niceeval 0.9.0 → 0.10.0

package/src/i18n/zh-CN.ts

@@ -58,9 +58,16 @@ export const zhCN = {
   "cli.flag.invalidOutput": "标志 --output 需要 auto|human|agent|ci 之一,收到 \"{{value}}\"。\n",
   "runner.budgetUnenforceable":
     "{{budgetKey}} 的 budget:连续多个 attempt 完成后都拿不到成本数据(agent 不上报用量且模型不在价格表)——该 agent 的 budget 无法执行,取消护栏继续跑。\n",
+  "runner.experimentTeardownFailed":
+    "实验 {{experimentId}} 的 teardown 执行失败:{{message}}。结果不受影响,但该实验起的宿主机资源可能没有回收,请手动检查。\n",
+  "runner.cleanupTimeout": "cleanup 执行超过 {{timeoutMs}}ms 清理超时\n",
+  "runner.setupReturnedCleanup":
+    "{{layer}} 返回了一个函数。setup 不承载收尾,返回值不会被执行——收尾写在同层成对的 teardown 里({{hint}});见 docs-site 的实验教程或 docs/runner.md「环境预置」。\n",
+  "runner.experimentTeardownLate":
+    "实验 {{experimentId}} 的 teardown 未被正常计数路径触发,已在运行收尾兜底执行。结果不受影响;这行出现说明命中了一个未定位的调度间歇问题,请把本次运行信息记入 memory 台账。\n",
   "judge.modelMissing":
     "judge 未配置模型:在 defineConfig({ judge: { model: \"...\" } })、eval 的 judge 配置或环境变量 NICEEVAL_JUDGE_MODEL 里指定裁判模型(没有内置默认模型)。\n" +
-    "  文档:node_modules/niceeval/docs-site/zh/how-to/scoring-guide.mdx",
+    "  文档:node_modules/niceeval/docs-site/zh/tutorials/scoring-guide.mdx",
   "loaders.yamlMissing":
     "loadYaml(\"{{path}}\") 需要 YAML 解析器:请先 `pnpm add yaml`(或改用 loadJson + JSON 数据集)。",
   "cli.flag.parseError": "{{message}}\n运行 `niceeval --help` 查看用法。\n",
@@ -68,7 +75,7 @@ export const zhCN = {
   "cli.help":
     "niceeval — agent-native evals\n\n" +
     "用法:\n" +
-    "  niceeval exp [组|实验] [eval-id 前缀…]   跑实验\n" +
+    "  niceeval exp [路径|实验] [eval-id 前缀…]   跑实验\n" +
     "  niceeval show [eval-id 前缀… | @<locator>]   终端读结果\n" +
     "      裸跑:每个 experiment 的现刻判定(跨 run 合成),每行带紧凑 attempt 索引\n" +
     "        (locator + 失败原因)\n" +
@@ -80,14 +87,14 @@ export const zhCN = {
     "      --timing      整个 attempt 的统一时间树(阶段 + hook/命令/turn + 轮内 OTel)\n" +
     "      --diff[=文件] agent 归因的文件改动摘要;=文件 按窗口展开单个文件\n" +
     "      --history   逐 experiment × eval 的执行时间轴(与 --report 互斥)\n" +
-    "      --results <目录> 钉死结果根   --experiment <id> 只看该实验\n" +
+    "      --results <目录> 钉死结果根   --exp <id> 只看该实验\n" +
     "      --report <文件> 自定义报告   --page <id> 定初始页(多页报告渲染该页,\n" +
     "        尾部再附其余页索引)\n" +
     "  niceeval list                            列出发现到的 eval\n" +
     "  niceeval view [eval-id 前缀…] [--out 目录] [--port n] [--no-open]\n" +
     "      报告页 + 证据室;--report <文件> 整槽换成自定义报告(与 show 同一文件)\n" +
     "      --page <id> 定初始页   --results <目录> 钉死结果根\n" +
-    "      --snapshot <文件> 只打开这一份快照   --experiment <id> 只看该实验\n" +
+    "      --snapshot <文件> 只打开这一份快照   --exp <id> 只看该实验\n" +
     "      --out <目录> 静态导出:index.html 连同查看器 artifact,可直接静态托管\n" +
     "  niceeval sandbox list|enter|history|diff|stop  查看与销毁 --keep-sandbox 留下的现场\n" +
     "  niceeval clean                           删除 .niceeval/ 历史 artifact\n" +
@@ -104,7 +111,7 @@ export const zhCN = {
   "cli.show.noResults": "{{root}} 下没有结果。先 `niceeval exp` 跑一轮,再 `niceeval show`。\n",
   "cli.show.runDirMissing": "Results directory not found: {{dir}}\n",
   "cli.show.noEvalMatch": "No results matched: {{patterns}}. Evals with results: {{evals}}\n",
-  "cli.show.noExperimentMatch": "No experiment matched --experiment {{arg}}. Experiments with results: {{experiments}}\n",
+  "cli.show.noExperimentMatch": "No experiment matched --exp {{arg}}. Experiments with results: {{experiments}}\n",
   "cli.show.historyReportConflict":
     "`--history` and `--report` are mutually exclusive: both take over the main output. --history is the host's per-attempt execution timeline; for snapshot-level trends, compose exp.snapshots inside your report file instead.\n",
   "cli.show.evidenceNeedsEval":
@@ -116,10 +123,14 @@ export const zhCN = {
   "cli.eval.noMatchKnown": "已发现 {{count}} 个 eval:{{evals}}\n",
   "cli.exp.agentModelFlagUnsupported": "experiment 运行不支持 --agent / --model。请新增或复制一个 experiment 文件并修改 model。\n",
   "cli.exp.viewerFlagUnsupported": "`{{flag}}` 只适用于 niceeval {{command}},不能用于 niceeval exp。\n",
-  "cli.experiment.noMatch": "没有匹配的实验:{{arg}}。已发现:{{experiments}}\n",
+  "cli.experiment.noMatch":
+    "没有匹配的实验:{{arg}}。可用路径:{{experiments}}。\n" +
+    "运行 `niceeval exp <路径> --dry` 预览计划。\n",
   "cli.experiment.viewerCommandHint": "你可能想运行:niceeval {{command}}{{args}}\n",
-  "cli.experiment.noEvalsSelected": "未选择任何 eval:{{selection}} 匹配到 0 个 eval。可用实验:{{experiments}}。\n",
-  "cli.experimentGroup": "组",
+  "cli.experiment.noEvalsSelected":
+    "未选择任何 eval:{{selection}} 匹配到 0 个 eval。可用的 eval 前缀:{{experiments}}。\n" +
+    "运行 `niceeval exp {{selection}} --dry` 查看它覆盖了什么,或去掉 eval 过滤跑这些实验选中的全部 eval。\n",
+  "cli.experimentGroup": "路径",
   "cli.fallbackCleanupTimeout": "\ngraceful 清理超时,强制清理沙箱…\n",
   "cli.forceCleanupExit": "\n强制清理沙箱并退出…\n",
