iterate-plugin 3.15.0 → 3.16.0

package/README.md

@@ -107,7 +107,7 @@ Both modes share:
 ### UI layer (build-free client slots, v3.12: 12 tabs, bilingual)
 
 - **ConvergenceDashboard** (`conversation.input.dock`) — live round progress bar, severity stats, dimension badges, trend mini-chart above the input; normal mode also shows fix-count badges; plus a live workflow-phase chip (current phase + running/stopped) and a **task_mode indicator (code / iterate)**, now backed by the persisted transcript (`iterate_status` / `iterate_transcript` emit `taskMode`). While a run is active the dashboard switches to a running state (round + phase) instead of the static empty state, and the start buttons are hidden.
-- **ObservatoryPanel** (`conversation.input.dock`) — a **twelve-tab** runtime observatory below the input: **live activity stream** (default tab, type filter) + **F1** review threads (expand / collapse all), **F2** convergence trend (incl. the per-round comparison table: findings raised / fixed / severity mix / validation outcomes), **F3** finding locations (severity / dimension / search filter), **F4** fixes + rollback, **F5** checkpoint resume, **F6** run console, **F7** decision timeline (type / round filter + search), **F8** Quality Gate, **F9** Experience Bank, **F10** Defense Events, **F11** Preflight. Number-key shortcuts: `1`–`9` jump to F1–F9, `0` to F10, `-` back to the live stream (F11 is click-only). **F8/F9/F10 are live** — they scan the session stream for the latest `iterate_quality_gate` / `iterate_experience` / `iterate_defense_events` results and render real data (PASS/FAIL chip + per-dimension score bars, experience entries with adopt buttons + client-side search, defense type-count chips + filtered event stream); when the session stream has none they distinguish "no data on disk" from "not pulled yet" and offer a **拉取磁盘快照 (pull disk snapshot)** instruction that has the model call the read-side tools and echo the `.iterate/` artifacts back into the stream. One-click JSON export of all observatory data — manifest + live + `qualityGate` / `experienceBank` / `defenseEvents` / `report` (download, copy fallback). **v3.9: report-export buttons** (导出报告 / 导出 SARIF, **v3.11 + 导出 JSON 报告**) render the *same shared data model* as the `iterate_report` tool — a PR-pasteable Markdown run report, a SARIF 2.1.0 log for CI / GitHub Code Scanning, and (v3.11) the complete normalized run model as machine-readable JSON. **v3.5.3: reset actions surfaced** — F5 copies an `iterate_checkpoint clear` instruction, F8 an `iterate_quality_gate clear` instruction, F10 an `iterate_defense_events clear` instruction. **v3.12: pre-run readiness (Preflight)** — F11 renders the latest `iterate_preflight` checklist scanned from the session stream (validation commands configured / git repo / dirty tree / stale FAIL gate / interrupted checkpoint), each warn/fail row offering the exact fix instruction (configure `validation.commands`, clear the stale gate, resume the checkpoint) so the environment is repaired BEFORE a round is spent on it; the empty state offers a **运行预检** instruction that has the model call `iterate_preflight`. **v3.13: the blocking validation row auto-detects commands from the project manifest and offers a one-click 「应用建议」 (copies an `iterate_config write` payload).**
+- **ObservatoryPanel** (`conversation.input.dock`) — a **twelve-tab** runtime observatory below the input: **live activity stream** (default tab, type filter) + **F1** review threads (expand / collapse all), **F2** convergence trend (incl. the per-round comparison table: findings raised / fixed / severity mix / validation outcomes, plus **v3.16 a cross-session comparison**: save the current run as a baseline (`存为对比基准`), then compare any later run against it side by side (`与基准对比`) with per-metric deltas tinted by direction), **F3** finding locations (severity / dimension / search filter), **F4** fixes + rollback, **F5** checkpoint resume, **F6** run console, **F7** decision timeline (type / round filter + search), **F8** Quality Gate, **F9** Experience Bank, **F10** Defense Events, **F11** Preflight. Number-key shortcuts: `1`–`9` jump to F1–F9, `0` to F10, `-` back to the live stream (F11 is click-only). **F8/F9/F10 are live** — they scan the session stream for the latest `iterate_quality_gate` / `iterate_experience` / `iterate_defense_events` results and render real data (PASS/FAIL chip + per-dimension score bars, experience entries with adopt buttons + client-side search, defense type-count chips + filtered event stream); when the session stream has none they distinguish "no data on disk" from "not pulled yet" and offer a **拉取磁盘快照 (pull disk snapshot)** instruction that has the model call the read-side tools and echo the `.iterate/` artifacts back into the stream. One-click JSON export of all observatory data — manifest + live + `qualityGate` / `experienceBank` / `defenseEvents` / `report` (download, copy fallback). **v3.9: report-export buttons** (导出报告 / 导出 SARIF, **v3.11 + 导出 JSON 报告**) render the *same shared data model* as the `iterate_report` tool — a PR-pasteable Markdown run report, a SARIF 2.1.0 log for CI / GitHub Code Scanning, and (v3.11) the complete normalized run model as machine-readable JSON. **v3.5.3: reset actions surfaced** — F5 copies an `iterate_checkpoint clear` instruction, F8 an `iterate_quality_gate clear` instruction, F10 an `iterate_defense_events clear` instruction. **v3.12: pre-run readiness (Preflight)** — F11 renders the latest `iterate_preflight` checklist scanned from the session stream (validation commands configured / git repo / dirty tree / stale FAIL gate / interrupted checkpoint), each warn/fail row offering the exact fix instruction (configure `validation.commands`, clear the stale gate, resume the checkpoint) so the environment is repaired BEFORE a round is spent on it; the empty state offers a **运行预检** instruction that has the model call `iterate_preflight`. **v3.13: the blocking validation row auto-detects commands from the project manifest and offers a one-click 「应用建议」 (copies an `iterate_config write` payload).**
 - **TriagePanel** (`conversation.chat.turnTail`) — per-finding triage with `y` (keep → fix), `n` (skip), `a` (known-intentional) keyboard shortcuts and `↑`/`↓` navigation (focus-guarded: only active while the panel has focus), filtering, batch ops (incl. select-all), localStorage persistence. **Write-back closed loop**: the primary button copies an `iterate_triage apply` instruction for the marked entries (the tool validates, dedupes, backs up, rolls back on failure); a secondary button copies the raw `known_intentional` YAML; **回读已写入条目** copies an `iterate_triage list` instruction to confirm what actually landed. Copy payloads are aligned to the target tool schemas. **v3.2: native command buttons** (approve architectural fix, trigger new round, rollback to checkpoint). **v3.3: §8 指派修复 (assign)** — one click copies an `iterate_fix` instruction carrying every in-scope finding (respecting the batch toggle).
 - **StatsCard** (`conversation.chat.turnTail`) — convergence stats, round history, trend chart and completion summary when no findings remain.
 - **iterate theme skin** (`theme.overrideTokens`) — a warm-amber 13-token `--dsw-*` override, light/dark modes, togglable in settings.
@@ -202,6 +202,8 @@ The observatory below the input carries **12 tabs**: the live activity stream (d
 
 `F1` review threads · `F2` convergence trend (incl. per-round comparison) · `F3` finding locations · `F4` fixes + rollback · `F5` checkpoint resume · `F6` run console · `F7` decision timeline · `F8` quality gate · `F9` experience bank · `F10` defense events · `F11` preflight
 
+**Cross-session comparison (v3.16)** — the F2 tab also carries a **跨会话对比 (cross-session comparison)** block: **存为对比基准** saves the current run's summary (rounds / findings / critical+high / fixes applied / failed fixes / validation pass rate / quality-gate score, persisted to `localStorage`) and **与基准对比** renders the next run against it side by side with a directional delta (better / worse tinted). This replaces the old "export two JSONs and diff by hand" chore; a corrupt / stale saved baseline reads as "no baseline", never a phantom run.
+
 Press `1`–`9` to jump to F1–F9, `0` for F10, and `-` to return to the live stream.
 
