
+- `store/` — Pinia stores: `user` (auth/roles), `config` (site-wide settings), `authModal` (login/signup form state), `loginSummary` (AI activity summary), `collab` (help-request queue + collab room), `achievement` (解锁弹窗队列), `myFlowchart` (题面「你画的流程图」那张 A/S 图的 mermaid 源码)
 - `composables/` — `aiStream`, `pagination` (URL-synced), `websocket` (reconnect + heartbeat), `collabDoc` (Yjs binding for the collab channel), `configUpdate` (WS-pushed config sync), `useMermaid` (lazy Mermaid render), `darkTransition` (View Transitions，111 以下走降级分支), `hiddenStudents` (统计面板的「请假隐藏」), `chartTheme`, `breakpoints`, `maxkb`, `learnProgress`, `rarity`
 - `layout/` — `default.vue` and `admin.vue` layout wrappers
 - `api.ts` — shared API calls (auth, profile, tags, captcha)
diff --git a/apps/web/src/oj/api.ts b/apps/web/src/oj/api.ts
index f902f08..b01da55 100644
--- a/apps/web/src/oj/api.ts
+++ b/apps/web/src/oj/api.ts
@@ -23,8 +23,7 @@ import {
   type YearlyAc,
   type ProblemList,
   type CreateFlowchartResponse,
-  type FlowchartCurrent,
-  type FlowchartDetail,
+  type FlowchartScores,
   type FlowchartList,
   type FlowchartSubmission,
   type AiDetail,
@@ -437,14 +436,9 @@ export function retryFlowchartSubmission(submissionId: string) {
   return api.post<{ status: string }>(`flowcharts/${encodeURIComponent(submissionId)}/retry`)
 }
 
-export function getCurrentProblemFlowchartSubmission(problemId: number) {
-  return api.get<FlowchartCurrent>(`problems/${problemId}/flowchart/current`)
-}
-
-export function getFlowchartSubmissionDetail(problemId: number, page = 0) {
-  return api.get<FlowchartDetail>(`problems/${problemId}/flowchart/history`, {
-    params: { page },
-  })
+/** 自己在这道题上评完的每一次（早的在前），加入题单之前的已经滤掉 */
+export function getFlowchartScores(problemId: number) {
+  return api.get<FlowchartScores>(`problems/${problemId}/flowchart/scores`)
 }
 
 // ==================== 题单相关API ====================
diff --git a/apps/web/src/oj/problem/components/ContextBar.vue b/apps/web/src/oj/problem/components/ContextBar.vue
index 2d6b29f..424afa0 100644
--- a/apps/web/src/oj/problem/components/ContextBar.vue
+++ b/apps/web/src/oj/problem/components/ContextBar.vue
@@ -3,6 +3,7 @@ import { storeToRefs } from "pinia"
 import type { RouteLocationRaw } from "vue-router"
 import { getProblemSetDetail, getProblemSetProblems } from "oj/api"
 import { useContestStore } from "oj/store/contest"
+import { isFlowchartPass, useFlowchartStore } from "oj/store/flowchart"
 import { useLessonStore } from "oj/store/lesson"
 import { useProblemStore } from "oj/store/problem"
 import { useUserStore } from "shared/store/user"
@@ -30,7 +31,11 @@ const userStore = useUserStore()
 const { problem } = storeToRefs(useProblemStore())
 const currentId = computed(() => problem.value?._id ?? "")
 
-/** 刚在这一页上做出来（不是本来就做过）：换题时 id 跟着变，那一次不算 */
+/**
+ * 刚在这一页上做出来（不是本来就做过）：代码通过，或者流程图刚评到 A / S。
+ * 换题时 id 跟着变，那一次不算；流程图只认这一页上刚评完的（evaluatedSeq）
+ */
+const flowchartStore = useFlowchartStore()
 function onSolvedHere(callback: () => void) {
   watch(
     () => [problem.value?.id, problem.value?.myStatus] as const,
@@ -38,6 +43,12 @@ function onSolvedHere(callback: () => void) {
       if (id !== undefined && id === previousId && status === 0 && previousStatus !== 0) callback()
     },
   )
+  watch(
+    () => flowchartStore.evaluatedSeq,
+    () => {
+      if (isFlowchartPass(flowchartStore.latestRating.grade)) callback()
+    },
+  )
 }
 
 // ==================== 课堂 ====================
@@ -124,13 +135,18 @@ const setBar = computed(() => {
           params: { problemSetId: ctx.value.problemSetId, problemID: item.problem._id },
         }
       : null
-  const done = list.filter((item) => item.isCompleted).length
+  // 进度只数必做题，和题单页的「完成进度 2 / 8（另有 3 道选做）」同一个口径；
+  // 一道必做都没有的题单才数全部
+  const required = list.some((item) => item.isRequired)
+    ? list.filter((item) => item.isRequired)
+    : list
+  const done = required.filter((item) => item.isCompleted).length
   return {
     title: problemSet.value.title,
     home: { name: "problemset", params: { problemSetId: ctx.value.problemSetId } },
     done,
-    total: list.length,
-    percentage: list.length ? (done / list.length) * 100 : 0,
+    total: required.length,
+    percentage: required.length ? (done / required.length) * 100 : 0,
     position: index >= 0 ? index + 1 : null,
     previous: index >= 0 ? link(list[index - 1]) : null,
     next: index >= 0 ? link(list[index + 1]) : null,
diff --git a/apps/web/src/oj/problem/components/EditorToolbar.vue b/apps/web/src/oj/problem/components/EditorToolbar.vue
index edfd939..694ffc6 100644
--- a/apps/web/src/oj/problem/components/EditorToolbar.vue
+++ b/apps/web/src/oj/problem/components/EditorToolbar.vue
@@ -44,7 +44,20 @@ const ctx = useProblemPageContext()
 const userStore = useUserStore()
 const codeStore = useCodeStore()
 const problemStore = useProblemStore()
-const { problem, languages } = storeToRefs(problemStore)
+const { problem, codeLanguages, canDraw } = storeToRefs(problemStore)
+
+/**
+ * 「写代码 / 画流程图」是两种作业，不是两门语言。原来流程图藏在语言下拉里，叫一个名叫
+ * Flowchart 的「语言」—— 有的题流程图交了几百次、代码个位数，入口却在下拉的第一项里。
+ */
+const drawing = computed(() => codeStore.code.language === "Flowchart")
+const mode = computed({
+  get: () => (drawing.value ? "draw" : "code"),
+  set: (value: "draw" | "code") => {
+    if (value === "draw") problemStore.switchLanguage("Flowchart")
+    else problemStore.switchLanguage(problemStore.supportedLanguage(codeStore.preferredLanguage()))
+  },
+})
 
 const { isDesktop } = useBreakpoints()
 
@@ -183,7 +196,7 @@ const handleMenuSelect = (key: string) => {
 
 // computed：换题时组件复用，选项得跟着新题的语言走
 const languageOptions = computed<DropdownOption[]>(() =>
-  languages.value.map((it) => ({
+  codeLanguages.value.map((it) => ({
     label: () =>
       h(NFlex, { align: "center" }, () => [
         h(Icon, {
@@ -239,8 +252,15 @@ const goEdit = () => {
 