   "cli.init.done": "已就绪:evals/、niceeval.config.ts,以及 AGENTS.md 里的 niceeval agent 指引区块(指向 node_modules/niceeval/docs-site/zh)。\n",
@@ -130,8 +141,8 @@ export const zhCN = {
   "cli.pressCtrlC": "按 Ctrl+C 退出。\n",
   "cli.resultsPath": "结构化结果:{{path}}(snapshot.json + 每 attempt 的 result.json / events.json / trace.json / diff.json)\n",
   "cli.run.experimentRequired":
-    "运行 eval 必须通过 experiment:用 `niceeval exp [实验组|配置] [eval id 前缀]`。\n" +
-    "  文档:node_modules/niceeval/docs-site/zh/how-to/write-experiment.mdx\n",
+    "运行 eval 必须通过 experiment:用 `niceeval exp [路径|配置] [eval id 前缀]`。\n" +
+    "  文档:node_modules/niceeval/docs-site/zh/tutorials/write-experiment.mdx\n",
   "cli.run.experimentRequiredHint": "提示:\"{{pattern}}\" 是实验{{kind}},你大概想跑:niceeval exp {{pattern}}\n",
   "cli.run.experimentRequiredKnown": "已发现实验:{{experiments}}\n",
   "cli.unimplemented": "命令 \"{{command}}\" 暂未实现(MVP)。\n",
@@ -142,7 +153,7 @@ export const zhCN = {
   "cli.view.url": "niceeval view: {{url}}\n",
   "context.capabilityMissing":
     "agent \"{{agent}}\" 不是沙箱型(defineSandboxAgent 构造),t.{{method}} 这类断言只有沙箱型 agent 可用。换用 defineSandboxAgent 构造的 agent,或去掉这条断言。\n" +
-    "  文档:node_modules/niceeval/docs-site/zh/how-to/sandbox-agent.mdx",
+    "  文档:node_modules/niceeval/docs-site/zh/tutorials/sandbox-agent.mdx",
   "context.skipEmpty": "skip() 需要一个非空理由。",
   "context.turnFailed": "本轮 send 返回 failed(turn status = failed):{{message}}",
   "context.turnFailedDefault": "本轮 send 返回 failed(turn status = failed)",
@@ -152,6 +163,8 @@ export const zhCN = {
   "define.evalTestRequired": "defineEval 需要一个 async test(t) 函数。",
   "define.experimentAgentRequired": "defineExperiment 需要 agent。",
   "define.experimentFlagNotJson": "experiment.flags.{{key}} 不是可 JSON 序列化的值(函数 / undefined / 循环引用 / bigint 不允许);flags 会原样进入结果快照,必须是纯 JSON。",
+  "define.experimentLabelInvalid": "experiment.labels.{{key}} 必须是字符串或有限数字;labels 是报告侧的归类坐标,会原样进入结果快照。",
+  "define.experimentSetupNotFunction": "experiment.setup 必须是函数((ctx) => void);要清理请挂 experiment.teardown;要按实验准备沙箱内环境请挂 sandbox spec 的 .setup() 钩子链。",
   "define.experimentIdRejected": "defineExperiment 不接受 id —— id 由文件路径推导。",
   "define.sandboxAgentNameRequired": "defineSandboxAgent 需要 name。",
   "define.sandboxCreateRequired": "defineSandbox 需要一个 create() 函数。",
@@ -164,7 +177,7 @@ export const zhCN = {
   "docker.unsupportedRuntime": "Unsupported runtime: {{runtime}}",
   "feedback.human.active": "ACTIVE",
   "feedback.human.budgetExhausted": "{{experimentId}} 预算已耗尽(已花 {{spent}},未跑 {{unstarted}})",
-  "feedback.human.compare": "Compare: niceeval view {{group}}",
+  "feedback.human.compare": "Compare: niceeval view",
   "feedback.human.counts": "共 {{total}} · 复用 {{reused}} · 运行中 {{running}} · 排队 {{queued}} · 已完成 {{completed}}",
   "feedback.human.diffHint": "Diff:    niceeval show {{locator}} --diff",
   "feedback.human.evalHint": "Eval:    niceeval show {{locator}} --source",
@@ -189,6 +202,10 @@ export const zhCN = {
   "feedback.phase.evalSetup": "eval 预置",
   "feedback.phase.sandboxCreate": "创建沙箱",
   "feedback.phase.sandboxQueue": "排队等沙箱",
+  "feedback.phase.experimentSetup": "实验预置",
+  "feedback.phase.experimentTeardown": "实验收尾",
+  "feedback.human.hookDone": "完成",
+  "feedback.human.hookFailed": "失败",
   "feedback.phase.sandboxSetup": "沙箱预置",
   "feedback.phase.scoring": "评分",
   "feedback.phase.teardown": "清理中",
@@ -268,13 +285,11 @@ export const zhCN = {
   "runner.startSandboxTeardown": "sandbox teardown(环境预置钩子)…",
   "runner.timeout": "attempt 超时({{timeoutMs}}ms)\n最近进度:\n{{recentLogs}}",
   "runner.traceSelected": " → 留 {{count}}(按语义)",
-  "runner.resumeCarry": "  · 复用上次 {{carried}} 个已判定的结果,重跑 {{retry}} 个 eval\n",
-  "runner.resumeCarryDetail": "      复用 [{{experiment}}] {{evals}}\n",
   "runner.useRemoteAgent": "使用 remote agent(不创建沙箱)…",
   "sandbox.providerNotImplemented": "{{provider}} sandbox provider not implemented; use docker, vercel, or e2b",
   "sandbox.missingSpec":
     "沙箱型 agent 需要一个 sandbox,但没有提供。niceeval 不再自动选默认 provider——请在 defineExperiment()/defineConfig() 里把 sandbox 设成 dockerSandbox() / vercelSandbox() / e2bSandbox()(从 \"niceeval/sandbox\" 导入)。\n" +
-    "  文档:node_modules/niceeval/docs-site/zh/how-to/sandbox-providers.mdx",
+    "  文档:node_modules/niceeval/docs-site/zh/tutorials/sandbox-providers.mdx",
   "sandbox.dependencyMissing.docker": "Docker sandbox requires 'dockerode'. Install it with: pnpm add dockerode @types/dockerode",
   "sandbox.dependencyMissing.e2b": "E2B sandbox requires 'e2b'. Install it with: pnpm add e2b",
   "sandbox.dependencyMissing.vercel": "Vercel sandbox requires '@vercel/sandbox'. Install it with: pnpm add @vercel/sandbox",