 ### Disk snapshot pull & export

package/README.zh-CN.md

@@ -106,7 +106,7 @@ dsh plugin --profile web add iterate-plugin
 ### UI 层（客户端免构建槽位，v3.12：12 个页签，双语）
 
 - **收敛看板 `ConvergenceDashboard`**（`conversation.input.dock`）— 输入框上方实时显示轮次进度条、严重度统计、维度徽章、趋势迷你图；normal 模式另显示修复计数徽章；还有运行阶段芯片（当前工作流阶段 + 运行中/已结束）与 **task_mode 指示器（code / iterate）**。运行期间看板切换为"运行中"状态（显示轮次与阶段），并隐藏启动按钮。
-- **运行时观测台 `ObservatoryPanel`**（`conversation.input.dock`）— 输入框下方**十二个页签**运行时观测台：**实时活动流**（默认页，支持按活动类型筛选）+ **F1** 审查线程（全部展开/全部收起）、**F2** 收敛趋势（含跨轮对比表：每轮 findings / fixed / 严重度分布 / 验证结果）、**F3** 发现定位（按严重度/维度/关键词筛选）、**F4** 修复与回滚、**F5** 断点恢复、**F6** 运行控制台、**F7** 决策时间线（按类型/轮次筛选与关键词搜索）、**F8** 质量门禁、**F9** 经验银行、**F10** 防御事件、**F11** 开跑预检。数字键快捷键：`1`–`9` 跳到 F1–F9，`0` 到 F10，`-` 回实时活动流（F11 仅点击进入）。**F8/F9/F10 为实时数据**——扫描会话流中最近一次 `iterate_quality_gate` / `iterate_experience` / `iterate_defense_events` 结果并渲染真实数据（PASS/FAIL 徽章 + 分维度评分条、经验条目（含采纳按钮 + 客户端搜索）、防御类型计数芯片 + 筛选后的事件流）；会话流中没有数据时区分"磁盘无数据"与"尚未拉取"，并提供**拉取磁盘快照**指令（让模型调用只读工具、把 `.iterate/` 产物回显进会话流）。一键导出全部观测数据为 JSON——manifest + live + `qualityGate` / `experienceBank` / `defenseEvents` / `report`（优先下载，失败回退复制）。**v3.9 新增报告导出按钮**（导出报告 / 导出 SARIF，**v3.11 新增导出 JSON 报告**），与 `iterate_report` 工具共用同一份数据模型——分别产出可直接粘贴进 PR 的 Markdown 运行报告、供 CI / GitHub Code Scanning 消费的 SARIF 2.1.0 日志，以及（v3.11）完整的归一化运行模型 JSON（无损格式）。**v3.5.3 新增重置操作**：F5 复制 `iterate_checkpoint clear`、F8 复制 `iterate_quality_gate clear`、F10 复制 `iterate_defense_events clear` 指令。**v3.12 新增开跑预检（Preflight）**——F11 渲染会话流中最近一次 `iterate_preflight` 清单（验证命令是否配置 / 是否 Git 仓库 / 脏工作区 / 陈旧 FAIL 门禁 / 中断断点），每个 warn/fail 行都给出精确的修复指令（配置 `validation.commands`、清除陈旧门禁、恢复断点），让环境在**花掉一轮之前**就被修好；空态提供**运行预检**指令（让模型调用 `iterate_preflight`）。**v3.13：阻断的验证命令行自动检测项目 manifest，提供一键「应用建议」**（复制 `iterate_config` 的 `operation:"write"` 指令，携带完整 `validation.commands.default` 载荷）。
+- **运行时观测台 `ObservatoryPanel`**（`conversation.input.dock`）— 输入框下方**十二个页签**运行时观测台：**实时活动流**（默认页，支持按活动类型筛选）+ **F1** 审查线程（全部展开/全部收起）、**F2** 收敛趋势（含跨轮对比表：每轮 findings / fixed / 严重度分布 / 验证结果；**v3.16 新增跨会话对比**：把当前运行**存为对比基准**，下次运行一键**与基准对比**，逐项并排 + 带方向的差值着色）、**F3** 发现定位（按严重度/维度/关键词筛选）、**F4** 修复与回滚、**F5** 断点恢复、**F6** 运行控制台、**F7** 决策时间线（按类型/轮次筛选与关键词搜索）、**F8** 质量门禁、**F9** 经验银行、**F10** 防御事件、**F11** 开跑预检。数字键快捷键：`1`–`9` 跳到 F1–F9，`0` 到 F10，`-` 回实时活动流（F11 仅点击进入）。**F8/F9/F10 为实时数据**——扫描会话流中最近一次 `iterate_quality_gate` / `iterate_experience` / `iterate_defense_events` 结果并渲染真实数据（PASS/FAIL 徽章 + 分维度评分条、经验条目（含采纳按钮 + 客户端搜索）、防御类型计数芯片 + 筛选后的事件流）；会话流中没有数据时区分"磁盘无数据"与"尚未拉取"，并提供**拉取磁盘快照**指令（让模型调用只读工具、把 `.iterate/` 产物回显进会话流）。一键导出全部观测数据为 JSON——manifest + live + `qualityGate` / `experienceBank` / `defenseEvents` / `report`（优先下载，失败回退复制）。**v3.9 新增报告导出按钮**（导出报告 / 导出 SARIF，**v3.11 新增导出 JSON 报告**），与 `iterate_report` 工具共用同一份数据模型——分别产出可直接粘贴进 PR 的 Markdown 运行报告、供 CI / GitHub Code Scanning 消费的 SARIF 2.1.0 日志，以及（v3.11）完整的归一化运行模型 JSON（无损格式）。**v3.5.3 新增重置操作**：F5 复制 `iterate_checkpoint clear`、F8 复制 `iterate_quality_gate clear`、F10 复制 `iterate_defense_events clear` 指令。**v3.12 新增开跑预检（Preflight）**——F11 渲染会话流中最近一次 `iterate_preflight` 清单（验证命令是否配置 / 是否 Git 仓库 / 脏工作区 / 陈旧 FAIL 门禁 / 中断断点），每个 warn/fail 行都给出精确的修复指令（配置 `validation.commands`、清除陈旧门禁、恢复断点），让环境在**花掉一轮之前**就被修好；空态提供**运行预检**指令（让模型调用 `iterate_preflight`）。**v3.13：阻断的验证命令行自动检测项目 manifest，提供一键「应用建议」**（复制 `iterate_config` 的 `operation:"write"` 指令，携带完整 `validation.commands.default` 载荷）。
 - **Findings 分诊面板 `TriagePanel`**（`conversation.chat.turnTail`）— 逐条分诊，键盘快捷键 `y`（保留→修复）、`n`（跳过）、`a`（已知有意）与 `↑`/`↓` 移动（焦点守卫：仅面板内焦点时生效），支持筛选、批量（含一键全选所有 findings）、localStorage 持久化。**写回闭环**：主按钮复制已标记条目的 `iterate_triage apply` 指令（工具负责校验、去重、备份、失败回滚）；次按钮复制裸 `known_intentional` YAML；**回读已写入条目**复制 `iterate_triage list` 指令核对实际落盘内容。复制载荷与目标工具 schema 对齐。**v3.2：原生命令按钮**（批准架构修复、触发新一轮、回滚到断点）。
 - **收敛统计卡片 `StatsCard`**（`conversation.chat.turnTail`）— 无 findings 时显示收敛统计、历史轮次、趋势图、完成摘要。
 - **iterate 主题皮肤**（`theme.overrideTokens`）— 暖琥珀配色的 13 个 `--dsw-*` token 覆盖，明暗双模式，可在设置页开关。
@@ -201,6 +201,8 @@ pnpm add /path/to/iterate-skill/harness/iterate-plugin
 
 `F1` 审查线程 · `F2` 收敛趋势（含跨轮对比）· `F3` 发现定位 · `F4` 修复与回滚 · `F5` 断点恢复 · `F6` 运行控制台 · `F7` 决策时间线 · `F8` 质量门禁 · `F9` 经验银行 · `F10` 防御事件 · `F11` 开跑预检
 
+**跨会话对比（v3.16）**——F2 页签还带一个**跨会话对比**块：**存为对比基准**把当前运行摘要（轮次 / 发现 / critical+high / 修复成功 / 修复失败 / 验证通过率 / 门禁得分，持久化到 `localStorage`）保存下来，**与基准对比**把下一次运行与它并排对比，差值按「更好 / 更差」方向着色。这取代了以前「导出两份 JSON 手工并排对比」的繁琐做法；损坏 / 过期的已存基准读作「无基准」，绝不变成幽灵运行。
+
 按 `1`–`9` 跳到 F1–F9，`0` 到 F10，`-` 回实时活动流。
 