 <template>
   <n-flex align="center">
+    <!-- 协作中的老师不会落到画图：求助入口本身就排掉了流程图 -->
+    <n-radio-group v-if="canDraw && !showCollabBar" v-model:value="mode" :size="buttonSize">
+      <n-radio-button value="code">写代码</n-radio-button>
+      <n-radio-button value="draw">画流程图</n-radio-button>
+    </n-radio-group>
+
     <!-- 协作中编辑器的语言跟着学生走，这个选择器改了也不会生效，索性禁掉 -->
     <n-select
+      v-if="!drawing"
       v-model:value="codeStore.code.language"
       style="width: 120px"
       :size="buttonSize"
diff --git a/apps/web/src/oj/problem/components/FlowchartResult.vue b/apps/web/src/oj/problem/components/FlowchartResult.vue
new file mode 100644
index 0000000..3f9b30f
--- /dev/null
+++ b/apps/web/src/oj/problem/components/FlowchartResult.vue
@@ -0,0 +1,388 @@
+<script setup lang="ts">
+import { storeToRefs } from "pinia"
+import { useCodeStore } from "oj/store/code"
+import { isFlowchartPass, useFlowchartStore } from "oj/store/flowchart"
+import { useProblemStore } from "oj/store/problem"
+import { sortFlowchartCriteria } from "utils/constants"
+import { decompressFromBase64, parseTime } from "utils/functions"
+
+/**
+ * 「结果」页签里流程图作业的那一块：AI 在评 / 评完了几分 / 历次分数 / 评语和改进建议。
+ *
+ * 原来评分详情是提交按钮旁一个 1000px 的弹框：盖住画布，学生没法一边看改进建议一边改图；
+ * 翻历次评分要一页一页点分页器。现在摆在左栏，画布整列都在右边（设计文档第 6 节）。
+ */
+
+const flowchartStore = useFlowchartStore()
+const { phase, scores, hiddenCount, selectedId, details, detailLoading, editor } =
+  storeToRefs(flowchartStore)
+const problemStore = useProblemStore()
+const codeStore = useCodeStore()
+const message = useMessage()
+
+/** 正在看的那一次：点过哪次就是哪次，否则是最新的 */
+const current = computed(() => {
+  const id = selectedId.value ?? scores.value.at(-1)?.id
+  return id ? (scores.value.find((row) => row.id === id) ?? null) : null
+})
+const currentIndex = computed(() =>
+  current.value ? scores.value.findIndex((row) => row.id === current.value!.id) : -1,
+)
+const detail = computed(() => (current.value ? details.value[current.value.id] : undefined))
+const viewingLatest = computed(() => selectedId.value === null)
+
+// 评完的那次、点到的那次，评语都要现拉（历次分数里只有分数）
+watch(
+  () => current.value?.id,
+  (id) => {
+    if (id && !details.value[id]) flowchartStore.select(selectedId.value)
+  },
+  { immediate: true },
+)
+
+const criteria = computed(() =>
+  sortFlowchartCriteria(
+    (detail.value?.aiCriteriaDetails ?? {}) as Record<
+      string,
+      { score: number; max: number; comment?: string }
+    >,
+  ),
+)
+const suggestions = computed(() =>
+  (detail.value?.aiSuggestions ?? "")
+    .split("\n")
+    .map((line) => line.trim())
+    .filter(Boolean),
+)
+
+function gradeType(grade: string) {
+  if (grade === "S" || grade === "A") return "success"
+  if (grade === "B") return "warning"
+  return "error"
+}
+
+function percentType(ratio: number) {
+  if (ratio >= 0.8) return "success"
+  if (ratio >= 0.6) return "info"
+  if (ratio >= 0.4) return "warning"
+  return "error"
+}
+
+// ==================== 载回画布 ====================
+/** 这一版的画布数据（压缩前的 JSON 字符串）；老提交的压缩数据可能是坏的 */
+const versionJson = computed(() => {
+  const data = detail.value?.flowchartData?.data
+  if (typeof data !== "string" || !data) return null
+  try {
+    return decompressFromBase64(data)
+  } catch {
+    return null
+  }
+})
+
+/**
+ * 一张图「长什么样」：节点的 id / 类型 / data（字写在里面）/ 位置，连线的两端和标签。不直接比 JSON 字符串 ——
+ * 载回画布时编辑器会补上默认字段（sourceHandle: null 之类），老提交还带着 vue-flow 的运行时
+ * 字段，字符串永远对不上，按钮就一直在。
+ */
+type Shape = { nodes?: unknown[]; edges?: unknown[] }
+function signature(data: Shape) {
+  const nodes = (data.nodes ?? []).map((raw) => {
+    const node = raw as {
+      id?: string
+      type?: string
+      data?: unknown
+      position?: { x?: number; y?: number }
+    }
+    return [
+      node.id,
+      node.type,
+      // 节点上的字（customLabel）和原始类型都在 data 里，整个带上
+      JSON.stringify(node.data ?? {}),
+      Math.round(node.position?.x ?? 0),
+      Math.round(node.position?.y ?? 0),
+    ].join("|")
+  })
+  const edges = (data.edges ?? []).map((raw) => {
+    const edge = raw as {
+      source?: string
+      target?: string
+      sourceHandle?: string | null
+      targetHandle?: string | null
+      label?: unknown
+    }
+    return [
+      edge.source,
+      edge.target,
+      edge.sourceHandle ?? "",
+      edge.targetHandle ?? "",
+      edge.label ?? "",
+    ].join("|")
+  })
+  return JSON.stringify([nodes.sort(), edges.sort()])
+}
+
+const versionSignature = computed(() => {
+  if (!versionJson.value) return null
+  try {
+    return signature(JSON.parse(versionJson.value) as Shape)
+  } catch {
+    return null
+  }
+})
+
+/**
+ * 只有画布和这一版不一样时才给「载回画布」：刚交完、还没动过的时候，这个按钮点了什么
+ * 都不会变，只会让人以为它坏了。读 getFlowchartData() 会顺带跟踪画布的节点和边。
+ */
+const canvasDiffers = computed(() => {
+  if (!editor.value || !versionSignature.value) return false
+  return signature(editor.value.getFlowchartData()) !== versionSignature.value
+})
+
+function loadToCanvas() {
+  if (!editor.value || !versionJson.value) return
+  try {
+    const json = JSON.parse(versionJson.value)
+    editor.value.setFlowchartData({ nodes: json.nodes || [], edges: json.edges || [] })
+    message.success("已载回画布，按 Ctrl+Z 可以撤回")
+  } catch {
+    message.error("这一版的图数据坏了，载不回画布")
+  }
+}
+
+/** A / S 之后：照着图写代码。切回学生习惯的那门语言 */
+const canCode = computed(() => problemStore.codeLanguages.length > 0)
+function goCode() {
+  problemStore.switchLanguage(problemStore.supportedLanguage(codeStore.preferredLanguage()))
+}
+</script>
+
+<template>
+  <div class="flowchart-result">
+    <!-- 正在评：评的是最新这一次，所以只在看最新时显示 -->
+    <n-alert
+      v-if="phase === 'evaluating' && viewingLatest"
+      type="info"
+      :show-icon="false"
+      class="block"
+    >
+      <n-flex align="center" :size="10" :wrap="false">
+        <n-spin :size="18" />
+        <div>
+          <b>AI 正在看你的图</b>（一般 5 秒左右）
+          <div class="note">评的时候画布还能改，改了不影响这一次</div>
+        </div>
+      </n-flex>
+    </n-alert>
+
+    <n-alert
+      v-else-if="phase === 'failed' && viewingLatest"
+      type="warning"
+      :show-icon="false"
+      class="block"
+    >
+      AI 这次没评出来，再点一次「交给 AI 点评」试试
+    </n-alert>
+
+    <n-empty
+      v-if="!scores.length && phase !== 'evaluating'"
+      class="empty"
+      description="还没交过流程图。画好之后点「交给 AI 点评」，评分和建议会出现在这里"
+    />
+
+    <template v-if="scores.length">
+      <!-- 历次分数：一眼看出是不是越改越好 -->
+      <div class="history" role="list" aria-label="历次分数">
+        <button
+          v-for="(row, index) in scores"
+          :key="row.id"
+          type="button"
+          role="listitem"
+          class="attempt"
+          :class="[gradeType(row.grade), { active: current?.id === row.id }]"
+          :title="parseTime(row.createTime, 'MM-DD HH:mm')"
+          @click="flowchartStore.select(index === scores.length - 1 ? null : row.id)"
+        >
+          <span class="attempt-no">第 {{ index + 1 }} 次</span>
+          <b>{{ row.score }}</b>
+          <span>{{ row.grade }}</span>
+        </button>
+      </div>
+      <p v-if="hiddenCount" class="note">
+        另外 {{ hiddenCount }} 次是加入题单之前画的，先藏起来了：在题单里画到 A 或 S 就解锁
+      </p>
+
+      <section v-if="current" class="block">
+        <n-flex align="baseline" :size="10">
+          <n-text :type="gradeType(current.grade)" class="score">
+            {{ current.score }} 分 · {{ current.grade }} 级
+          </n-text>
+          <n-text depth="3" class="note">
+            第 {{ currentIndex + 1 }} 次 · {{ parseTime(current.createTime, "MM-DD HH:mm") }}
+          </n-text>
+        </n-flex>
+
+        <n-flex v-if="isFlowchartPass(current.grade)" align="center" class="pass" :size="12">
+          <span>画到 {{ current.grade }} 级了，这道题的流程图算做完 ✓</span>
+          <n-button v-if="canCode" size="small" type="primary" @click="goCode">
+            照着它写代码 →
+          </n-button>
+        </n-flex>
+
+        <n-flex v-if="detailLoading && !detail" justify="center" class="pad">
+          <n-spin size="small" />
+        </n-flex>
+
+        <template v-if="detail">
+          <p v-if="detail.aiFeedback" class="feedback">{{ detail.aiFeedback }}</p>
+
+          <div v-if="suggestions.length" class="suggestions">
+            <h4>可以这样改</h4>
+            <ol>
+              <li v-for="(line, index) in suggestions" :key="index">{{ line }}</li>
+            </ol>
+          </div>
+
+          <div v-if="criteria.length" class="criteria">
+            <h4>四项得分</h4>
+            <div v-for="[name, item] in criteria" :key="name" class="criterion">
+              <n-flex justify="space-between" align="center">
+                <b>{{ name }}</b>
+                <n-tag :type="percentType(item.score / item.max)" size="small" round>
+                  {{ item.score || 0 }} / {{ item.max }}
+                </n-tag>
+              </n-flex>
+              <n-text v-if="item.comment" depth="3" class="comment">{{ item.comment }}</n-text>
+            </div>
+          </div>
+
+          <n-button v-if="canvasDiffers" secondary class="load" @click="loadToCanvas">
+            把这一版载回画布
+          </n-button>
+        </template>
+      </section>
+    </template>
+  </div>
+</template>
+
+<style scoped>
+.flowchart-result {
+  padding-bottom: 16px;
+}
+
+.block {
+  margin-bottom: 14px;
+}
+
+.empty {
+  margin-top: 48px;
+}
+
+.note {
+  font-size: 13px;
+  opacity: 0.75;
+  margin: 2px 0 0;
+}
+
+.pad {
+  padding: 16px 0;
+}
+
+.history {
+  display: flex;
+  gap: 6px;
+  overflow-x: auto;
+  padding: 2px 2px 8px;
+}
+
+.attempt {
+  flex: none;
+  display: flex;
+  flex-direction: column;
+  align-items: center;
+  gap: 1px;
+  min-width: 58px;
+  padding: 6px 8px;
+  border-radius: 8px;
+  border: 1px solid rgba(128, 128, 128, 0.25);
+  background: transparent;
+  color: inherit;
+  font: inherit;
+  font-size: 13px;
+  cursor: pointer;
+}
+
+.attempt b {
+  font-size: 17px;
+  font-variant-numeric: tabular-nums;
+}
+
+.attempt-no {
+  font-size: 11px;
+  opacity: 0.65;
+}
+
+.attempt.success b {
+  color: #18a058;
+}
+
+.attempt.warning b {
+  color: #f0a020;
+}
+
+.attempt.error b {
+  color: #d03050;
+}
+
+.attempt.active {
+  border-color: #2080f0;
+  box-shadow: 0 0 0 1px #2080f0;
+}
+
+.score {
+  font-size: 22px;
+  font-weight: 700;
+}
+
+.pass {
+  margin: 10px 0 4px;
+  padding: 8px 12px;
+  border-radius: 6px;
+  background-color: rgba(24, 160, 88, 0.1);
+}
+
+.feedback {
+  margin: 12px 0;
+  line-height: 1.7;
+}
+
+h4 {
+  margin: 14px 0 6px;
+  font-size: 14px;
+}
+
+.suggestions ol {
+  margin: 0;
+  padding-left: 1.4em;
+  line-height: 1.7;
+}
+
+.criterion {
+  padding: 6px 0;
+}
+
+.criterion + .criterion {
+  border-top: 1px dashed rgba(128, 128, 128, 0.2);
+}
+
+.comment {
+  display: block;
+  margin-top: 2px;
+  font-size: 13px;
+}
+
+.load {
+  margin-top: 14px;
+}
+</style>
diff --git a/apps/web/src/oj/problem/components/ProblemContent.vue b/apps/web/src/oj/problem/components/ProblemContent.vue
index 6da06e7..a2f3545 100644
--- a/apps/web/src/oj/problem/components/ProblemContent.vue
+++ b/apps/web/src/oj/problem/components/ProblemContent.vue
@@ -22,7 +22,7 @@ const MyFlowchart = defineAsyncComponent(() => import("./MyFlowchart.vue"))
 