package/src/index.ts

@@ -24,7 +24,9 @@ export type {
   Severity,
   Verdict,
   EvalDef,
+  EvalDescriptor,
   ExperimentDef,
+  ExperimentHookContext,
   Config,
   LocalizedText,
   JudgeConfig,

package/src/report/{react → assets}/colors.ts

@@ -32,6 +32,28 @@ export function colorIndexForKey(key: string): number {
   return hash % NRE_PALETTE_SIZE;
 }
 
+/**
+ * 同一张图内 series 键集合的配色:每个键仍以稳定散列为起点(无冲突时与跨图配色一致),
+ * 图内撞色时按传入顺序(图例顺序)线性探测下一个空色格——跨图稳定让位给图内可辨;
+ * 键数超过色板后无空格可探,回落散列格(复用不可避免)。返回键 → 色板下标。
+ * 契约见 docs/feature/reports/library/metric-views.md「MetricScatter」;
+ * 修法台账见 memory/scatter-series-color-collision.md。
+ */
+export function colorIndicesForKeys(keys: readonly string[]): Map<string, number> {
+  const used = new Set<number>();
+  const out = new Map<string, number>();
+  for (const key of keys) {
+    if (out.has(key)) continue;
+    let idx = colorIndexForKey(key);
+    if (used.size < NRE_PALETTE_SIZE) {
+      while (used.has(idx)) idx = (idx + 1) % NRE_PALETTE_SIZE;
+    }
+    used.add(idx);
+    out.set(key, idx);
+  }
+  return out;
+}
+
 /** 键对应的稳定 class 名("nre-c3"),配 styles.css 的 .nre-cN 上文字色。 */
 export function colorClassForKey(key: string): string {
   return `nre-c${colorIndexForKey(key)}`;

