niceeval 0.5.4 → 0.6.1

package/src/view/app/lib/rows.ts

@@ -1,73 +1,17 @@
-// viewData(官方数据契约 + 快照明细)→ 页面模型的纯拼接,零聚合口径:
-// 榜单数字全部来自官方 MetricTable.data 的格子(server 侧算好),这里只把格子和
-// 对应 latest 快照的证据室明细(attempt 列表、判决计数)接在一起;Runs / Traces 的
-// 平铺列表来自全部快照(server 侧已跨快照去重)。
+// viewData 快照明细 → 证据室页面模型的纯拼接,零聚合口径:Runs / Traces 的平铺列表
+// 来自全部快照(server 侧已跨快照去重);统计口径整体住在报告槽的静态 HTML 里,
+// 前端不再拼榜单行。修复 prompt 的失败清单(CopyControls 的 fixPromptEntries)
+// 也从这份快照明细现算。
 
-import type { ReactNode } from "react";
-import type { RowRun, T } from "../shared.ts";
-import type { SortKey, ViewData, ViewResult, ViewRow, ViewSnapshot } from "../types.ts";
-import { evalLevelStats } from "../../../shared/outcome.ts";
-import { displayExperimentName, fallbackExperimentLabel, sumMaybe } from "../../../shared/aggregate.ts";
-import { formatConfigValue } from "./format.ts";
+import type { RowRun } from "../shared.ts";
+import type { ViewResult, ViewSnapshot } from "../types.ts";
+import { displayExperimentName } from "../../../shared/aggregate.ts";
 
-/** 默认报告格子的列键(= 内置指标的 metric.name,见 src/report/metrics.ts)。 */
-export const CELL_KEYS = {
-  passRate: "pass-rate",
-  duration: "duration",
-  tokens: "tokens",
-  cost: "cost",
-} as const;
-
-/** 快照的展示标签:experiment id 的末段;合成键退化到 agent/model 兜底。 */
-function snapshotLabel(snapshot: ViewSnapshot): string {
-  if (snapshot.synthetic) {
-    const first = snapshot.results[0];
-    return first ? fallbackExperimentLabel(first) : snapshot.experimentId;
-  }
+/** 快照的展示标签:experiment id 的末段。 */
+export function snapshotLabel(snapshot: ViewSnapshot): string {
   return displayExperimentName(snapshot.experimentId) ?? snapshot.experimentId;
 }
 