 const isDark = useDark()
 const problemStore = useProblemStore()
-const { problem, languages } = storeToRefs(problemStore)
+const { problem } = storeToRefs(problemStore)
 const ctx = useProblemPageContext()
 
 // SQL 题：隐藏输入/输出/例子，改为渲染数据表与期望结果
@@ -103,11 +103,17 @@ const flowchartStore = useFlowchartStore()
 const userStore = useUserStore()
 const myFlowchartZoom = ref(false)
 
-// 这道题能画流程图，就去查画到 A/S 没有 —— 原来要先切到「流程图」这门「语言」才查
+// 这道题能画流程图，就去查画到 A/S 没有 —— 原来要先切到「流程图」这门「语言」才查。
+// 不看 canDraw：手机上画不了，但以前在电脑上画好的那张照样该摆出来
 watch(
-  () => [problem.value?.id, languages.value.includes("Flowchart"), userStore.isAuthed] as const,
-  ([id, canDraw, authed]) => {
-    if (id && canDraw && authed) flowchartStore.ensureLoaded()
+  () =>
+    [
+      problem.value?.id,
+      !!problem.value?.allowFlowchart && ctx.value.entry !== "contest",
+      userStore.isAuthed,
+    ] as const,
+  ([id, drawable, authed]) => {
+    if (id && drawable && authed) flowchartStore.ensureLoaded()
   },
   { immediate: true },
 )
diff --git a/apps/web/src/oj/problem/components/ProblemEditor.vue b/apps/web/src/oj/problem/components/ProblemEditor.vue
index f1628e0..8793fdb 100644
--- a/apps/web/src/oj/problem/components/ProblemEditor.vue
+++ b/apps/web/src/oj/problem/components/ProblemEditor.vue
@@ -11,6 +11,7 @@ import type { LANGUAGE } from "utils/types"
 import { beginEditTrace, editTraceExtensions } from "oj/problem/utils/editTrace"
 import { errorMarkExtensions } from "oj/problem/utils/errorMark"
 import EditorToolbar from "./EditorToolbar.vue"
+import { useFlowchartStore } from "oj/store/flowchart"
 
 const FlowchartEditor = defineAsyncComponent(
   () => import("shared/components/FlowchartEditor/index.vue"),
@@ -84,8 +85,16 @@ const { isDesktop } = useBreakpoints()
 const editorExtensions = [...editTraceExtensions, ...errorMarkExtensions]
 
 const contestID = route.params.contestID || null
-const storageKey = computed(
-  () => `problem_${problem.value!._id}_contest_${contestID}_lang_${codeStore.code.language}`,
+const problemSetId = route.params.problemSetId || null
+/**
+ * 本地草稿的键。题单入口单独一份：原来题库和题单共用 `problem_题号_contest_null_语言`，
+ * 加入题单之前在题库里写对的代码，一进题单就摆在编辑器里（设计文档第 11 节 4）。
+ * 键的最后一段必须是语言，changeLanguage 靠它判断草稿是不是这门语言的。
+ */
+const storageKey = computed(() =>
+  problemSetId
+    ? `problem_${problem.value!._id}_problemset_${problemSetId}_lang_${codeStore.code.language}`
+    : `problem_${problem.value!._id}_contest_${contestID}_lang_${codeStore.code.language}`,
 )
 
 const editorHeight = computed(() =>
@@ -130,7 +139,28 @@ const changeLanguage = (v: LANGUAGE) => {
   )
 }
 
-// 提供FlowchartEditor的ref给子组件
+/**
+ * 换语言的请求：工具栏的「写代码 / 画流程图」、结果页签里的「照着它写代码」。
+ * 和语言下拉走同一条路（存语言偏好 → 读那门语言的草稿）。
+ */
+watch(
+  () => problemStore.languageRequest,
+  (language) => {
+    if (!language) return
+    problemStore.languageRequest = null
+    if (language === codeStore.code.language) return
+    codeStore.setLanguage(language)
+    changeLanguage(language)
+  },
+)
+
+// 流程图编辑器交给 store：结果页签（在左栏，不在编辑器底下）要拿当前画布比一比、
+// 把旧版本载回去。子组件 SubmitFlowchart 还是从 inject 拿
+const flowchartStore = useFlowchartStore()
+watch(flowchartEditorRef, (editor) => flowchartStore.attachEditor(editor ?? null), {
+  immediate: true,
+})
+onBeforeUnmount(() => flowchartStore.attachEditor(null))
 provide("flowchartEditorRef", flowchartEditorRef)
 </script>
 
diff --git a/apps/web/src/oj/problem/components/ResultPane.vue b/apps/web/src/oj/problem/components/ResultPane.vue
index 77a174d..6e4b457 100644
--- a/apps/web/src/oj/problem/components/ResultPane.vue
+++ b/apps/web/src/oj/problem/components/ResultPane.vue
@@ -2,6 +2,7 @@
 import { storeToRefs } from "pinia"
 import { useProblemStore } from "oj/store/problem"
 import { useSubmissionStore } from "oj/store/submission"
+import { useCodeStore } from "oj/store/code"
 