package/src/report/{react → assets}/enhance.js

@@ -1,5 +1,5 @@
 // niceeval 报告的渐进增强 runtime:纯 vanilla JS、零依赖、IIFE、幂等。
-// 只作用于 .nre DOM 与 data-nre-* 属性;六个行为——实验组切换、Tabs 单选切换、表格排序、
+// 只作用于 .nre DOM 与 data-nre-* 属性;五个行为——Tabs 单选切换、表格排序、
 // 行过滤、SVG 点 tooltip、警告命令复制。全部只改浏览状态,不改数据、指标口径或初始 HTML 数值。
 // 静态 HTML 无 JS 时内容完整可读是硬约束:排序有数据侧预排、tooltip 退化为原生 <title>、
 // 过滤输入框静默无功能、Tabs 退化为原生 <details> 手风琴、命令块退化为点击全选。
@@ -19,38 +19,6 @@
     return target && target.closest ? target.closest(selector) : null;
   }
 
-  // ───────────────────────── 实验组:[data-nre-experiment-group-select] ─────────────────────────
-  // 数据侧已经把每组的摘要、散点和列表分别算好；这里仅在同一报告节点内切换 <details>，
-  // 不重新请求、不重算，也不触碰 Runs / Traces / Attempt 证据室。
-
-  function selectExperimentGroup(control) {
-    var root = control.closest("[data-nre-experiment-groups]");
-    if (!root) return;
-    var selected = control.getAttribute("data-nre-experiment-group-select");
-    var controls = root.querySelectorAll("[data-nre-experiment-group-select]");
-    var panels = root.querySelectorAll("[data-nre-experiment-group-panel]");
-    for (var i = 0; i < controls.length; i++) {
-      var active = controls[i].getAttribute("data-nre-experiment-group-select") === selected;
-      controls[i].setAttribute("aria-selected", active ? "true" : "false");
-      controls[i].tabIndex = active ? 0 : -1;
-    }
-    for (var j = 0; j < panels.length; j++) {
-      panels[j].open = panels[j].getAttribute("data-nre-experiment-group-panel") === selected;
-    }
-  }
-
-  document.addEventListener("click", function (e) {
-    var control = closest(e.target, "[data-nre-experiment-group-select]");
-    if (control) selectExperimentGroup(control);
-  });
-
-  document.addEventListener("keydown", function (e) {
-    var control = closest(e.target, "[data-nre-experiment-group-select]");
-    if (!control || (e.key !== "Enter" && e.key !== " ")) return;
-    e.preventDefault();
-    selectExperimentGroup(control);
-  });
-
   // ───────────────────────── Tabs:[data-nre-tabs] 单选切换 ─────────────────────────
   // 静态 HTML 每 tab 一个 <details> 且仅首个 open;点击 summary 时接管原生 toggle:
   // 打开所点 tab、收起同组其余,点已开的 tab 保持打开(单选语义,恒有一个面板可见)。

package/src/report/{react → assets}/styles.css

@@ -65,98 +65,6 @@
 .nre-attempt-list .nre-truncated { color:var(--warn); font-size:.9em; }
 .nre-attempt-list .nre-attempt-list-empty { border:1px dashed var(--line-strong); border-radius:8px; padding:18px; color:var(--muted); text-align:center; background:var(--panel); }
 