-function experimentGroup(snapshot: ViewSnapshot): string | undefined {
-  if (snapshot.synthetic || !snapshot.experimentId.includes("/")) return undefined;
-  return snapshot.experimentId.split("/").slice(0, -1).join("/");
-}
-
-/** 官方 TableData 行 × latest 快照 → 榜单行。行集合以官方表为准,快照只补证据室明细。 */
-export function buildRows(data: ViewData): ViewRow[] {
-  const latestByExperiment = new Map<string, ViewSnapshot>();
-  for (const snapshot of data.snapshots) {
-    if (snapshot.latest) latestByExperiment.set(snapshot.experimentId, snapshot);
-  }
-  const rows: ViewRow[] = [];
-  for (const tableRow of data.table.rows) {
-    const snapshot = latestByExperiment.get(tableRow.key);
-    if (!snapshot) continue; // 官方表与快照来自同一次挑选,理论上不缺;缺了宁可不渲染半行
-    const results = snapshot.results;
-    const stats = evalLevelStats(results, (r) => r.id);
-    rows.push({
-      key: tableRow.key,
-      experimentId: snapshot.experimentId,
-      ...(snapshot.synthetic ? { synthetic: true } : {}),
-      experiment: results[0]?.experiment,
-      group: experimentGroup(snapshot),
-      label: snapshotLabel(snapshot),
-      agent: snapshot.agent,
-      model: snapshot.model,
-      lastRunAt: snapshot.startedAt,
-      cells: tableRow.cells,
-      runs: results.length,
-      evals: stats.evals,
-      passed: stats.passed,
-      failed: stats.failed,
-      errored: stats.errored,
-      skipped: stats.skipped,
-      totalCostUSD: sumMaybe(results.map((r) => r.usage?.costUSD ?? r.estimatedCostUSD)),
-      results,
-    });
-  }
-  return rows;
-}
-
 /** Runs / Traces 的平铺列表:全部快照(含历史)的 attempt,带所属实验的展示标注。 */
 export function flattenAttempts(snapshots: ViewSnapshot[]): RowRun[] {
   return snapshots.flatMap((snapshot) => {
@@ -94,50 +38,3 @@ export function resultFromUrl(snapshots: ViewSnapshot[]): ViewResult | null {
   }
   return null;
 }
-
-export function buildGroupMap(rows: ViewRow[]): Map<string, ViewRow[]> {
-  const map = new Map<string, ViewRow[]>();
-  for (const row of rows) {
-    if (!row.group) continue;
-    if (!map.has(row.group)) map.set(row.group, []);
-    map.get(row.group)?.push(row);
-  }
-  return map;
-}
-
-export function compareRows(a: ViewRow, b: ViewRow, key: SortKey): number {
-  const av = valueFor(a, key);
-  const bv = valueFor(b, key);
-  if (typeof av === "string" || typeof bv === "string") return String(av).localeCompare(String(bv));
-  // 缺数据(null)沉底,与官方组件「缺数据不画 0」同一姿势。
-  if (av === null && bv === null) return 0;
-  if (av === null) return 1;
-  if (bv === null) return -1;
-  return Number(av) - Number(bv);
-}
-
-export function valueFor(row: ViewRow, key: SortKey): string | number | null {
-  if (key === "experiment") return row.label;
-  if (key === "model") return row.model || "";
-  if (key === "agent") return row.agent;
-  const cellKey =
-    key === "passRate" ? CELL_KEYS.passRate : key === "duration" ? CELL_KEYS.duration : key === "tokens" ? CELL_KEYS.tokens : CELL_KEYS.cost;
-  return row.cells[cellKey]?.value ?? null;
-}
-
-export function configChips(row: ViewRow, t: T): [string, ReactNode][] {
-  const exp = row.experiment || {};
-  const flags = exp.flags && Object.keys(exp.flags).length
-    ? Object.entries(exp.flags).map(([k, v]) => k + "=" + formatConfigValue(v)).join(", ")
-    : t("config.flagsNone");
-  return [
-    [t("config.experiment"), row.synthetic ? row.label : row.experimentId],
-    [t("table.model"), row.model || t("config.default")],
-    ["agent", row.agent],
-    ["runs", exp.runs ?? row.runs],
-    ["earlyExit", exp.earlyExit === undefined ? t("config.notApplicable") : String(exp.earlyExit)],
-    ["sandbox", exp.sandbox || t("config.default")],
-    ["budget", exp.budget === undefined ? t("config.none") : "$" + exp.budget],
-    ["flags", flags],
-  ];
-}

package/src/view/app/lib/verdict.ts

@@ -0,0 +1,25 @@
+import type { Assertion, Verdict, ViewResult } from "../types.ts";
+import type { T } from "../shared.ts";
+
+export function verdictClass(verdict: Verdict): string {
+  return verdict === "passed" ? "good" : verdict === "errored" ? "infra-err" : verdict === "failed" ? "bad" : "warn";
+}
+
+export function verdictLabel(verdict: Verdict, t: T): string {
+  if (verdict === "passed") return t("status.pass");
+  if (verdict === "failed") return t("status.fail");
+  if (verdict === "errored") return t("status.error");
+  if (verdict === "skipped") return t("status.skipped");
+  return verdict || "—";
+}
+
+// Only gate-severity failures are eval "failure reasons"; soft failures show as scores
+export function failingAssertions(result: ViewResult): Assertion[] {
+  return (result.assertions || []).filter((a: Assertion) => !a.passed && a.severity === "gate");
+}
+
+export function reasonFor(result: ViewResult, failedGates: Assertion[]): string {
+  if (result.error) return result.error;
+  if (result.skipReason) return result.skipReason;
+  return failedGates.map((a: Assertion) => (a.detail ? `${a.name}: ${a.detail}` : a.name)).join(", ");
+}

package/src/view/app/main.tsx

@@ -1,26 +1,23 @@
 import { createRoot } from "react-dom/client";
-import type { ViewData } from "./types.ts";
+import type { ReportSlotHtml, ViewData } from "./types.ts";
 import { App } from "./App.tsx";
 import "../styles.css";
 