 ### 磁盘快照拉取与导出

package/lib/client.js

@@ -1,4 +1,4 @@
-/* iterate-plugin client source-digest: 4f3bcac77cef83b80ea8fa2387f254121b9f94f7a2a2ca9422dd6342581c3463 */
+/* iterate-plugin client source-digest: 55b24ad8a8a5d8abf400d42314a6bf75394e8e529fc587ac991bb4e50d868393 */
 window.__ModuleLoader__.load({ id: "iterate-plugin", factory: (require) => {
 var module = { exports: {} };
 var exports = module.exports;
@@ -2645,7 +2645,24 @@ var EN = {
   "\u5404\u8F6E\u53D1\u73B0\u6570\u91CF\uFF1A{a0}{a1}": "Findings per round: {a0}{a1}",
   " \xB7 \u5F53\u524D Round {a0}": " \xB7 current round {a0}",
   "\u8DE8\u8F6E\u5BF9\u6BD4\uFF08\u672C\u4F1A\u8BDD\uFF09": "Cross-round comparison (this session)",
-  "\u8DE8\u4F1A\u8BDD\u5BF9\u6BD4\uFF1A\u672C\u8868\u53EA\u8986\u76D6\u5F53\u524D\u4F1A\u8BDD\uFF1B\u70B9\u300C\u62C9\u53D6\u78C1\u76D8\u5FEB\u7167\u300D\u8BA9\u6A21\u578B\u56DE\u663E iterate_history\uFF0C\u518D\u7528\u300C\u5BFC\u51FA JSON\u300D\u5E76\u6392\u6BD4\u8F83\u4E24\u6B21\u8FD0\u884C": 'Cross-session comparison: this table only covers the current session \u2014 click "Pull disk snapshot" to have the model echo iterate_history, then use "Export JSON" to compare two runs side by side',
+  "\u8DE8\u4F1A\u8BDD\u5BF9\u6BD4\uFF1A\u672C\u8868\u53EA\u8986\u76D6\u5F53\u524D\u4F1A\u8BDD\uFF1B\u8981\u8DE8\u4E24\u6B21\u8FED\u4EE3\u5BF9\u6BD4\uFF0C\u7528\u4E0B\u65B9\u300C\u5B58\u4E3A\u5BF9\u6BD4\u57FA\u51C6\u300D\u4FDD\u5B58\u672C\u6B21\u8FD0\u884C\uFF0C\u518D\u4E0E\u4E0B\u4E00\u6B21\u8FED\u4EE3\u5BF9\u6BD4\u3002": 'Cross-session comparison: this table only covers the current session \u2014 to compare two iterations, save this run with "Save as comparison baseline" below, then compare the next iteration against it.',
+  \u8DE8\u4F1A\u8BDD\u5BF9\u6BD4: "Cross-session comparison",
+  "\u57FA\u51C6\u5DF2\u4FDD\u5B58\uFF1A{a0}": "Baseline saved: {a0}",
+  "\u5148\u4FDD\u5B58\u5F53\u524D\u8FD0\u884C\u4E3A\u57FA\u51C6\uFF0C\u518D\u8DD1\u4E0B\u4E00\u6B21\u8FED\u4EE3\u540E\u4E00\u952E\u5BF9\u6BD4": "Save the current run as the baseline, then one-click compare the next iteration against it",
+  "\u628A\u5F53\u524D\u8FD0\u884C\u6458\u8981\u4FDD\u5B58\u4E3A\u5BF9\u6BD4\u57FA\u51C6\uFF08\u542B\u8F6E\u6B21/\u53D1\u73B0/\u4FEE\u590D/\u9A8C\u8BC1/\u95E8\u7981\uFF09": "Save the current run summary as the comparison baseline (rounds / findings / fixes / validation / gate)",
+  "\u5F53\u524D\u6CA1\u6709\u53EF\u4FDD\u5B58\u7684\u8FD0\u884C\u6570\u636E": "No run data to save yet",
+  \u5DF2\u5B58\u4E3A\u57FA\u51C6: "Saved as baseline",
+  \u5B58\u4E3A\u5BF9\u6BD4\u57FA\u51C6: "Save as comparison baseline",
+  "\u663E\u793A\u6216\u9690\u85CF\u4E0E\u5DF2\u4FDD\u5B58\u57FA\u51C6\u7684\u5BF9\u6BD4\u8868": "Show or hide the comparison against the saved baseline",
+  \u6536\u8D77\u5BF9\u6BD4: "Hide comparison",
+  \u4E0E\u57FA\u51C6\u5BF9\u6BD4: "Compare with baseline",
+  "\u6E05\u9664\u5DF2\u4FDD\u5B58\u7684\u5BF9\u6BD4\u57FA\u51C6": "Clear the saved comparison baseline",
+  \u6E05\u9664\u57FA\u51C6: "Clear baseline",
+  "\u5F53\u524D\u4F1A\u8BDD\u6CA1\u6709\u8FD0\u884C\u6570\u636E\u53EF\u5BF9\u6BD4": "This session has no run data to compare yet",
+  \u6307\u6807: "Metric",
+  \u57FA\u51C6: "Baseline",
+  \u5F53\u524D: "Current",
+  \u53D8\u5316: "Delta",
   // ── F3 finding locations ─────────────────────────────────────────────────
   \u6682\u65E0\u53D1\u73B0: "No findings",
   "{a0}\uFF08{a1} \u9879\uFF09": "{a0} ({a1} findings)",
@@ -3042,6 +3059,101 @@ function registerIterateContextMenus(service, deps, instructionByEntry) {
   };
 }
 