 /**
  * 左栏「结果」页签：三段 —— 提交结果 / 运行例子 / 自己输入。
@@ -17,10 +18,15 @@ const SubmissionResult = defineAsyncComponent(() => import("./SubmissionResult.v
 const PythonErrorExplain = defineAsyncComponent(() => import("./PythonErrorExplain.vue"))
 const SampleRunResult = defineAsyncComponent(() => import("./SampleRunResult.vue"))
 const CustomRun = defineAsyncComponent(() => import("./CustomRun.vue"))
+const FlowchartResult = defineAsyncComponent(() => import("./FlowchartResult.vue"))
 
 const submissionStore = useSubmissionStore()
 const { submission, syntaxErrorInfo, resultSegment } = storeToRefs(submissionStore)
 const { problem } = storeToRefs(useProblemStore())
+const codeStore = useCodeStore()
+
+/** 正在画流程图：结果就是 AI 的点评，没有例子可跑 */
+const drawing = computed(() => codeStore.code.language === "Flowchart")
 
 // 试跑走 Judge0，它跑不了 SQL；流程图也没得跑。这两种只留「提交结果」
 const canTrial = computed(() => {
@@ -30,7 +36,8 @@ const canTrial = computed(() => {
 </script>
 
 <template>
-  <div class="result-pane">
+  <FlowchartResult v-if="drawing" />
+  <div v-show="!drawing" class="result-pane">
     <n-radio-group v-if="canTrial" v-model:value="resultSegment" size="small" class="segments">
       <n-radio-button value="submit">提交结果</n-radio-button>
       <n-radio-button value="samples">运行例子</n-radio-button>
diff --git a/apps/web/src/oj/problem/components/ResultTabLabel.vue b/apps/web/src/oj/problem/components/ResultTabLabel.vue
index 478121b..84f8616 100644
--- a/apps/web/src/oj/problem/components/ResultTabLabel.vue
+++ b/apps/web/src/oj/problem/components/ResultTabLabel.vue
@@ -3,6 +3,8 @@ import { Icon } from "@iconify/vue"
 import { useThemeVars } from "naive-ui"
 import { storeToRefs } from "pinia"
 import { useSubmissionStore } from "oj/store/submission"
+import { useCodeStore } from "oj/store/code"
+import { isFlowchartPass, useFlowchartStore } from "oj/store/flowchart"
 import { SubmissionStatus } from "utils/constants"
 
 /**
@@ -14,7 +16,23 @@ const theme = useThemeVars()
 const { submission, syntaxErrorInfo, judging, pending, submitting } =
   storeToRefs(useSubmissionStore())
 
+const codeStore = useCodeStore()
+const flowchart = storeToRefs(useFlowchartStore())
+
+/** 画流程图时：评中转圈，A/S 打勾，B/C 是「还没过关」的 !，评失败是 ✕ */
+const flowchartStatus = computed(() => {
+  const phase = flowchart.phase.value
+  if (phase === "evaluating")
+    return { icon: "ph:circle-notch-bold", color: theme.value.infoColor, spin: true }
+  if (phase === "failed") return { icon: "ph:x-bold", color: theme.value.errorColor }
+  if (phase !== "done") return null
+  return isFlowchartPass(flowchart.latestRating.value.grade)
+    ? { icon: "ph:check-bold", color: theme.value.successColor }
+    : { icon: "ph:warning-bold", color: theme.value.warningColor }
+})
+
 const status = computed(() => {
+  if (codeStore.code.language === "Flowchart") return flowchartStatus.value
   if (syntaxErrorInfo.value) return { icon: "ph:warning-bold", color: theme.value.warningColor }
   if (!submission.value) return null
   if (judging.value || pending.value || submitting.value)
diff --git a/apps/web/src/oj/problem/components/SubmissionEffects.vue b/apps/web/src/oj/problem/components/SubmissionEffects.vue
index 20c9e28..4c4a7bd 100644
--- a/apps/web/src/oj/problem/components/SubmissionEffects.vue
+++ b/apps/web/src/oj/problem/components/SubmissionEffects.vue
@@ -6,6 +6,7 @@ import { useProblemStore } from "oj/store/problem"
 import { useSubmissionStore } from "oj/store/submission"
 import { useFireworks } from "oj/problem/composables/useFireworks"
 import { useLessonStore } from "oj/store/lesson"
+import { isFlowchartPass, useFlowchartStore } from "oj/store/flowchart"
 import { SubmissionStatus } from "utils/constants"
 import { useProblemPageContext } from "../composables/problemPageContext"
 
@@ -225,6 +226,19 @@ watch(
     }
   },
 )
+
+// ==================== 流程图画到 A / S ====================
+// 题单里流程图作业拿到 A 或 S 就算这道题做完（后端评分那一路已经记了账），
+// 和代码通过一样 1.5 秒后回题单（设计文档第 3 节决定 4、5）。只认这一页上刚评完的那一次
+const flowchartStore = useFlowchartStore()
+watch(
+  () => flowchartStore.evaluatedSeq,
+  () => {
+    if (!isFlowchartPass(flowchartStore.latestRating.grade)) return
+    celebrate()
+    if (ctx.value.backToProblemSet) goToProblemSetDelayed()
+  },
+)
 </script>
 
 <template>
diff --git a/apps/web/src/oj/problem/components/SubmitFlowchart.vue b/apps/web/src/oj/problem/components/SubmitFlowchart.vue
index fb39f52..cbb90f5 100644
--- a/apps/web/src/oj/problem/components/SubmitFlowchart.vue
+++ b/apps/web/src/oj/problem/components/SubmitFlowchart.vue
@@ -1,346 +1,54 @@
 <script lang="ts" setup>
 import { storeToRefs } from "pinia"
-
-// 工具函数
-import { decompressFromBase64, compressToBase64 } from "utils/functions"
-import { sortFlowchartCriteria } from "utils/constants"
-
-// 组合式函数
+import { compressToBase64 } from "utils/functions"
 import { useBreakpoints } from "shared/composables/breakpoints"
-import { useMermaid } from "shared/composables/useMermaid"
 import { useMermaidConverter } from "../composables/useMermaidConverter"
-
-// API 和状态管理
-import { getFlowchartSubmissionDetail } from "oj/api"
 import { useProblemStore } from "oj/store/problem"
-import { useFlowchartStore, type FlowchartRating } from "oj/store/flowchart"
+import { useFlowchartStore, type FlowchartEditorInstance } from "oj/store/flowchart"
 
 /**
- * 「提交流程图」按钮、分数按钮和评分详情弹框。
+ * 「交给 AI 点评」按钮。
  *
- * 提交和评分（WebSocket + 轮询 + 超时、评完的提示、A/S 亮到「我的流程图」）都在
- * flowchart store：这个组件只在语言选成 Flowchart 时才挂载，切去写代码它就卸载，
- * 正在评的那次不能跟着没。这里只剩从编辑器里取图、以及翻看历次评分的弹框。
+ * 提交和评分（WebSocket + 轮询 + 超时）在 flowchart store；评分结果、历次分数、载回旧版
+ * 都在左栏「结果」页签（FlowchartResult）。原来这里还有一个分数按钮和 1000px 的评分弹框，
+ * 弹框盖住画布，学生没法一边看改进建议一边改图。
  */
 
-// ==================== 类型定义 ====================
-interface Evaluation extends FlowchartRating {
-  feedback: string
-  suggestions: string
-  criteria_details: {
-    [key: string]: { score: number; max: number; comment: string }
-  }
-}
-
-interface FlowchartEditorInstance {
-  getFlowchartData: () => { nodes: unknown[]; edges: unknown[] }
-  setFlowchartData: (data: { nodes: unknown[]; edges: unknown[] }) => void
-}
-
-// 通过inject获取FlowchartEditor组件的引用
 const flowchartEditorRef = inject<Ref<FlowchartEditorInstance | null>>("flowchartEditorRef")
-const mermaidContainer = useTemplateRef<HTMLElement>("mermaidContainer")
 