-/* ---- ExperimentComparison:完整组索引 + 当前组独立 panel ---- */
-.nre-experiment-comparison .nre-experiment-group-tabs {
-  display:grid;
-  grid-template-columns:repeat(auto-fit,minmax(320px,1fr));
-  gap:12px;
-  margin-bottom:20px;
-}
-.nre-experiment-comparison .nre-experiment-group-tab {
-  min-width:0;
-  border:1px solid var(--line);
-  border-radius:9px;
-  padding:14px 16px 8px;
-  background:var(--panel);
-  cursor:pointer;
-  transition:border-color .12s,background .12s;
-}
-.nre-experiment-comparison .nre-experiment-group-tab:hover {
-  border-color:var(--line-strong);
-  background:color-mix(in oklch,var(--panel-2),var(--panel) 35%);
-}
-.nre-experiment-comparison .nre-experiment-group-tab:focus-visible {
-  outline:2px solid var(--nre-c0);
-  outline-offset:2px;
-}
-.nre-experiment-comparison .nre-experiment-group-name {
-  display:block;
-  margin-bottom:10px;
-  color:var(--text);
-  font-size:15px;
-  overflow-wrap:anywhere;
-}
-/* 组卡里嵌的 ScopeSummary 降为紧凑行:去掉满宽 KPI 条的边框,收小数字 */
-.nre-experiment-comparison .nre-experiment-group-tab .nre-scope-summary { margin:0; }
-.nre-experiment-comparison .nre-experiment-group-tab .nre-scope-kpis {
-  display:flex;
-  flex-wrap:wrap;
-  column-gap:16px;
-  row-gap:6px;
-  margin:0 0 6px;
-  border:0;
-  border-radius:0;
-  background:transparent;
-  overflow:visible;
-}
-.nre-experiment-comparison .nre-experiment-group-tab .nre-scope-kpi {
-  min-width:58px;
-  padding:0;
-  border-right:0;
-}
-.nre-experiment-comparison .nre-experiment-group-tab .nre-scope-kpi dd {
-  margin:3px 0 0;
-  font-size:16px;
-  line-height:1.15;
-  font-weight:650;
-}
-.nre-experiment-comparison .nre-experiment-group-tab .nre-scope-kpi-rate dd { font-size:18px; }
-.nre-experiment-comparison .nre-experiment-group-panel {
-  padding:16px 0;
-}
-.nre-experiment-comparison .nre-experiment-group-panel > summary {
-  cursor:pointer;
-  font-size:16px;
-  font-weight:700;
-  overflow-wrap:anywhere;
-}
-.nre-experiment-comparison .nre-experiment-group-panel[open] > summary {
-  margin-bottom:16px;
-}
-.nre-experiment-comparison .nre-experiment-group-panel > .nre {
-  margin:0 0 18px;
-}
-/* group panel 已经是内容的唯一外框;内嵌图表不再重复套一层卡片边框 */
-.nre-experiment-comparison .nre-experiment-group-panel .nre-metric-scatter,
-.nre-experiment-comparison .nre-experiment-group-panel .nre-metric-line,
-.nre-experiment-comparison .nre-experiment-group-panel .nre-metric-bars {
-  border:0;
-  border-radius:0;
-  background:transparent;
-  padding:14px 0 8px;
-}
-.nre-experiment-comparison .nre-experiment-group-panel > .nre:last-child {
-  margin-bottom:0;
-}
-.nre-experiment-comparison .nre-experiment-groups-empty {
-  border:1px dashed var(--line-strong);
-  border-radius:8px;
-  padding:18px;
-  color:var(--muted);
-  text-align:center;
-  background:var(--panel);
-}
-
 /* ---- EvalList:实体列表的第二级,一条 <details> 一个 experimentId × evalId ---- */
 .nre-eval-list .nre-evals { list-style:none; margin:0; padding:0; }
 .nre-eval-list .nre-eval { border-bottom:1px solid var(--line); }
@@ -176,13 +84,19 @@
 /* ---- ExperimentList:web 面固定八列比较表,一条 <details> 一个 experiment ---- */
 .nre-experiment-table { border:1px solid var(--line); border-radius:9px; overflow:hidden; background:var(--panel); }
 .nre-experiment-head,
-.nre-experiment-summary { display:grid; grid-template-columns:minmax(220px,1.25fr) minmax(120px,.8fr) minmax(90px,.6fr) minmax(100px,.65fr) minmax(90px,.55fr) minmax(110px,.7fr) minmax(90px,.6fr) minmax(160px,.9fr); column-gap:16px; align-items:center; }
-.nre-experiment-head { padding:12px 20px; color:var(--muted); font-size:12px; font-weight:560; text-transform:uppercase; letter-spacing:.05em; border-bottom:1px solid var(--line); }
-.nre-experiment-head button { appearance:none; border:0; padding:0; background:transparent; color:inherit; font:inherit; text-transform:inherit; letter-spacing:inherit; text-align:left; cursor:pointer; }
-.nre-experiment-head button::after { content:"↕"; margin-left:6px; color:var(--soft); font-size:11px; }
-.nre-experiment-head button.nre-sort-asc::after { content:"↑"; color:var(--text); }
-.nre-experiment-head button.nre-sort-desc::after { content:"↓"; color:var(--text); }
-.nre-experiment-table-wrap > .nre-filter { height:34px; min-width:260px; margin-bottom:16px; border:1px solid var(--line); border-radius:7px; padding:0 11px; background:var(--panel); color:var(--text); font:inherit; }