+// src/client/run-compare.ts
+var RUN_BASELINE_VERSION = 1;
+var RUN_BASELINE_STORAGE_KEY = "iterate.run-baseline";
+function buildRunBaseline(input, savedAt) {
+  const convergence = Array.isArray(input.convergence) ? input.convergence : [];
+  const findings = (Array.isArray(input.findings) ? input.findings : []).filter(
+    (f) => typeof f === "object" && f !== null
+  );
+  const fixes = (Array.isArray(input.fixes) ? input.fixes : []).filter(
+    (f) => typeof f === "object" && f !== null
+  );
+  const validations = (Array.isArray(input.validations) ? input.validations : []).filter(
+    (v) => typeof v === "object" && v !== null
+  );
+  let criticalHigh = 0;
+  for (const f of findings) {
+    const sev = String(f && f.severity || "").toLowerCase();
+    if (sev === "critical" || sev === "high") criticalHigh += 1;
+  }
+  let fixed = 0;
+  let failedFixes = 0;
+  for (const x of fixes) {
+    if (x && x.success === true) fixed += 1;
+    else if (x && x.success === false) failedFixes += 1;
+  }
+  let passed = 0;
+  for (const v of validations) {
+    if (v && v.allowed === true) passed += 1;
+  }
+  const validationTotal = validations.length;
+  const validationPassRate = validationTotal > 0 ? Math.round(passed / validationTotal * 100) : null;
+  return {
+    version: RUN_BASELINE_VERSION,
+    savedAt,
+    ...input.project ? { project: input.project } : {},
+    ...input.mode ? { mode: input.mode } : {},
+    rounds: convergence.length,
+    findings: findings.length,
+    criticalHigh,
+    fixed,
+    failedFixes,
+    validationPassRate,
+    validationTotal,
+    qualityGateStatus: input.qualityGateStatus || null,
+    qualityGateScore: typeof input.qualityGateScore === "number" ? input.qualityGateScore : null,
+    stoppedReason: input.stoppedReason || null,
+    convergence
+  };
+}
+function parseRunBaseline(raw) {
+  if (!raw) return null;
+  let parsed;
+  try {
+    parsed = JSON.parse(raw);
+  } catch {
+    return null;
+  }
+  if (typeof parsed !== "object" || parsed === null) return null;
+  const b = parsed;
+  if (b.version !== RUN_BASELINE_VERSION) return null;
+  if (typeof b.savedAt !== "string" || b.savedAt.length === 0) return null;
+  if (typeof b.findings !== "number" || !Number.isFinite(b.findings)) return null;
+  const convergence = Array.isArray(b.convergence) ? b.convergence.filter((n) => typeof n === "number" && Number.isFinite(n)) : [];
+  return {
+    version: RUN_BASELINE_VERSION,
+    savedAt: b.savedAt,
+    ...typeof b.project === "string" && b.project ? { project: b.project } : {},
+    ...typeof b.mode === "string" && b.mode ? { mode: b.mode } : {},
+    rounds: typeof b.rounds === "number" && Number.isFinite(b.rounds) ? b.rounds : convergence.length,
+    findings: b.findings,
+    criticalHigh: typeof b.criticalHigh === "number" && Number.isFinite(b.criticalHigh) ? b.criticalHigh : 0,
+    fixed: typeof b.fixed === "number" && Number.isFinite(b.fixed) ? b.fixed : 0,
+    failedFixes: typeof b.failedFixes === "number" && Number.isFinite(b.failedFixes) ? b.failedFixes : 0,
+    validationPassRate: typeof b.validationPassRate === "number" && Number.isFinite(b.validationPassRate) ? b.validationPassRate : null,
+    validationTotal: typeof b.validationTotal === "number" && Number.isFinite(b.validationTotal) ? b.validationTotal : 0,
+    qualityGateStatus: typeof b.qualityGateStatus === "string" && b.qualityGateStatus ? b.qualityGateStatus : null,
+    qualityGateScore: typeof b.qualityGateScore === "number" && Number.isFinite(b.qualityGateScore) ? b.qualityGateScore : null,
+    stoppedReason: typeof b.stoppedReason === "string" && b.stoppedReason ? b.stoppedReason : null,
+    convergence
+  };
+}
+function compareRuns(baseline, current3) {
+  const num = (v) => typeof v === "number" && Number.isFinite(v) ? v : null;
+  const rows = [
+    { key: "rounds", baseline: num(baseline.rounds), current: num(current3.rounds), higherIsBetter: null },
+    { key: "findings", baseline: num(baseline.findings), current: num(current3.findings), higherIsBetter: false },
+    { key: "criticalHigh", baseline: num(baseline.criticalHigh), current: num(current3.criticalHigh), higherIsBetter: false },
+    { key: "fixed", baseline: num(baseline.fixed), current: num(current3.fixed), higherIsBetter: true },
+    { key: "failedFixes", baseline: num(baseline.failedFixes), current: num(current3.failedFixes), higherIsBetter: false },
+    { key: "validationPassRate", baseline: num(baseline.validationPassRate), current: num(current3.validationPassRate), higherIsBetter: true },
+    { key: "qualityGateScore", baseline: num(baseline.qualityGateScore), current: num(current3.qualityGateScore), higherIsBetter: true }
+  ];
+  return rows;
+}
+
 // src/client/index.ts
 var React = require("react");
 var name = "iterate-plugin";
@@ -4737,6 +4849,15 @@ var OBS_LIVE_META = {
   info: { label: { zh: "\u4FE1\u606F", en: "Info" }, color: "var(--dsw-alias-label-secondary)" }
 };
 var OBS_LIVE_FALLBACK_COLOR = "var(--dsw-alias-label-secondary)";
+var RUN_COMPARE_METRIC_LABELS = {
+  rounds: { zh: "\u8F6E\u6B21", en: "Rounds" },
+  findings: { zh: "\u53D1\u73B0", en: "Findings" },
+  criticalHigh: { zh: "\u4E25\u91CD\uFF08critical+high\uFF09", en: "Critical + high" },
+  fixed: { zh: "\u4FEE\u590D\u6210\u529F", en: "Fixes applied" },
+  failedFixes: { zh: "\u4FEE\u590D\u5931\u8D25", en: "Failed fixes" },
+  validationPassRate: { zh: "\u9A8C\u8BC1\u901A\u8FC7\u7387", en: "Validation pass rate" },
+  qualityGateScore: { zh: "\u95E8\u7981\u5F97\u5206", en: "Quality gate score" }
+};
 function formatObsLiveTime(ts) {
   if (!ts) return "";
   let date = null;
@@ -4804,6 +4925,15 @@ function ObservatoryPanel(props) {
   );
   const [expSearch, setExpSearch] = React.useState("");
   const [defenseFilter, setDefenseFilter] = React.useState("");
+  const [baseline, setBaseline] = React.useState(() => {
+    try {
+      return parseRunBaseline(storage2 ? storage2.get(RUN_BASELINE_STORAGE_KEY) : null);
+    } catch {
+      return null;
+    }
+  });
+  const [showCompare, setShowCompare] = React.useState(false);
+  const [baselineFlash, setBaselineFlash] = React.useState(false);
   const copyTimer = React.useRef(null);
   React.useEffect(() => () => {
     if (copyTimer.current) clearTimeout(copyTimer.current);
@@ -5047,11 +5177,51 @@ function ObservatoryPanel(props) {
       })
     );
   };
+  const currentRunSummary = () => {
+    const hasRun = (manifest.convergence || []).length > 0 || (manifest.findings || []).length > 0 || (manifest.fixes || []).length > 0 || typeof manifest.round === "number" && manifest.round > 0;
+    if (!hasRun) return null;
+    return buildRunBaseline(
+      {
+        project: manifest.project || void 0,
+        mode: manifest.mode || void 0,
+        convergence: manifest.convergence || [],
+        findings: manifest.findings || [],
+        fixes: manifest.fixes || [],
+        validations: manifest.validations || [],
+        qualityGateStatus: qualityGate && qualityGate.overallStatus || null,
+        qualityGateScore: qualityGate && typeof qualityGate.overallScore === "number" ? qualityGate.overallScore : null,
+        stoppedReason: manifest.stoppedReason || null
+      },
+      (/* @__PURE__ */ new Date()).toISOString()
+    );
+  };
+  const saveBaseline = () => {
+    const summary = currentRunSummary();
+    if (!summary || !storage2) return;
+    try {
+      storage2.set(RUN_BASELINE_STORAGE_KEY, JSON.stringify(summary));
+      setBaseline(summary);
+      setShowCompare(true);
+      setBaselineFlash(true);
+      if (copyTimer.current) clearTimeout(copyTimer.current);
+      copyTimer.current = setTimeout(() => setBaselineFlash(false), 1600);
+    } catch (err) {
+      log("failed to persist run baseline", err);
+    }
+  };
+  const clearBaseline = () => {
+    if (storage2) {
+      try {
+        storage2.remove(RUN_BASELINE_STORAGE_KEY);
+      } catch (err) {
+        log("failed to clear run baseline", err);
+      }
+    }
+    setBaseline(null);
+    setShowCompare(false);
+  };
   const renderTrend = () => {
     const conv = manifest.convergence || [];
-    if (conv.length === 0) {
-      return React.createElement("div", { className: "iterate-obs-empty" }, t("\u6682\u65E0\u6536\u655B\u6570\u636E"));
-    }
     const points = conv.map((n, i) => ({ round: i + 1, count: n }));
     const report = latestReport(session);
     const comparison = buildRoundComparison(report, manifest.validations || [], manifest.fixes || []);
@@ -5092,7 +5262,7 @@ function ObservatoryPanel(props) {
         })
       )
     );
-    return React.createElement(
+    const trendBody = conv.length === 0 ? React.createElement("div", { className: "iterate-obs-empty" }, t("\u6682\u65E0\u6536\u655B\u6570\u636E")) : React.createElement(
       "div",
       {},
       React.createElement(TrendChart, { points }),
@@ -5111,12 +5281,111 @@ function ObservatoryPanel(props) {
           React.createElement(
             "span",
             { className: "iterate-obs-head-meta" },
-            t("\u8DE8\u4F1A\u8BDD\u5BF9\u6BD4\uFF1A\u672C\u8868\u53EA\u8986\u76D6\u5F53\u524D\u4F1A\u8BDD\uFF1B\u70B9\u300C\u62C9\u53D6\u78C1\u76D8\u5FEB\u7167\u300D\u8BA9\u6A21\u578B\u56DE\u663E iterate_history\uFF0C\u518D\u7528\u300C\u5BFC\u51FA JSON\u300D\u5E76\u6392\u6BD4\u8F83\u4E24\u6B21\u8FD0\u884C")
+            t("\u8DE8\u4F1A\u8BDD\u5BF9\u6BD4\uFF1A\u672C\u8868\u53EA\u8986\u76D6\u5F53\u524D\u4F1A\u8BDD\uFF1B\u8981\u8DE8\u4E24\u6B21\u8FED\u4EE3\u5BF9\u6BD4\uFF0C\u7528\u4E0B\u65B9\u300C\u5B58\u4E3A\u5BF9\u6BD4\u57FA\u51C6\u300D\u4FDD\u5B58\u672C\u6B21\u8FD0\u884C\uFF0C\u518D\u4E0E\u4E0B\u4E00\u6B21\u8FED\u4EE3\u5BF9\u6BD4\u3002")
           )
         ),
         React.createElement("div", { className: "iterate-obs-block-body" }, comparisonTable)
       ) : null
     );
+    return React.createElement("div", {}, trendBody, renderCrossSessionCompare());
+  };
+  const renderCrossSessionCompare = () => {
+    const current3 = currentRunSummary();
+    const hasRun = current3 !== null;
+    const metricLabel = (key) => {
+      const m = RUN_COMPARE_METRIC_LABELS[key];
+      return m ? m[getLang()] : key;
+    };
+    const fmtCell = (v, key) => {
+      if (v === null) return "\u2014";
+      return key === "validationPassRate" ? `${v}%` : String(v);
+    };
+    const deltaText = (row) => {
+      if (row.baseline === null || row.current === null) return "\u2014";
+      const d = row.current - row.baseline;
+      return d === 0 ? "0" : `${d > 0 ? "+" : ""}${d}`;
+    };
+    const deltaTint = (row) => {
+      if (row.baseline === null || row.current === null || row.higherIsBetter === null) return void 0;
+      const d = row.current - row.baseline;
+      if (d === 0) return void 0;
+      const improved = row.higherIsBetter ? d > 0 : d < 0;
+      return improved ? { color: "var(--dsw-alias-state-success-primary)", fontWeight: 600 } : { color: "var(--dsw-alias-state-error-primary)", fontWeight: 600 };
+    };
+    let compareBody = null;
+    if (baseline && current3 && showCompare) {
+      const rows = compareRuns(baseline, current3);
+      compareBody = React.createElement(
+        "table",
+        { className: "iterate-obs-table" },
+        React.createElement(
+          "thead",
+          {},
+          React.createElement(
+            "tr",
+            {},
+            React.createElement("th", {}, t("\u6307\u6807")),
+            React.createElement("th", {}, t("\u57FA\u51C6")),
+            React.createElement("th", {}, t("\u5F53\u524D")),
+            React.createElement("th", {}, t("\u53D8\u5316"))
+          )
+        ),
+        React.createElement(
+          "tbody",
+          {},
+          ...rows.map(
+            (row) => React.createElement(
+              "tr",
+              { key: `rc-${row.key}` },
+              React.createElement("td", {}, metricLabel(row.key)),
+              React.createElement("td", {}, fmtCell(row.baseline, row.key)),
+              React.createElement("td", {}, fmtCell(row.current, row.key)),
+              React.createElement("td", { style: deltaTint(row) }, deltaText(row))
+            )
+          )
+        )
+      );
+    }
+    return React.createElement(
+      "div",
+      { className: "iterate-obs-block", style: { marginTop: 8 } },
+      React.createElement(
+        "div",
+        { className: "iterate-obs-block-head" },
+        React.createElement("span", {}, t("\u8DE8\u4F1A\u8BDD\u5BF9\u6BD4")),
+        baseline ? React.createElement("span", { className: "iterate-obs-head-meta" }, tf(`\u57FA\u51C6\u5DF2\u4FDD\u5B58\uFF1A{a0}`, { a0: baseline.savedAt })) : React.createElement("span", { className: "iterate-obs-head-meta" }, t("\u5148\u4FDD\u5B58\u5F53\u524D\u8FD0\u884C\u4E3A\u57FA\u51C6\uFF0C\u518D\u8DD1\u4E0B\u4E00\u6B21\u8FED\u4EE3\u540E\u4E00\u952E\u5BF9\u6BD4"))
+      ),
+      React.createElement(
+        "div",
+        { className: "iterate-obs-block-body" },
+        React.createElement(
+          "div",
+          { className: "iterate-obs-bar", style: { flexWrap: "wrap" } },
+          React.createElement("button", {
+            className: "iterate-btn",
+            "data-primary": "",
+            disabled: !hasRun,
+            title: hasRun ? t("\u628A\u5F53\u524D\u8FD0\u884C\u6458\u8981\u4FDD\u5B58\u4E3A\u5BF9\u6BD4\u57FA\u51C6\uFF08\u542B\u8F6E\u6B21/\u53D1\u73B0/\u4FEE\u590D/\u9A8C\u8BC1/\u95E8\u7981\uFF09") : t("\u5F53\u524D\u6CA1\u6709\u53EF\u4FDD\u5B58\u7684\u8FD0\u884C\u6570\u636E"),
+            onClick: saveBaseline
+          }, baselineFlash ? t("\u5DF2\u5B58\u4E3A\u57FA\u51C6") : t("\u5B58\u4E3A\u5BF9\u6BD4\u57FA\u51C6")),
+          baseline ? React.createElement("button", {
+            className: "iterate-btn",
+            "data-copied": showCompare ? "" : void 0,
+            disabled: !hasRun,
+            title: t("\u663E\u793A\u6216\u9690\u85CF\u4E0E\u5DF2\u4FDD\u5B58\u57FA\u51C6\u7684\u5BF9\u6BD4\u8868"),
+            onClick: () => setShowCompare((v) => !v)
+          }, showCompare ? t("\u6536\u8D77\u5BF9\u6BD4") : t("\u4E0E\u57FA\u51C6\u5BF9\u6BD4")) : null,
+          baseline ? React.createElement("button", {
+            className: "iterate-btn",
+            "data-ghost": "",
+            title: t("\u6E05\u9664\u5DF2\u4FDD\u5B58\u7684\u5BF9\u6BD4\u57FA\u51C6"),
+            onClick: clearBaseline
+          }, t("\u6E05\u9664\u57FA\u51C6")) : null
+        ),
+        showCompare && baseline && !hasRun ? React.createElement("div", { className: "iterate-obs-empty", style: { marginTop: 8 } }, t("\u5F53\u524D\u4F1A\u8BDD\u6CA1\u6709\u8FD0\u884C\u6570\u636E\u53EF\u5BF9\u6BD4")) : null,
+        compareBody
+      )
+    );
   };
   const renderFindings = () => {
     const findings = manifest.findings || [];