-// 基础组合式函数
 const message = useMessage()
-const problemStore = useProblemStore()
-const { problem } = storeToRefs(problemStore)
+const { problem } = storeToRefs(useProblemStore())
 const flowchartStore = useFlowchartStore()
-const { loading, latestRating, submissionCount } = storeToRefs(flowchartStore)
+const { loading } = storeToRefs(flowchartStore)
 const { isDesktop } = useBreakpoints()
 const { convertToMermaid } = useMermaidConverter()
-const { renderError, renderFlowchart } = useMermaid()
 
-// 评分详情弹框
-const rendering = ref(false)
-const modalRating = ref<FlowchartRating>({ score: 0, grade: "" })
-const myFlowchartZippedStr = ref("")
-const myMermaidCode = ref("")
-const showDetailModal = ref(false)
-const evaluation = ref<Evaluation>({
-  score: 0,
-  grade: "",
-  feedback: "",
-  suggestions: "",
-  criteria_details: {},
-})
-const page = ref(1)
-const suggestionLines = computed(() => splitSuggestionLines(evaluation.value.suggestions))
-
-// jsonb 不保留键序，直接遍历会把 40 分的「逻辑正确性」排到最后
-const sortedCriteria = computed(() => sortFlowchartCriteria(evaluation.value.criteria_details))
-
-function splitSuggestionLines(suggestions?: string | null) {
-  return suggestions
-    ? suggestions
-        .split("\n")
-        .map((suggestion) => suggestion.trim())
-        .filter(Boolean)
-    : []
-}
-
-// ==================== 提交 ====================
 function submit() {
-  if (!flowchartEditorRef?.value) return
-
-  // 获取流程图的JSON数据
-  const flowchartData = flowchartEditorRef.value.getFlowchartData()
-
+  const editor = flowchartEditorRef?.value
+  if (!editor) return
+  const flowchartData = editor.getFlowchartData()
   if (!flowchartData?.nodes?.length || !flowchartData?.edges?.length) {
-    message.error("流程图节点或边不能为空")
+    message.warning("画布上还没有图：从左边拖几个框进来，再用线连起来")
     return
   }
-
   flowchartStore.submit(
     convertToMermaid(flowchartData),
     compressToBase64(JSON.stringify(flowchartData)),
   )
 }
 
-// ==================== 评分详情 ====================
-async function getSubmission(submissionPage = 0) {
-  const problemId = problem.value?.id
-  if (!problemId) return
-  const data = await getFlowchartSubmissionDetail(problemId, submissionPage)
-  if (problem.value?.id !== problemId) return
-  submissionCount.value = data.count
-  const submission = data.submission
-  // 翻到没有提交的页时后端返回 null（契约里 submission 是 nullable）——
-  // 原来的 any 让这里看起来非空，真翻到那一页会直接抛
-  if (!submission) {
-    myFlowchartZippedStr.value = ""
-    myMermaidCode.value = ""
-    modalRating.value = { score: 0, grade: "" }
-    evaluation.value = {
-      score: 0,
-      grade: "",
-      feedback: "",
-      suggestions: "",
-      criteria_details: {},
-    }
-    return
-  }
-  myFlowchartZippedStr.value = String(submission.flowchartData.data ?? "")
-  myMermaidCode.value = submission.mermaidCode || ""
-  modalRating.value = {
-    score: submission.aiScore ?? 0,
-    grade: submission.aiGrade ?? "",
-  }
-  evaluation.value = {
-    score: submission.aiScore ?? 0,
-    grade: submission.aiGrade ?? "",
-    feedback: submission.aiFeedback ?? "",
-    suggestions: submission.aiSuggestions ?? "",
-    criteria_details: submission.aiCriteriaDetails as Evaluation["criteria_details"],
-  }
-}
-
-// 请求失败时 rendering 必须复位：拦截器对普通业务错误是静默 reject 的，
-// 少了这个 finally，弹框就会永远停在转圈状态，也没有任何提示
-async function updatePage(val: number) {
-  page.value = val
-  rendering.value = true
-  try {
-    await getSubmission(val)
-    // 等待 DOM 更新
-    await nextTick()
-    await renderFlowchart(mermaidContainer.value, myMermaidCode.value)
-  } catch (error) {
-    message.error("加载这次提交失败，请稍后重试")
-    console.error("加载流程图提交失败:", error)
-  } finally {
-    rendering.value = false
-  }
-}
-
-async function openDetailModal() {
-  showDetailModal.value = true
-  rendering.value = true
-  try {
-    await getSubmission()
-    page.value = submissionCount.value
-    // 等待 DOM 更新，确保弹框已经渲染
-    await nextTick()
-    await renderFlowchart(mermaidContainer.value, myMermaidCode.value)
-  } catch (error) {
-    message.error("加载评分详情失败，请稍后重试")
-    console.error("加载流程图评分详情失败:", error)
-  } finally {
-    rendering.value = false
-  }
-}
-
-function closeModal() {
-  showDetailModal.value = false
-}
-
-function loadToEditor() {
-  if (myFlowchartZippedStr.value) {
-    // 老提交的压缩数据可能是坏的（格式换过、存了一半），
-    // 不兜住的话 decompressFromBase64/JSON.parse 直接抛，按钮点了毫无反应
-    try {
-      const json = JSON.parse(decompressFromBase64(myFlowchartZippedStr.value))
-      flowchartEditorRef?.value?.setFlowchartData({
-        nodes: json.nodes || [],
-        edges: json.edges || [],
-      })
-    } catch (error) {
-      message.error("这份流程图数据已损坏，无法加载到编辑器")
-      console.error("解析流程图数据失败:", error)
-      return
-    }
-  }
-  closeModal()
-}
-
-// ==================== 工具函数 ====================
-const getGradeType = (grade: string) => {
-  if (grade === "S") return "primary"
-  if (grade === "A") return "info"
-  if (grade === "B") return "warning"
-  return "error"
-}
-
-const getPercentType = (percent: number) => {
-  if (percent >= 0.8) return "primary"
-  else if (percent >= 0.6) return "info"
-  else if (percent >= 0.4) return "warning"
-  return "error"
-}
-
-// ==================== 生命周期 ====================
-onMounted(flowchartStore.ensureLoaded)
-
-// 换题：分数和还在评的那次由 store 自己收（见 flowchart store 的 reset），
-// 这里只关掉弹框，再拉新题的最近一次分数
-watch(
-  () => problem.value?.id,
-  (next, previous) => {
-    if (!previous || next === previous) return
-    showDetailModal.value = false
-    page.value = 1
-    flowchartStore.ensureLoaded()
-  },
-)
+// 进题、换题都读一次历次分数（store 里同一道题只读一次）
+watch(() => problem.value?.id, flowchartStore.ensureLoaded, { immediate: true })
 </script>
 