+.nre-experiment-summary { display:grid; grid-template-columns:minmax(240px,1.35fr) minmax(115px,.72fr) minmax(95px,.58fr) minmax(108px,.62fr) minmax(105px,.58fr) minmax(112px,.66fr) minmax(92px,.52fr) minmax(160px,.86fr); column-gap:14px; align-items:center; }
+.nre-experiment-head { padding:13px 20px; color:var(--muted); font-size:12px; line-height:1.2; font-weight:620; border-bottom:1px solid var(--line); }
+.nre-experiment-head button { appearance:none; display:inline-flex; width:max-content; max-width:100%; align-items:center; gap:5px; border:0; padding:0; background:transparent; color:inherit; font:inherit; line-height:inherit; text-align:left; white-space:nowrap; cursor:pointer; }
+.nre-experiment-head button:hover,
+.nre-experiment-head button:focus-visible { color:var(--text); }
+.nre-experiment-head .nre-num-head { justify-self:end; text-align:right; }
+.nre-experiment-head .nre-sort-icon { display:inline-block; flex:0 0 11px; width:11px; color:var(--soft); font-size:11px; text-align:center; }
+.nre-experiment-head .nre-sort-icon::before { content:"↕"; opacity:0; transition:opacity .12s; }
+.nre-experiment-head button:hover .nre-sort-icon::before,
+.nre-experiment-head button:focus-visible .nre-sort-icon::before { opacity:1; }
+.nre-experiment-head button.nre-sort-asc .nre-sort-icon::before { content:"↑"; opacity:1; color:var(--text); }
+.nre-experiment-head button.nre-sort-desc .nre-sort-icon::before { content:"↓"; opacity:1; color:var(--text); }
+.nre-experiment-table-wrap > .nre-filter { display:block; width:min(100%,280px); height:34px; min-width:0; margin:0 0 12px auto; border:1px solid var(--line); border-radius:7px; padding:0 11px; background:var(--panel); color:var(--text); font:inherit; }
 .nre-experiment-list .nre-experiment-entry { border-bottom:1px solid var(--line); }
 .nre-experiment-list .nre-experiment-entry:last-child { border-bottom:0; }
 .nre-experiment-list .nre-experiment-summary { padding:13px 20px; cursor:pointer; list-style:none; transition:background .12s; }
@@ -194,11 +108,11 @@
 .nre-experiment-list .nre-experiment-entry[open] .nre-experiment-name::before { transform:rotate(45deg); }
 .nre-experiment-list .nre-experiment-id { display:block; overflow-wrap:anywhere; font-weight:650; }
 .nre-experiment-list .nre-experiment-name small { display:block; margin-top:3px; color:var(--muted); font-size:12px; font-weight:400; }
-.nre-experiment-list .nre-num { font-variant-numeric:tabular-nums; font-feature-settings:"tnum"; white-space:nowrap; }
+.nre-experiment-list .nre-num { text-align:right; font-variant-numeric:tabular-nums; font-feature-settings:"tnum"; white-space:nowrap; }
 .nre-experiment-list .nre-good { color:var(--good); }
 .nre-experiment-list .nre-warn { color:var(--warn); }
 .nre-experiment-list .nre-bad { color:var(--bad); }
-.nre-experiment-list .nre-experiment-pill { display:inline-flex; min-height:22px; align-items:center; border:1px solid var(--line); border-radius:999px; padding:0 8px; color:var(--muted); font-size:12px; white-space:nowrap; }
+.nre-experiment-list .nre-experiment-pill { display:inline-flex; min-height:22px; align-items:center; color:var(--muted); font-size:12px; white-space:nowrap; }
 .nre-experiment-list .nre-experiment-detail { padding:18px 30px 22px; background:color-mix(in oklch,var(--panel-2),var(--panel) 24%); border-top:1px solid var(--line); }
 .nre-experiment-list .nre-experiment-flags { display:flex; flex-wrap:wrap; gap:8px; margin:0 0 14px; color:var(--soft); font-size:12px; }
 .nre-experiment-list .nre-experiment-flags b { border:1px solid color-mix(in oklch,var(--line),transparent 16%); border-radius:999px; padding:3px 8px; color:var(--text); font-weight:560; }
@@ -219,7 +133,7 @@
 .nre-experiment-list .nre-eval-attempt-metrics,
 .nre-experiment-list .nre-eval-reason { color:var(--muted); font-size:.92em; }
 .nre-experiment-list .nre-experiment-list-empty { border:1px dashed var(--line-strong); border-radius:8px; padding:18px; color:var(--muted); text-align:center; background:var(--panel); }
-@media(max-width:1000px){.nre-experiment-table{overflow-x:auto}.nre-experiment-head,.nre-experiment-summary{min-width:1040px}.nre-experiment-detail{min-width:1040px}}
+@media(max-width:1160px){.nre-experiment-table{overflow-x:auto}.nre-experiment-head,.nre-experiment-summary{min-width:1120px}.nre-experiment-detail{min-width:1120px}}
 