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "iterate-plugin",
-  "version": "3.15.0",
+  "version": "3.16.0",
   "description": "dsh plugin that turns the iterate skill into an autonomous closed-loop harness with quality command center and experience bank (v3.3). Features: plan -> parallel review xN -> atomic fixes -> validate -> loop -> auto-stop, plus dry-run pure-review mode, quality gate compute/persist, writable experience bank, defense events stream (record + bilingual labels), and native command buttons — with a live quality command center (F8/F9/F10 render real session data), §8 assign-fix instruction, and task_mode indicator wired end to end.",
   "type": "module",
   "license": "MIT",

package/src/client/i18n.ts

@@ -296,8 +296,27 @@ export const EN: Record<string, string> = {
   '各轮发现数量：{a0}{a1}': 'Findings per round: {a0}{a1}',
   ' · 当前 Round {a0}': ' · current round {a0}',
   '跨轮对比（本会话）': 'Cross-round comparison (this session)',
-  '跨会话对比：本表只覆盖当前会话；点「拉取磁盘快照」让模型回显 iterate_history，再用「导出 JSON」并排比较两次运行':
-    'Cross-session comparison: this table only covers the current session — click "Pull disk snapshot" to have the model echo iterate_history, then use "Export JSON" to compare two runs side by side',
+  '跨会话对比：本表只覆盖当前会话；要跨两次迭代对比，用下方「存为对比基准」保存本次运行，再与下一次迭代对比。':
+    'Cross-session comparison: this table only covers the current session — to compare two iterations, save this run with \"Save as comparison baseline\" below, then compare the next iteration against it.',
+  跨会话对比: 'Cross-session comparison',
+  '基准已保存：{a0}': 'Baseline saved: {a0}',
+  '先保存当前运行为基准，再跑下一次迭代后一键对比':
+    'Save the current run as the baseline, then one-click compare the next iteration against it',
+  '把当前运行摘要保存为对比基准（含轮次/发现/修复/验证/门禁）':
+    'Save the current run summary as the comparison baseline (rounds / findings / fixes / validation / gate)',
+  '当前没有可保存的运行数据': 'No run data to save yet',
+  已存为基准: 'Saved as baseline',
+  存为对比基准: 'Save as comparison baseline',
+  '显示或隐藏与已保存基准的对比表': 'Show or hide the comparison against the saved baseline',
+  收起对比: 'Hide comparison',
+  与基准对比: 'Compare with baseline',
+  '清除已保存的对比基准': 'Clear the saved comparison baseline',
+  清除基准: 'Clear baseline',
+  '当前会话没有运行数据可对比': 'This session has no run data to compare yet',
+  指标: 'Metric',
+  基准: 'Baseline',
+  当前: 'Current',
+  变化: 'Delta',
 