 <template>
-  <!-- 主要操作区域 -->
-  <n-flex align="center">
-    <!-- 提交按钮 -->
-    <n-button
-      :size="isDesktop ? 'medium' : 'small'"
-      type="primary"
-      :loading="loading"
-      :disabled="loading"
-      @click="submit"
-    >
-      {{ loading ? "AI 点评中..." : "提交流程图" }}
-    </n-button>
-
-    <!-- 评分结果按钮 -->
-    <n-button
-      secondary
-      v-if="latestRating.grade"
-      @click="openDetailModal"
-      :type="getGradeType(latestRating.grade)"
-    >
-      {{ latestRating.score }}分 {{ latestRating.grade }}级
-    </n-button>
-
-    <!-- 流程图评分详情模态框 -->
-    <n-modal v-model:show="showDetailModal" preset="card" style="width: 1000px">
-      <template #header>
-        <n-flex align="center">
-          <n-text>流程图评分详情</n-text>
-          <n-text :type="getGradeType(modalRating.grade)">
-            {{ modalRating.score }}分 {{ modalRating.grade }}级
-          </n-text>
-        </n-flex>
-      </template>
-      <n-grid :cols="5" :x-gap="16">
-        <!-- 左侧：流程图预览区域 -->
-        <n-gi :span="3">
-          <div class="flowchart">
-            <n-spin :show="rendering">
-              <n-alert v-if="renderError" type="error" title="流程图渲染失败">
-                {{ renderError }}
-              </n-alert>
-              <div class="flowchart" v-show="!renderError" ref="mermaidContainer"></div>
-            </n-spin>
-          </div>
-          <!-- 加载到编辑器按钮 -->
-          <n-flex style="margin-top: 16px" justify="center">
-            <n-button @click="loadToEditor" type="primary"> 加载到流程图编辑器 </n-button>
-          </n-flex>
-        </n-gi>
-
-        <!-- 右侧：评分详情区域 -->
-        <n-gi :span="2" style="max-height: 550px; overflow: auto">
-          <!-- AI反馈 -->
-          <n-card
-            v-if="evaluation.feedback"
-            size="small"
-            title="AI反馈"
-            style="margin-bottom: 16px"
-          >
-            <n-text>{{ evaluation.feedback }}</n-text>
-          </n-card>
-
-          <!-- 改进建议 -->
-          <n-card
-            v-if="suggestionLines.length"
-            size="small"
-            title="改进建议"
-            style="margin-bottom: 16px"
-          >
-            <n-flex vertical :size="6">
-              <n-text
-                v-for="(suggestion, index) in suggestionLines"
-                :key="`${index}-${suggestion}`"
-              >
-                {{ suggestion }}
-              </n-text>
-            </n-flex>
-          </n-card>
-
-          <!-- 详细评分 -->
-          <n-card v-if="sortedCriteria.length" size="small" title="详细评分">
-            <div v-for="[key, detail] in sortedCriteria" :key="key" style="margin-bottom: 12px">
-              <!-- 评分项标题和分数 -->
-              <n-flex justify="space-between" align="center" style="margin-bottom: 4px">
-                <n-text strong>{{ key }}</n-text>
-                <n-tag :type="getPercentType(detail.score / detail.max)" size="small" round>
-                  {{ detail.score || 0 }}分 / {{ detail.max }}分
-                </n-tag>
-              </n-flex>
-              <!-- 评分项详细说明 -->
-              <n-text v-if="detail.comment" depth="3" style="font-size: 12px">
-                {{ detail.comment }}
-              </n-text>
-            </div>
-          </n-card>
-        </n-gi>
-      </n-grid>
-
-      <!-- 分页组件 -->
-      <n-flex justify="center" style="margin-top: 24px" v-if="submissionCount > 1">
-        <n-pagination v-model:page="page" :page-count="submissionCount" @update-page="updatePage" />
-      </n-flex>
-    </n-modal>
-  </n-flex>
+  <n-button
+    :size="isDesktop ? 'medium' : 'small'"
+    type="primary"
+    :loading="loading"
+    :disabled="loading"
+    @click="submit"
+  >
+    {{ loading ? "AI 正在看…" : "交给 AI 点评" }}
+  </n-button>
 </template>
-<style scoped>
-/* ==================== 流程图样式 ==================== */
-.flowchart {
-  height: 500px;
-  display: flex;
-  justify-content: center;
-  align-items: center;
-}
-
-/* 确保 SVG 图表占满容器 */
-:deep(.flowchart > svg) {
-  height: 100%;
-}
-</style>
diff --git a/apps/web/src/oj/store/code.ts b/apps/web/src/oj/store/code.ts
index f0405b4..8ff80f6 100644
--- a/apps/web/src/oj/store/code.ts
+++ b/apps/web/src/oj/store/code.ts
@@ -8,13 +8,19 @@ import type { Code, LANGUAGE } from "utils/types"
  * 代码编辑器状态管理 Store
  * 管理全局的代码、输入、输出状态
  */
+function savedCodeLanguage() {
+  const saved = normalizeLanguage(storage.get(STORAGE_KEY.LANGUAGE))
+  return saved && saved !== "Flowchart" ? saved : null
+}
+
 export const useCodeStore = defineStore("code", () => {
   // ==================== 状态 ====================
   const code = reactive<Code>({
     value: "",
     // 过一道 normalizeLanguage：上线那一刻学生浏览器的 localStorage 里存的还是
     // 旧值 Python3，直接拿来用会被后端的契约挡掉（而且报错看不出是这个原因）
-    language: normalizeLanguage(storage.get(STORAGE_KEY.LANGUAGE)) ?? "Python",
+    // 以前的版本会把 Flowchart 也记成语言偏好，那种老值在这里一并不认
+    language: savedCodeLanguage() ?? "Python",
   })
 
   const input = ref("")
@@ -36,7 +42,6 @@ export const useCodeStore = defineStore("code", () => {
    */
   function setLanguage(language: LANGUAGE) {
     code.language = language
-    storage.set(STORAGE_KEY.LANGUAGE, language)
   }
 
   /**
@@ -69,14 +74,23 @@ export const useCodeStore = defineStore("code", () => {
     output.value = ""
   }
 
-  // 监听语言变化，保存到本地存储
+  /**
+   * 语言变了就记下来，下一道题默认用它。**不记 Flowchart**：那是「画流程图」这种作业，
+   * 不是编程语言偏好 —— 记下来的话，画完一道流程图题，下一道能画的题一打开就是画布，
+   * 而学生这节课要写的是代码。切回「写代码」时用的也是这份偏好。
+   */
   watch(
     () => code.language,
     (newLanguage) => {
-      storage.set(STORAGE_KEY.LANGUAGE, newLanguage)
+      if (newLanguage !== "Flowchart") storage.set(STORAGE_KEY.LANGUAGE, newLanguage)
     },
   )
 
+  /** 学生习惯用的编程语言（从来不会是 Flowchart） */
+  function preferredLanguage(): LANGUAGE {
+    return savedCodeLanguage() ?? "Python"
+  }
+
   return {
     // 状态
     code,
@@ -89,6 +103,7 @@ export const useCodeStore = defineStore("code", () => {
     // 操作
     setCode,
     setLanguage,
+    preferredLanguage,
     setInput,
     setOutput,
     reset,
diff --git a/apps/web/src/oj/store/flowchart.ts b/apps/web/src/oj/store/flowchart.ts
index 7467cc6..857be14 100644
--- a/apps/web/src/oj/store/flowchart.ts
+++ b/apps/web/src/oj/store/flowchart.ts
@@ -1,13 +1,11 @@
 import { createDiscreteApi } from "naive-ui"
 import { defineStore } from "pinia"
 import { errorCode } from "utils/api"
-import {
-  getCurrentProblemFlowchartSubmission,
-  getFlowchartSubmission,
-  getFlowchartSubmissionDetail,
-  submitFlowchart,
-} from "oj/api"
+import type { FlowchartScores, FlowchartSubmission } from "@oj2/contract"
+import type { Edge, Node } from "@vue-flow/core"
+import { getFlowchartScores, getFlowchartSubmission, submitFlowchart } from "oj/api"
 import { useProblemStore } from "oj/store/problem"
+import { useSubmissionStore } from "oj/store/submission"
 import { useFlowchartWebSocket, type FlowchartEvaluationUpdate } from "shared/composables/websocket"
 import { useMyFlowchartStore } from "shared/store/myFlowchart"
 import { useUserStore } from "shared/store/user"
@@ -22,18 +20,35 @@ export interface FlowchartRating {
 
 type Outcome = { ok: true; score: number; grade: string } | { ok: false; error?: string }
 
+/** 编辑器对外的两个方法（FlowchartEditor 的 defineExpose） */
+export interface FlowchartEditorInstance {
+  getFlowchartData: () => { nodes: unknown[]; edges: unknown[] }
+  setFlowchartData: (data: { nodes: Node[]; edges: Edge[] }) => void
+}
+
+/** 画到这两档算这道题做完（后端 flowchart/grade.ts 的 FLOWCHART_PASS_GRADES） */
+export function isFlowchartPass(grade: string | null | undefined) {
+  return grade === "S" || grade === "A"
+}
+
+/**
+ * 这一页上最近一次交的图走到哪了。idle = 这一页还没交过（历史分数照样有）
+ */
+export type FlowchartPhase = "idle" | "evaluating" | "done" | "failed"
+
 /** AI 评分比判题慢得多，轮询放缓一点 */
 const POLL_INTERVAL = 3000
 /** 到点还没结果就收手，别无限轮询下去 */
 const POLL_TIMEOUT = 3 * 60 * 1000
 