 /* ---- verdict 判定色(EvalList/ExperimentList 的判定符;errored 在这里按 bad 收敛) ---- */
 .nre .nre-eval-verdict.nre-verdict-passed { color:var(--good); }
@@ -425,6 +339,15 @@ table.nre,
   font-variant-numeric: tabular-nums;
   white-space: nowrap;
 }
+.nre-scope-summary .nre-scope-kpi-note {
+  display:block;
+  margin-top:5px;
+  color:var(--soft);
+  font-size:11px;
+  font-weight:450;
+  line-height:1.25;
+  white-space:nowrap;
+}
 /* 计票格装的是 pill 组:允许换行,行高回到常规 */
 .nre-scope-summary .nre-scope-kpi-verdicts dd {
   white-space: normal;
@@ -516,20 +439,26 @@ table.nre,
 }
 
 /* ---- Hero / PoweredBy(站点标题区与品牌行) ---- */
-.nre-hero { margin: 0 0 18px; }
+/* 居中大标题:hero 是页首的门面块,整块占满内容列宽、文本水平居中,标题随视口 clamp 放大,
+   meta 与品牌行居中跟随。不设 max-width——窄框居中在宽内容列里会读成左偏,占满列宽才真居中。 */
+.nre-hero {
+  text-align: center;
+  margin: 0 0 clamp(28px, 5vw, 44px);
+}
 .nre-hero .nre-hero-title {
   margin: 0;
-  font-size: 24px;
-  font-weight: 700;
+  font-size: clamp(38px, 5vw, 60px);
+  line-height: 0.98;
+  font-weight: 760;
   letter-spacing: -0.01em;
 }
 .nre-hero .nre-hero-meta {
-  margin: 4px 0 0;
+  margin: 14px 0 0;
   color: var(--muted);
-  font-size: 13px;
+  font-size: 14px;
 }
 .nre-powered-by {
-  margin: 6px 0 0;
+  margin: 10px 0 0;
   font-size: 12px;
 }
 .nre-powered-by a {
@@ -714,9 +643,10 @@ table.nre,
 .nre-metric-scatter .nre-scatter-svg,
 .nre-metric-line .nre-line-svg,
 .nre-metric-bars .nre-bars-svg {
-  max-width: 720px;
+  max-width: 880px;
   width: 100%;
   height: auto;
+  margin-inline: auto;
   display: block;
   overflow: visible;
 }
@@ -759,11 +689,16 @@ table.nre,
 }
 .nre-metric-scatter .nre-scatter-legend,
 .nre-metric-bars .nre-bars-legend {
+  display: flex;
+  max-width: 880px;
+  flex-wrap: wrap;
+  gap: 0.35rem 0.9rem;
   color: var(--muted);
   font-size: 0.85em;
-  margin-top: 0.25rem;
+  margin: 0.5rem auto 0;
 }
-.nre .nre-legend-key { margin-right: 0.75em; }
+.nre .nre-legend-key { display:inline-flex; align-items:center; gap:0.4em; }
+.nre .nre-legend-key::before { content:""; width:0.55em; height:0.55em; border-radius:50%; background:currentColor; }
 .nre-metric-bars .nre-bars-metric {
   margin-right: 1em;
   font-weight: 600;
@@ -771,9 +706,10 @@ table.nre,
 }
 .nre-metric-scatter .nre-scatter-missing,
 .nre-metric-line .nre-line-missing {
+  max-width: 880px;
   color: var(--warn);
   font-size: 0.85em;
-  margin: 0.25rem 0 0;
+  margin: 0.35rem auto 0;
 }
 /* 散点空态(0 可画点)/ 不足以成图(1 可画点):明说缺什么,不画空坐标系 */
 .nre-metric-scatter .nre-scatter-empty {
@@ -862,6 +798,94 @@ table.nre,
   margin: 0.25rem 0;
 }
 
+/* ---- Section header:标题行右侧短元信息(meta),同一行右对齐,空间不足自动换行
+        (docs/feature/reports/library/layout.md「Grid 与 Stat」)。省略 meta 时 Section
+        渲染的是裸 <h2>(既有结构),不套这层 header——旧行为逐字节不变。 ---- */
+.nre-section-header {
+  display: flex;
+  flex-wrap: wrap;
+  align-items: baseline;
+  justify-content: space-between;
+  gap: 4px 16px;
+  margin: 0 0 10px;
+}
+.nre-section-header .nre-section-title { margin: 0; }
+.nre-section-header .nre-section-meta {
+  margin: 0;
+  color: var(--muted);
+  font-size: 13px;
+  white-space: nowrap;
+}
+
+/* ---- Grid:自由摘要面板的格子容器(docs/feature/reports/architecture.md「排版原语的语义层
+        与面内布局」)。max-width 钳在「声明列数 × 最小格宽 + 格间距」——auto-fit 因此永远
+        摆不出比声明列数更多的列;容器比这个宽度窄时,minmax 的下限让列数按最小 inline size
+        自动减少,窄到个位数视口也不会产生页面级横向滚动。响应式换列全靠这几行 CSS,
+        不需要测量脚本或 hydration。 ---- */
+.nre-grid {
+  --nre-grid-cell-min: 220px;
+  --nre-grid-gap: 20px;
+  --nre-grid-cell-padding: 16px;
+  --nre-stat-value-size: 20px;
+  display: grid;
+  grid-template-columns: repeat(auto-fit, minmax(min(var(--nre-grid-cell-min), 100%), 1fr));
+  gap: var(--nre-grid-gap);
+  max-width: calc(
+    var(--nre-grid-max-columns) * var(--nre-grid-cell-min) + (var(--nre-grid-max-columns) - 1) * var(--nre-grid-gap)
+  );
+  margin: 1rem 0;
+}
+.nre-grid--compact {
+  --nre-grid-cell-min: 160px;
+  --nre-grid-gap: 10px;
+  --nre-grid-cell-padding: 10px;
+  --nre-stat-value-size: 16px;
+}
+.nre-grid-cell { min-width: 0; }
+/* boxed:每个 cell 独立完整四边框,靠 grid gap 分开——不是靠相邻项拼半边框,换行后
+   不会缺左边、缺底边或出现双线。 */
+.nre-grid--boxed .nre-grid-cell {
+  border: 1px solid var(--line);
+  border-radius: 8px;
+  padding: var(--nre-grid-cell-padding);
+  background: var(--panel);
+}
+.nre-grid--plain .nre-grid-cell { padding: 0; }
+/* 一个 cell 里嵌套多个 Stat(Col 归拢):只有纵向留白,没有内部分隔线。 */
+.nre-grid-cell > .nre-col { gap: calc(var(--nre-grid-gap) * 0.6); }
+
+/* ---- Stat:label / 主值 / detail 三段;label 与 detail 弱化层级,value 用 tabular
+        numerals;tone 是主值的语义色,只染 value——positive 不会把 label 和 detail
+        一并染色。 ---- */
+.nre-stat {
+  display: flex;
+  flex-direction: column;
+  gap: 4px;
+  min-width: 0;
+}
+.nre-stat .nre-stat-label {
+  color: var(--soft);
+  font-size: 12px;
+  font-weight: 560;
+  overflow-wrap: anywhere;
+}
+.nre-stat .nre-stat-value {
+  font-size: var(--nre-stat-value-size);
+  font-weight: 700;
+  line-height: 1.15;
+  font-variant-numeric: tabular-nums;
+  font-feature-settings: "tnum";
+  overflow-wrap: anywhere;
+}
+.nre-stat .nre-stat-detail {
+  color: var(--muted);
+  font-size: 12px;
+  overflow-wrap: anywhere;
+}
+.nre-stat--positive .nre-stat-value { color: var(--good); }
+.nre-stat--negative .nre-stat-value { color: var(--bad); }
+.nre-stat--warning .nre-stat-value { color: var(--warn); }
+
 /* ---- Tabs:静态 HTML 每 tab 一个 <details> 且仅首个 open(无 JS = 手风琴卡片,内容完整);
         enhance.js 在 <html> 上打 nre-js 后升级为单选 tab 条:收起的 tab 缩成一枚标题排在首行,
         唯一展开的 tab 以 100% 宽 + order:1 换行到下方,活动标题带高亮下划线、贴着面板顶部。