   // ── F3 finding locations ─────────────────────────────────────────────────
   暂无发现: 'No findings',

package/src/client/index.ts

@@ -146,6 +146,17 @@ import {
   registerIterateShortcuts,
   type ShortcutsService,
 } from './shortcuts.ts'
+// v3.16: cross-session run comparison — the F2 tab compares two iterations of
+// the same project (save the current run as a baseline, then compare any later
+// run against it), instead of the old export-two-JSONs-and-diff-by-hand chore.
+import {
+  buildRunBaseline,
+  parseRunBaseline,
+  compareRuns,
+  RUN_BASELINE_STORAGE_KEY,
+  type RunBaseline,
+  type RunCompareMetric,
+} from './run-compare.ts'
 
 // ─── Module contract ─────────────────────────────────────────────────────────
 
@@ -2318,6 +2329,17 @@ const OBS_LIVE_META: Record<string, { label: { zh: string; en: string }; color:
 }
 const OBS_LIVE_FALLBACK_COLOR = 'var(--dsw-alias-label-secondary)'
 
+/** v3.16: cross-session comparison metric labels (inline {zh,en} pair, like OBS_TABS). */
+const RUN_COMPARE_METRIC_LABELS: Record<string, { zh: string; en: string }> = {
+  rounds: { zh: '轮次', en: 'Rounds' },
+  findings: { zh: '发现', en: 'Findings' },
+  criticalHigh: { zh: '严重（critical+high）', en: 'Critical + high' },
+  fixed: { zh: '修复成功', en: 'Fixes applied' },
+  failedFixes: { zh: '修复失败', en: 'Failed fixes' },
+  validationPassRate: { zh: '验证通过率', en: 'Validation pass rate' },
+  qualityGateScore: { zh: '门禁得分', en: 'Quality gate score' },
+}
+
 /**
  * Format a live-event timestamp as DD/MM HH:mm:ss (0-padded). Accepts an epoch
  * ms number or an ISO-ish string; any unparseable value degrades to ''.
@@ -2428,6 +2450,20 @@ function ObservatoryPanel(props: SlotProps) {
   // v3.0: Defense events filter
   const [defenseFilter, setDefenseFilter] = React.useState('')
 
+  // v3.16: cross-session comparison — the baseline is read from localStorage
+  // once on mount (a corrupt/stale payload reads as "no baseline", never as a
+  // phantom run); `showCompare` toggles the side-by-side table, `baselineFlash`
+  // confirms the save action.
+  const [baseline, setBaseline] = React.useState<RunBaseline | null>(() => {
+    try {
+      return parseRunBaseline(storage ? storage.get(RUN_BASELINE_STORAGE_KEY) : null)
+    } catch {
+      return null
+    }
+  })
+  const [showCompare, setShowCompare] = React.useState(false)
+  const [baselineFlash, setBaselineFlash] = React.useState(false)
+
   // Single shared "copied" flash timer (reused across all copy buttons).
   const copyTimer = React.useRef<ReturnType<typeof setTimeout> | null>(null)
   React.useEffect(() => () => { if (copyTimer.current) clearTimeout(copyTimer.current) }, [])
@@ -2691,18 +2727,62 @@ function ObservatoryPanel(props: SlotProps) {
     }))
   }
 