 /**
- * 流程图提交与 AI 评分：正在评的那一次、最近一次的分数、提交次数。
+ * 流程图提交与 AI 评分：正在评的那一次、历次分数、结果页签正在看哪一次。
  *
  * 原来都在提交按钮那个组件（SubmitFlowchart）里，而它只在语言选成 Flowchart 时才挂载 ——
  * 交完切到写代码，组件一卸载就断开 WebSocket，正在评的那次结果就丢了（评分中位 4.8 秒、
  * 最长二十几秒，足够学生切走）。挪到 store 之后，切走了照样评完、照样提示、照样把
- * A/S 的图亮到「我的流程图」。评分详情弹框只是展示，还留在组件里。
+ * A/S 的图亮到「你画的流程图」。评分详情从 1000px 的弹框挪进了左栏「结果」页签
+ * （FlowchartResult），那边只管显示。
  */
 export const useFlowchartStore = defineStore("flowchart", () => {
   const problemStore = useProblemStore()
@@ -41,7 +56,27 @@ export const useFlowchartStore = defineStore("flowchart", () => {
 
   const loading = ref(false)
   const latestRating = ref<FlowchartRating>({ score: 0, grade: "" })
-  const submissionCount = ref(0)
+  const phase = ref<FlowchartPhase>("idle")
+  /**
+   * 这一页上**刚刚**评完一次就加一（不含进页面时读到的历史）。课堂条标做完、题单里
+   * 1.5 秒后跳回题单，都只认这一下 —— 进页面读到以前的 A 不该再跳一次
+   */
+  const evaluatedSeq = ref(0)
+  /** 这道题评完的每一次，早的在前；hidden = 加入题单之前、被藏起来的次数 */
+  const scores = ref<FlowchartScores["scores"]>([])
+  const hiddenCount = ref(0)
+  const submissionCount = computed(() => scores.value.length)
+  /** 结果页签正在看哪一次；null = 最新那次 */
+  const selectedId = ref<string | null>(null)
+  /** 看过的几次的完整评语，按 id 缓存（评完就不会再变） */
+  const details = ref<Record<string, FlowchartSubmission>>({})
+  const detailLoading = ref(false)
+
+  /** 画布本身。结果页签要拿当前画布和某一版比一比、把那一版载回去 */
+  const editor = shallowRef<FlowchartEditorInstance | null>(null)
+  function attachEditor(instance: FlowchartEditorInstance | null) {
+    editor.value = instance
+  }
   /** 最近一次交上去的那张图，评到 A/S 时亮到「我的流程图」 */
   const lastSubmittedMermaidCode = ref("")
   /** 最近一次分数是哪道题的；换题后置空，下次挂载时重新拉 */
@@ -122,16 +157,21 @@ export const useFlowchartStore = defineStore("flowchart", () => {
     scheduleDisconnect(15 * 60 * 1000)
 
     if (!outcome.ok) {
-      message.error(
-        outcome.error ? `流程图评分失败: ${outcome.error}` : "流程图评分失败，请稍后重试",
-      )
+      phase.value = "failed"
+      message.error("AI 这次没评出来，再交一次试试")
       return
     }
     latestRating.value = { score: outcome.score, grade: outcome.grade }
-    message.success(`流程图评分完成！得分: ${outcome.score}分 (${outcome.grade}级)`)
-    if ((outcome.grade === "A" || outcome.grade === "S") && lastSubmittedMermaidCode.value) {
+    phase.value = "done"
+    selectedId.value = null
+    evaluatedSeq.value++
+    // 学生这会儿可能在看题目：只提示一句、页签上亮状态，不强行把他切过去（设计文档 5.3）
+    message.success(`AI 点评完了：${outcome.score} 分，${outcome.grade} 级`)
+    if (isFlowchartPass(outcome.grade) && lastSubmittedMermaidCode.value) {
       myFlowchartStore.show(lastSubmittedMermaidCode.value)
     }
+    const problemId = problemStore.problem?.id
+    if (problemId) loadScores(problemId)
   }
 
   const handleWebSocketMessage = (data: FlowchartEvaluationUpdate) => {
@@ -162,7 +202,9 @@ export const useFlowchartStore = defineStore("flowchart", () => {
 
     lastSubmittedMermaidCode.value = mermaidCode
     loading.value = true
-    latestRating.value = { score: 0, grade: "" }
+    phase.value = "evaluating"
+    // 交上去就切到「结果」页签，和交代码一样
+    useSubmissionStore().revealResult()
 
     try {
       const response = await submitFlowchart({
@@ -183,10 +225,9 @@ export const useFlowchartStore = defineStore("flowchart", () => {
         startPollingFallback()
         startPollingDeadline()
       }
-
-      message.success("流程图已提交，请耐心等待评分")
     } catch (error) {
       loading.value = false
+      phase.value = "idle"
       // 按错误码分支（见 utils/api.ts 的约定）。限流是最容易撞上的一种：
       // 只说「提交失败」的话，学生会以为是自己的图有问题，然后反复点，越点越久
       if (errorCode(error) === "too-many-submissions") {
@@ -200,32 +241,64 @@ export const useFlowchartStore = defineStore("flowchart", () => {
     }
   }
 
-  // ==================== 最近一次的分数 ====================
+  // ==================== 历次分数 ====================
+  async function loadScores(problemId: number) {
+    const data = await getFlowchartScores(problemId)
+    // 换题之后才回来的是上一道题的，别盖掉新题的
+    if (problemStore.problem?.id !== problemId) return
+    scores.value = data.scores
+    hiddenCount.value = data.hidden
+    const latest = data.scores.at(-1)
+    if (latest && phase.value !== "evaluating") {
+      latestRating.value = { score: latest.score, grade: latest.grade }
+    }
+  }
+
   /**
-   * 这道题最近一次的分数；画到了 A/S 的话，把那张图亮到「我的流程图」。
-   * 同一道题只拉一次（组件卸了再挂回来不重拉），换题后 loadedFor 清空。
+   * 进这道题时读一次历次分数；画到过 A/S 的话，把最近那张 A/S 亮到题面的「你画的流程图」。
+   * 同一道题只读一次（组件卸了再挂回来不重读），换题后 loadedFor 清空。
    */
   async function ensureLoaded() {
     const problemId = problemStore.problem?.id
     if (!problemId || loadedFor === problemId) return
     loadedFor = problemId
     try {
-      const data = await getCurrentProblemFlowchartSubmission(problemId)
-      // 换题之后才回来的是上一道题的分数，别盖掉新题的
+      await loadScores(problemId)
       if (problemStore.problem?.id !== problemId) return
-      submissionCount.value = data.count
-      latestRating.value = { score: data.score, grade: data.grade }
-      const grade = data.grade
-      if ((grade === "A" || grade === "S") && data.count > 0) {
-        const detail = await getFlowchartSubmissionDetail(problemId, data.count)
+      const passed = scores.value.filter((row) => isFlowchartPass(row.grade)).at(-1)
+      if (passed) {
+        const detail = await fetchDetail(passed.id)
         // 查的这一会儿又换了题，这张图不是新题的
         if (problemStore.problem?.id !== problemId) return
-        if (detail.submission?.mermaidCode) myFlowchartStore.show(detail.submission.mermaidCode)
+        if (detail?.mermaidCode) myFlowchartStore.show(detail.mermaidCode)
       }
     } catch (error) {
       // 拉不到就当没交过；下次挂载再试
       loadedFor = null
-      console.error("[Flowchart] 读取最近一次评分失败:", error)
+      console.error("[Flowchart] 读取历次评分失败:", error)
+    }
+  }
+
+  async function fetchDetail(id: string) {
+    const cached = details.value[id]
+    if (cached) return cached
+    const detail = await getFlowchartSubmission(id)
+    details.value[id] = detail
+    return detail
+  }
+
+  /** 结果页签里点了某一次（null = 回到最新）：把那次的完整评语拉回来 */
+  async function select(id: string | null) {
+    selectedId.value = id
+    const target = id ?? scores.value.at(-1)?.id
+    if (!target || details.value[target]) return
+    detailLoading.value = true
+    try {
+      await fetchDetail(target)
+    } catch {
+      message.error("这一次的评语读不出来，过一会儿再试")
+    } finally {
+      detailLoading.value = false
     }
   }
 