package/src/report/built-in/index.tsx

@@ -3,8 +3,8 @@
 // (docs/feature/reports/library/built-in.md)。新增内建视图 = 新文件 + 新具名导出,
 // 不需要注册表,也不改变装载管线。
 
-import { standard } from "./standard.tsx";
+import { standard, standardAttemptPage } from "./standard.tsx";
 
-export { standard };
+export { standard, standardAttemptPage };
 
 export default standard;

package/src/report/built-in/standard.tsx

@@ -1,14 +1,17 @@
-// standard —— 内建视图之一:报告 / Attempts / 追踪三页的通用结果站,没有任何私有钩子
-// (docs/feature/reports/library/built-in.md)。裸 `niceeval show` 与 `niceeval view`
-// 装载的就是它:一份普通 defineReport,与用户的 --report 文件同层、走同一条
+// standard —— 内建视图之一:报告 / Attempts / 追踪三页的通用结果站 + 一张隐藏的 attempt
+// 详情页,没有任何私有钩子(docs/feature/reports/library/built-in.md)。裸 `niceeval show` 与
+// `niceeval view` 装载的就是它:一份普通 defineReport,与用户的 --report 文件同层、走同一条
 // 装载 → resolve → validate → render 管线。「builtin」不是装载逻辑里的类别,
 // 只是宿主默认拿哪个值的事实;用户报告经 defineReport({ extends: standard, … }) 整站复用。
 //
-// 三页站点:report(Hero + 警告 + 批量修复 prompt + 实验对比)、attempts(带过滤的
-// attempt 全列表)、traces(执行瀑布)。裸宿主导航上能看到的一切内容都在这份定义里;
-// 宿主保留的只有机器(docs/feature/reports/architecture.md「宿主保留的只有机器」)。
+// 四页:report(Hero + 警告 + 批量修复 prompt + 实验对比)、attempts(带过滤的 attempt
+// 全列表)、traces(执行瀑布)三页进导航;第四张 standardAttemptPage 以 locator 为输入、
+// 不进导航,locator 深链打开它,content 就是公开的 AttemptDetail 组合组件。裸宿主导航上能
+// 看到的一切内容都在这份定义里;宿主保留的只有机器(docs/feature/reports/architecture.md
+// 「宿主保留的只有机器」)。
 
 import {
+  AttemptDetail,
   AttemptList,
   Col,
   CopyFixPrompt,
@@ -19,6 +22,14 @@ import {
   defineReport,
 } from "../index.ts";
 
+export const standardAttemptPage = {
+  id: "attempt",
+  title: "Attempt",
+  input: "attempt",
+  navigation: false,
+  content: <AttemptDetail />,
+} as const;
+
 export const standard = defineReport({
   pages: [
     {
@@ -55,5 +66,6 @@ export const standard = defineReport({
         </Col>
       ),
     },
+    standardAttemptPage,
   ],
 });