+  // ── v3.16: cross-session run comparison handlers ────────────────────────
+  // The summary is built from the SAME manifest the F2 tab renders, so a saved
+  // baseline can never disagree with what the user is looking at.
+  const currentRunSummary = (): RunBaseline | null => {
+    const hasRun =
+      (manifest.convergence || []).length > 0 ||
+      (manifest.findings || []).length > 0 ||
+      (manifest.fixes || []).length > 0 ||
+      (typeof manifest.round === 'number' && manifest.round > 0)
+    if (!hasRun) return null
+    return buildRunBaseline(
+      {
+        project: manifest.project || undefined,
+        mode: manifest.mode || undefined,
+        convergence: manifest.convergence || [],
+        findings: (manifest.findings || []) as Array<{ severity?: string }>,
+        fixes: (manifest.fixes || []) as Array<{ success?: boolean }>,
+        validations: (manifest.validations || []) as Array<{ allowed?: boolean }>,
+        qualityGateStatus: (qualityGate && qualityGate.overallStatus) || null,
+        qualityGateScore: qualityGate && typeof qualityGate.overallScore === 'number' ? qualityGate.overallScore : null,
+        stoppedReason: manifest.stoppedReason || null,
+      },
+      new Date().toISOString(),
+    )
+  }
+  const saveBaseline = () => {
+    const summary = currentRunSummary()
+    if (!summary || !storage) return
+    try {
+      storage.set(RUN_BASELINE_STORAGE_KEY, JSON.stringify(summary))
+      setBaseline(summary)
+      setShowCompare(true)
+      setBaselineFlash(true)
+      if (copyTimer.current) clearTimeout(copyTimer.current)
+      copyTimer.current = setTimeout(() => setBaselineFlash(false), 1600)
+    } catch (err) {
+      log('failed to persist run baseline', err)
+    }
+  }
+  const clearBaseline = () => {
+    if (storage) {
+      try { storage.remove(RUN_BASELINE_STORAGE_KEY) } catch (err) { log('failed to clear run baseline', err) }
+    }
+    setBaseline(null)
+    setShowCompare(false)
+  }
+
   // ── F2: convergence trend (inline bars, reusing TrendChart) ─────────────
   const renderTrend = () => {
     const conv = manifest.convergence || []
-    if (conv.length === 0) {
-      return React.createElement('div', { className: 'iterate-obs-empty' }, t('暂无收敛数据'))
-    }
     const points = conv.map((n, i) => ({ round: i + 1, count: n }))
     // Gap #13 (step ①): single-session cross-round comparison. One row per
     // round — findings raised / fixed / severity mix / validation outcome.
-    // Cross-SESSION comparison rides on the #4 pull instruction instead (the
-    // client has no filesystem access, so `iterate_history` aggregates have to
-    // be echoed back into the session first).
+    // v3.16: the cross-SESSION half is now the `renderCrossSessionCompare`
+    // block below (save the current run as a baseline, then compare any later
+    // run against it) instead of the old export-two-JSONs-and-diff-by-hand chore.
     const report = latestReport(session)
     const comparison = buildRoundComparison(report, manifest.validations || [], manifest.fixes || [])
     const comparisonTable = comparison.length === 0
@@ -2742,22 +2822,117 @@ function ObservatoryPanel(props: SlotProps) {
             }),
           ),
         )
-    return React.createElement('div', {},
-      React.createElement(TrendChart, { points }),
-      React.createElement('div', { className: 'iterate-obs-bar', style: { marginTop: 6 } },
-        tf(`各轮发现数量：{a0}{a1}`, { a0: conv.join(' → '), a1: typeof manifest.round === 'number' ? tf(` · 当前 Round {a0}`, { a0: manifest.round }) : '' }),
-      ),
-      comparisonTable
-        ? React.createElement('div', { className: 'iterate-obs-block', style: { marginTop: 8 } },
-            React.createElement('div', { className: 'iterate-obs-block-head' },
-              React.createElement('span', {}, t('跨轮对比（本会话）')),
-              React.createElement('span', { className: 'iterate-obs-head-meta' },
-                t('跨会话对比：本表只覆盖当前会话；点「拉取磁盘快照」让模型回显 iterate_history，再用「导出 JSON」并排比较两次运行'),
-              ),
+    const trendBody = conv.length === 0
+      ? React.createElement('div', { className: 'iterate-obs-empty' }, t('暂无收敛数据'))
+      : React.createElement('div', {},
+          React.createElement(TrendChart, { points }),
+          React.createElement('div', { className: 'iterate-obs-bar', style: { marginTop: 6 } },
+            tf(`各轮发现数量：{a0}{a1}`, { a0: conv.join(' → '), a1: typeof manifest.round === 'number' ? tf(` · 当前 Round {a0}`, { a0: manifest.round }) : '' }),
+          ),
+          comparisonTable
+            ? React.createElement('div', { className: 'iterate-obs-block', style: { marginTop: 8 } },
+                React.createElement('div', { className: 'iterate-obs-block-head' },
+                  React.createElement('span', {}, t('跨轮对比（本会话）')),
+                  React.createElement('span', { className: 'iterate-obs-head-meta' },
+                    t('跨会话对比：本表只覆盖当前会话；要跨两次迭代对比，用下方「存为对比基准」保存本次运行，再与下一次迭代对比。'),
+                  ),
+                ),
+                React.createElement('div', { className: 'iterate-obs-block-body' }, comparisonTable),
+              )
+            : null,
+        )
+    return React.createElement('div', {}, trendBody, renderCrossSessionCompare())
+  }
+
+  // ── v3.16: cross-session run comparison (F2) ─────────────────────────────
+  // Side-by-side comparison of the CURRENT run against a previously saved
+  // baseline, persisted in localStorage. This replaces the manual
+  // export-JSON-and-diff chore the old cross-session hint pointed at: save the
+  // current run once, then compare any later run against it in one click.
+  const renderCrossSessionCompare = () => {
+    const current = currentRunSummary()
+    const hasRun = current !== null
+    const metricLabel = (key: string): string => {
+      const m = RUN_COMPARE_METRIC_LABELS[key]
+      return m ? m[getLang()] : key
+    }
+    const fmtCell = (v: number | null, key: string): string => {
+      if (v === null) return '—'
+      return key === 'validationPassRate' ? `${v}%` : String(v)
+    }
+    const deltaText = (row: RunCompareMetric): string => {
+      if (row.baseline === null || row.current === null) return '—'
+      const d = row.current - row.baseline
+      return d === 0 ? '0' : `${d > 0 ? '+' : ''}${d}`
+    }
+    const deltaTint = (row: RunCompareMetric): { color: string; fontWeight: number } | undefined => {
+      if (row.baseline === null || row.current === null || row.higherIsBetter === null) return undefined
+      const d = row.current - row.baseline
+      if (d === 0) return undefined
+      const improved = row.higherIsBetter ? d > 0 : d < 0
+      return improved
+        ? { color: 'var(--dsw-alias-state-success-primary)', fontWeight: 600 }
+        : { color: 'var(--dsw-alias-state-error-primary)', fontWeight: 600 }
+    }
+    let compareBody: React.ReactNode = null
+    if (baseline && current && showCompare) {
+      const rows = compareRuns(baseline, current)
+      compareBody = React.createElement('table', { className: 'iterate-obs-table' },
+        React.createElement('thead', {},
+          React.createElement('tr', {},
+            React.createElement('th', {}, t('指标')),
+            React.createElement('th', {}, t('基准')),
+            React.createElement('th', {}, t('当前')),
+            React.createElement('th', {}, t('变化')),
+          ),
+        ),
+        React.createElement('tbody', {},
+          ...rows.map((row) =>
+            React.createElement('tr', { key: `rc-${row.key}` },
+              React.createElement('td', {}, metricLabel(row.key)),
+              React.createElement('td', {}, fmtCell(row.baseline, row.key)),
+              React.createElement('td', {}, fmtCell(row.current, row.key)),
+              React.createElement('td', { style: deltaTint(row) }, deltaText(row)),
             ),
-            React.createElement('div', { className: 'iterate-obs-block-body' }, comparisonTable),
-          )
-        : null,
+          ),
+        ),
+      )
+    }
+    return React.createElement('div', { className: 'iterate-obs-block', style: { marginTop: 8 } },
+      React.createElement('div', { className: 'iterate-obs-block-head' },
+        React.createElement('span', {}, t('跨会话对比')),
+        baseline
+          ? React.createElement('span', { className: 'iterate-obs-head-meta' }, tf(`基准已保存：{a0}`, { a0: baseline.savedAt }))
+          : React.createElement('span', { className: 'iterate-obs-head-meta' }, t('先保存当前运行为基准，再跑下一次迭代后一键对比')),
+      ),
+      React.createElement('div', { className: 'iterate-obs-block-body' },
+        React.createElement('div', { className: 'iterate-obs-bar', style: { flexWrap: 'wrap' } },
+          React.createElement('button', {
+            className: 'iterate-btn', 'data-primary': '', disabled: !hasRun,
+            title: hasRun ? t('把当前运行摘要保存为对比基准（含轮次/发现/修复/验证/门禁）') : t('当前没有可保存的运行数据'),
+            onClick: saveBaseline,
+          }, baselineFlash ? t('已存为基准') : t('存为对比基准')),
+          baseline
+            ? React.createElement('button', {
+                className: 'iterate-btn', 'data-copied': showCompare ? '' : undefined,
+                disabled: !hasRun,
+                title: t('显示或隐藏与已保存基准的对比表'),
+                onClick: () => setShowCompare((v) => !v),
+              }, showCompare ? t('收起对比') : t('与基准对比'))
+            : null,
+          baseline
+            ? React.createElement('button', {
+                className: 'iterate-btn', 'data-ghost': '',
+                title: t('清除已保存的对比基准'),
+                onClick: clearBaseline,
+              }, t('清除基准'))
+            : null,
+        ),
+        showCompare && baseline && !hasRun
+          ? React.createElement('div', { className: 'iterate-obs-empty', style: { marginTop: 8 } }, t('当前会话没有运行数据可对比'))
+          : null,
+        compareBody,
+      ),
     )
   }
 