@@ -237,7 +310,11 @@ export const useFlowchartStore = defineStore("flowchart", () => {
   function reset() {
     stopMonitoring()
     latestRating.value = { score: 0, grade: "" }
-    submissionCount.value = 0
+    phase.value = "idle"
+    scores.value = []
+    hiddenCount.value = 0
+    selectedId.value = null
+    details.value = {}
     lastSubmittedMermaidCode.value = ""
     loadedFor = null
   }
@@ -253,9 +330,19 @@ export const useFlowchartStore = defineStore("flowchart", () => {
   return {
     loading,
     latestRating,
+    phase,
+    evaluatedSeq,
+    scores,
+    hiddenCount,
     submissionCount,
+    selectedId,
+    details,
+    detailLoading,
+    editor,
+    attachEditor,
     submit,
     ensureLoaded,
+    select,
     reset,
   }
 })
diff --git a/apps/web/src/oj/store/problem.ts b/apps/web/src/oj/store/problem.ts
index 88c28e0..6e132db 100644
--- a/apps/web/src/oj/store/problem.ts
+++ b/apps/web/src/oj/store/problem.ts
@@ -1,5 +1,7 @@
 import { defineStore } from "pinia"
+import { useBreakpoints } from "shared/composables/breakpoints"
 import type { LANGUAGE, ProblemDetail } from "utils/types"
+import { problemEntryOf } from "oj/problem/composables/problemPageContext"
 
 export const useProblemStore = defineStore("problem", () => {
   const problem = ref<ProblemDetail | null>(null)
@@ -25,20 +27,46 @@ export const useProblemStore = defineStore("problem", () => {
    */
   const failCount = computed(() => (problem.value?.myFailedCount ?? 0) + sessionFailCount.value)
 
+  const { isDesktop } = useBreakpoints()
+
+  /**
+   * 这道题能不能画流程图。比赛里不给（入口差异表）；题单里原来也不给，现在给了，画到 A/S
+   * 算完成（设计文档第 3 节决定 4）。手机上不给：节点库是 HTML5 拖放，手机上拖不动。
+   */
+  const canDraw = computed(
+    () => !!problem.value?.allowFlowchart && problemEntryOf(route) !== "contest" && isDesktop.value,
+  )
+
+  /**
+   * 编辑器能切到的全部「语言」。流程图在内部仍是一门叫 Flowchart 的语言（提交、草稿、编辑器
+   * 切换都按它分支），但界面上不再混在语言下拉里 —— 那是工具栏最左边的「写代码 / 画流程图」。
+   */
   const languages = computed<LANGUAGE[]>(() => {
-    if (route.name === "problem" && problem.value?.allowFlowchart) {
-      return ["Flowchart", ...problem.value.languages]
-    }
-    return problem.value?.languages ?? []
+    const own = problem.value?.languages ?? []
+    return canDraw.value ? ["Flowchart", ...own] : own
   })
 
+  /** 语言下拉里的：真正的编程语言 */
+  const codeLanguages = computed(() => languages.value.filter((it) => it !== "Flowchart"))
+
+  /**
+   * 请编辑器换语言，并读回那门语言的草稿（没有就是模板）。只有编辑器（ProblemEditor）知道
+   * 草稿存在哪，所以这里只留一个请求，由它接住 —— 结果页签里的「照着它写代码」不在编辑器里。
+   */
+  const languageRequest = ref<LANGUAGE | null>(null)
+  function switchLanguage(language: LANGUAGE) {
+    languageRequest.value = language
+  }
+
   /**
    * 这道题收不收当前语言，不收就退到它支持的第一种（SQL 题只有 "SQL"，硬编码的
    * Python 会被后端拒绝）。编辑器每次载入代码之前都要过一遍：草稿的键里带着语言，
    * 先载入再改语言，编辑器里摆的就是另一种语言的模板。
    */
   function supportedLanguage(current: LANGUAGE): LANGUAGE {
-    return languages.value.includes(current) ? current : (languages.value[0] ?? "Python")
+    if (languages.value.includes(current)) return current
+    // 兜底落在编程语言上：languages 的第一项可能是 Flowchart，不能默认把人扔到画布上
+    return codeLanguages.value[0] ?? languages.value[0] ?? "Python"
   }
 
   function incrementFailCount() {
@@ -55,7 +83,11 @@ export const useProblemStore = defineStore("problem", () => {
   return {
     problem,
     failCount,
+    canDraw,
     languages,
+    codeLanguages,
+    languageRequest,
+    switchLanguage,
     supportedLanguage,
     incrementFailCount,
   }
diff --git a/apps/web/src/shared/components/FlowchartEditor/index.vue b/apps/web/src/shared/components/FlowchartEditor/index.vue
index 45ec4c0..975754a 100644
--- a/apps/web/src/shared/components/FlowchartEditor/index.vue
+++ b/apps/web/src/shared/components/FlowchartEditor/index.vue
@@ -24,7 +24,22 @@ interface Props {
 const { height = "calc(100vh - 133px)" } = defineProps<Props>()
 
 // Vue Flow 实例
-const { addEdges, removeNodes, removeEdges } = useVueFlow()
+const { addEdges, removeNodes, removeEdges, fitView, onNodesInitialized } = useVueFlow()
+
+/**
+ * 进页面读回草稿、载回旧版之后缩放到正好装下整张图。原来画布停在默认的位置和缩放上，
+ * 图大一点就有一半在外面，学生以为草稿丢了。只在这两种时候缩：拖进一个新节点时也会
+ * 触发 nodes-initialized，那时候跟着缩放会让画布跳一下
+ */
+let fitPending = false
+function fitSoon() {
+  fitPending = true
+}
+onNodesInitialized(() => {
+  if (!fitPending) return
+  fitPending = false
+  fitView({ padding: 0.15, maxZoom: 1.2 })
+})
 
 // 节点和边的响应式数据
 // 显式标注成 Ref<Node[]>，不走 ref<T>() 的 UnwrapRef 推导 —— vue-flow 的
@@ -37,12 +52,18 @@ const { canUndo, canRedo, resetHistory, saveState, undo, redo } = useHistory()
 
 const problemStore = useProblemStore()
 const { problem } = storeToRefs(problemStore)
-// 缓存管理：用 computed key 支持题目 ID 异步加载后自动切换到正确的 storage
-const cacheKey = computed(() =>
-  problem.value?._id
-    ? `flowchart-editor-data-problem-${problem.value!._id}`
-    : "flowchart-editor-data",
-)
+const route = useRoute()
+/**
+ * 缓存管理：用 computed key 支持题目 ID 异步加载后自动切换到正确的 storage。
+ * 题单入口单独一份草稿：加入题单之前在题库里画到 A 的那张，不能一进题单就摆在画布上
+ * （设计文档第 11 节 3，和代码草稿同理）
+ */
+const cacheKey = computed(() => {
+  if (!problem.value?._id) return "flowchart-editor-data"
+  const base = `flowchart-editor-data-problem-${problem.value._id}`
+  const problemSetId = route.params.problemSetId
+  return problemSetId ? `${base}-problemset-${problemSetId}` : base
+})
 const { isSaving, lastSaved, hasUnsavedChanges, saveToCache, loadFromCache, clearCache } = useCache(
   nodes,
   edges,
@@ -51,6 +72,7 @@ const { isSaving, lastSaved, hasUnsavedChanges, saveToCache, loadFromCache, clea
     // 换题后画布已经被换成新题的草稿，历史必须跟着重建，
     // 否则一次撤销就会把上一题的图还原到这一题里
     resetHistory(nodes.value, edges.value)
+    fitSoon()
   },
 )
 
@@ -152,6 +174,7 @@ onMounted(() => {
   // 否则第一步操作没有可回退的目标，撤销按钮一直是灰的
   loadFromCache()
   resetHistory(nodes.value, edges.value)
+  fitSoon()
 })
 
 onUnmounted(() => {
@@ -177,6 +200,7 @@ const setFlowchartData = (data: { nodes: Node[]; edges: Edge[] }) => {
     nodes.value = processedNodes
     edges.value = processedEdges
     saveState(nodes.value, edges.value)
+    fitSoon()
   }
 }
 
diff --git a/docs/specs/2026-09-28-problem-page-redesign-design.md b/docs/specs/2026-09-28-problem-page-redesign-design.md
index d0f0fa4..0946955 100644
--- a/docs/specs/2026-09-28-problem-page-redesign-design.md
+++ b/docs/specs/2026-09-28-problem-page-redesign-design.md
@@ -1,7 +1,7 @@
 # OJ2 设计文档：题目详情页重设计
 
 日期：2026-09-28
-状态：**设计已定，第 14 节的 1–4 步和第 7 节已实施**。效果图在 claude.ai 的私有画布「题目详情页重设计」里（「机房（方案 B）」
+状态：**设计已定，第 14 节的 1–5 步和第 7 节已实施**（剩第 6 步：协作的几处规则、手机版）。效果图在 claude.ai 的私有画布「题目详情页重设计」里（「机房（方案 B）」
 那一页是定稿，「初稿 1280×800」那一页已作废，只留着对照）。效果图表达不了的规则都写在这里，
 **实施时以本文为准**。