package/src/client/run-compare.ts

@@ -0,0 +1,213 @@
+/**
+ * src/client/run-compare.ts — cross-session run comparison (the observatory F2 tab).
+ *
+ * The F2 convergence tab compares ROUNDS *within* one session, but comparing
+ * two ITERATIONS of the same project (before / after a refactor) used to be a
+ * manual chore: export JSON for each run, open both files, line up the columns.
+ * This module turns that into a first-class interaction. The current run is
+ * reduced to a compact, serializable summary and saved to localStorage as the
+ * comparison baseline; any later run can then be compared against it side by
+ * side, with a per-metric delta and a directional "better / worse" tint.
+ *
+ * Design constraints (mirrors `i18n.ts` / `session-cache.ts`):
+ *   - **Pure / dependency-free**: no React, no DOM, no Chinese literals (the
+ *     bilingual labels live next to `t()` in index.ts, which is the layer the
+ *     i18n dictionary-contract test scans). The summary builder and the
+ *     compare reducer are unit-testable on their own.
+ *   - **Versioned**: the stored payload carries a `version` so a future schema
+ *     change can migrate or reject stale baselines instead of mis-reading them.
+ *   - **Never fabricated**: every field is derived from real manifest fields
+ *     with defensive optional chaining; a run with no findings simply reads 0.
+ */
+
+/** Current schema version of the persisted baseline. */
+export const RUN_BASELINE_VERSION = 1
+
+/** localStorage key holding the single comparison baseline (per browser). */
+export const RUN_BASELINE_STORAGE_KEY = 'iterate.run-baseline'
+
+/**
+ * A compact, serializable summary of one run — everything the side-by-side
+ * comparison needs, and nothing the client cannot already see in the manifest.
+ */
+export interface RunBaseline {
+  version: 1
+  /** ISO timestamp of when this baseline was saved (NOT the run end time). */
+  savedAt: string
+  /** Project root the run was anchored to (informational). */
+  project?: string
+  /** 'normal' | 'dry-run' — the mode that produced the run. */
+  mode?: string
+  /** Number of review rounds captured. */
+  rounds: number
+  /** Total findings raised across the run. */
+  findings: number
+  /** critical + high findings (the "must fix" surface). */
+  criticalHigh: number
+  /** Fixes that succeeded. */
+  fixed: number
+  /** Fixes that failed (rolled back / validation-rejected). */
+  failedFixes: number
+  /** Fraction of validation command runs that passed, 0-100, or null if none ran. */
+  validationPassRate: number | null
+  /** Number of validation command runs. */
+  validationTotal: number
+  /** 'pass' | 'fail' | null — the quality gate verdict, when it was computed. */
+  qualityGateStatus: string | null
+  /** Quality gate overall score (0-100) when computed. */
+  qualityGateScore: number | null
+  /** Why the run ended: converged / max_rounds_reached / aborted_by_validation. */
+  stoppedReason: string | null
+  /** Findings per round (the convergence series), for a side-by-side trend. */
+  convergence: number[]
+}
+
+/** The manifest-shaped inputs the summary builder reads (defensively). */
+export interface RunBaselineInput {
+  project?: string
+  mode?: string
+  convergence?: number[]
+  findings?: Array<{ severity?: string }>
+  fixes?: Array<{ success?: boolean }>
+  validations?: Array<{ allowed?: boolean }>
+  qualityGateStatus?: string | null
+  qualityGateScore?: number | null
+  stoppedReason?: string | null
+}
+
+/**
+ * Reduce a run manifest (plus the session-scanned quality gate) to a compact,
+ * comparison-ready summary. Pure: `savedAt` is injected by the caller so this
+ * stays deterministic and testable.
+ */
+export function buildRunBaseline(input: RunBaselineInput, savedAt: string): RunBaseline {
+  const convergence = Array.isArray(input.convergence) ? input.convergence : []
+  // Only object entries count — a null / primitive in a malformed manifest
+  // must never be read as a finding / fix / validation (and skew the rate).
+  const findings = (Array.isArray(input.findings) ? input.findings : []).filter(
+    (f): f is { severity?: string } => typeof f === 'object' && f !== null,
+  )
+  const fixes = (Array.isArray(input.fixes) ? input.fixes : []).filter(
+    (f): f is { success?: boolean } => typeof f === 'object' && f !== null,
+  )
+  const validations = (Array.isArray(input.validations) ? input.validations : []).filter(
+    (v): v is { allowed?: boolean } => typeof v === 'object' && v !== null,
+  )
+
+  let criticalHigh = 0
+  for (const f of findings) {
+    const sev = String((f && f.severity) || '').toLowerCase()
+    if (sev === 'critical' || sev === 'high') criticalHigh += 1
+  }
+
+  let fixed = 0
+  let failedFixes = 0
+  for (const x of fixes) {
+    if (x && x.success === true) fixed += 1
+    else if (x && x.success === false) failedFixes += 1
+  }
+
+  let passed = 0
+  for (const v of validations) {
+    if (v && v.allowed === true) passed += 1
+  }
+  const validationTotal = validations.length
+  const validationPassRate = validationTotal > 0 ? Math.round((passed / validationTotal) * 100) : null
+
+  return {
+    version: RUN_BASELINE_VERSION,
+    savedAt,
+    ...(input.project ? { project: input.project } : {}),
+    ...(input.mode ? { mode: input.mode } : {}),
+    rounds: convergence.length,
+    findings: findings.length,
+    criticalHigh,
+    fixed,
+    failedFixes,
+    validationPassRate,
+    validationTotal,
+    qualityGateStatus: input.qualityGateStatus || null,
+    qualityGateScore: typeof input.qualityGateScore === 'number' ? input.qualityGateScore : null,
+    stoppedReason: input.stoppedReason || null,
+    convergence,
+  }
+}
+
+/**
+ * Parse a persisted baseline JSON string. Returns `null` for anything that is
+ * not a well-formed, version-matched baseline — a stale/corrupt payload must
+ * never be mis-read as a real run, and never throws.
+ */
+export function parseRunBaseline(raw: string | null): RunBaseline | null {
+  if (!raw) return null
+  let parsed: unknown
+  try {
+    parsed = JSON.parse(raw)
+  } catch {
+    return null
+  }
+  if (typeof parsed !== 'object' || parsed === null) return null
+  const b = parsed as Record<string, unknown>
+  if (b.version !== RUN_BASELINE_VERSION) return null
+  if (typeof b.savedAt !== 'string' || b.savedAt.length === 0) return null
+  if (typeof b.findings !== 'number' || !Number.isFinite(b.findings)) return null
+  // convergence must be a number array (defensive; a non-array becomes []).
+  const convergence = Array.isArray(b.convergence)
+    ? b.convergence.filter((n): n is number => typeof n === 'number' && Number.isFinite(n))
+    : []
+  return {
+    version: RUN_BASELINE_VERSION,
+    savedAt: b.savedAt,
+    ...(typeof b.project === 'string' && b.project ? { project: b.project } : {}),
+    ...(typeof b.mode === 'string' && b.mode ? { mode: b.mode } : {}),
+    rounds: typeof b.rounds === 'number' && Number.isFinite(b.rounds) ? b.rounds : convergence.length,
+    findings: b.findings,
+    criticalHigh: typeof b.criticalHigh === 'number' && Number.isFinite(b.criticalHigh) ? b.criticalHigh : 0,
+    fixed: typeof b.fixed === 'number' && Number.isFinite(b.fixed) ? b.fixed : 0,
+    failedFixes: typeof b.failedFixes === 'number' && Number.isFinite(b.failedFixes) ? b.failedFixes : 0,
+    validationPassRate:
+      typeof b.validationPassRate === 'number' && Number.isFinite(b.validationPassRate) ? b.validationPassRate : null,
+    validationTotal:
+      typeof b.validationTotal === 'number' && Number.isFinite(b.validationTotal) ? b.validationTotal : 0,
+    qualityGateStatus: typeof b.qualityGateStatus === 'string' && b.qualityGateStatus ? b.qualityGateStatus : null,
+    qualityGateScore:
+      typeof b.qualityGateScore === 'number' && Number.isFinite(b.qualityGateScore) ? b.qualityGateScore : null,
+    stoppedReason: typeof b.stoppedReason === 'string' && b.stoppedReason ? b.stoppedReason : null,
+    convergence,
+  }
+}
+
+/**
+ * One row of the side-by-side comparison.
+ *
+ * `higherIsBetter` tells the renderer how to tint the delta:
+ *   - `true`  → a larger current value is an improvement (fixed, pass rate, score)
+ *   - `false` → a smaller current value is an improvement (findings, failed fixes)
+ *   - `null`  → neutral (rounds — fewer is usually faster but not always better)
+ */
+export interface RunCompareMetric {
+  key: 'rounds' | 'findings' | 'criticalHigh' | 'fixed' | 'failedFixes' | 'validationPassRate' | 'qualityGateScore'
+  baseline: number | null
+  current: number | null
+  higherIsBetter: boolean | null
+}
+
+/**
+ * Compare a saved baseline against the current run. Both sides come from
+ * `buildRunBaseline`, so `null` means "that side had no such measurement"
+ * (e.g. a dry-run has no fixes; a run that never validated has no pass rate).
+ */
+export function compareRuns(baseline: RunBaseline, current: RunBaseline): RunCompareMetric[] {
+  const num = (v: number | null | undefined): number | null =>
+    typeof v === 'number' && Number.isFinite(v) ? v : null
+  const rows: RunCompareMetric[] = [
+    { key: 'rounds', baseline: num(baseline.rounds), current: num(current.rounds), higherIsBetter: null },
+    { key: 'findings', baseline: num(baseline.findings), current: num(current.findings), higherIsBetter: false },
+    { key: 'criticalHigh', baseline: num(baseline.criticalHigh), current: num(current.criticalHigh), higherIsBetter: false },
+    { key: 'fixed', baseline: num(baseline.fixed), current: num(current.fixed), higherIsBetter: true },
+    { key: 'failedFixes', baseline: num(baseline.failedFixes), current: num(current.failedFixes), higherIsBetter: false },
+    { key: 'validationPassRate', baseline: num(baseline.validationPassRate), current: num(current.validationPassRate), higherIsBetter: true },
+    { key: 'qualityGateScore', baseline: num(baseline.qualityGateScore), current: num(current.qualityGateScore), higherIsBetter: true },
+  ]
+  return rows
+}