-// 没有烘焙数据(比如直接打开裸产物)时的空页面兜底,形状与官方数据契约一致。
+// 没有烘焙数据(比如直接打开裸产物)时的空页面兜底。
 const emptyViewData: ViewData = {
   composedRuns: 0,
-  overview: {
-    snapshots: [],
-    totals: { evals: 0, attempts: 0, passed: 0, failed: 0, errored: 0, skipped: 0, costUSD: null, durationMs: 0 },
-    warnings: [],
-  },
-  table: { dimension: "experiment", columns: [], rows: [] },
-  overall: { dimension: "overall", columns: [], rows: [] },
   snapshots: [],
 };
 
 const initialData: ViewData = window.__NICEEVAL_VIEW_DATA__ ?? emptyViewData;
 
-// --report:server 把报告树的静态 HTML 烘成 <template id="niceeval-report"> 块
-// (__NICEEVAL_VIEW_DATA__ 旁的静态块)。前端只负责把它摆进报告槽位置,不解析、不 hydrate。
-const reportHtml = document.getElementById("niceeval-report")?.innerHTML;
+// 报告槽:server 把报告树的静态 HTML 烘成 en / zh-CN 两个 <template> 静态块
+// (__NICEEVAL_VIEW_DATA__ 旁)。前端按当前界面语言把对应块摆进报告槽位置,
+// 切语言即换块;不解析、不 hydrate。
+const reportHtml: ReportSlotHtml = {
+  en: document.getElementById("niceeval-report-en")?.innerHTML ?? "",
+  "zh-CN": document.getElementById("niceeval-report-zh-CN")?.innerHTML ?? "",
+};
 
 const rootEl = document.getElementById("root");
 if (!rootEl) throw new Error("Missing #root element");

package/src/view/app/pages/RunsPage.tsx

@@ -1,6 +1,6 @@
 import { useState } from "react";
 import type { RowRun, T } from "../shared.ts";
-import { outcomeClass, outcomeLabel } from "../lib/outcome.ts";
+import { verdictClass, verdictLabel } from "../lib/verdict.ts";
 import { formatCost, formatDateTime, formatDuration, formatTokens, totalTokens } from "../lib/format.ts";
 
 /** 全部历史 attempt 打平成一张表;列表在 App 里由 flattenAttempts(全部快照,已跨快照去重)算好。 */
@@ -35,7 +35,7 @@ export function RunsView({ attempts, t }: { attempts: RowRun[]; t: T }) {
               <tr>
                 <th>{t("table.evalId")}</th>
                 <th>{t("table.experiment")}</th>
-                <th>{t("table.outcome")}</th>
+                <th>{t("table.verdict")}</th>
                 <th>{t("table.agent")}</th>
                 <th>{t("table.model")}</th>
                 <th>{t("metric.duration")}</th>
@@ -47,14 +47,14 @@ export function RunsView({ attempts, t }: { attempts: RowRun[]; t: T }) {
             <tbody>
               {filtered.length ? (
                 filtered.map((r: RowRun) => {
-                  const outcome = r.outcome;
+                  const verdict = r.verdict;
                   return (
                     <tr key={`${r.id}-${r.rowLabel}-${r.attempt}`}>
                       <td>
                         <span className="name">{r.id}</span>
                       </td>
                       <td>{r.rowLabel}</td>
-                      <td className={outcomeClass(outcome)}>{outcomeLabel(outcome, t)}</td>
+                      <td className={verdictClass(verdict)}>{verdictLabel(verdict, t)}</td>
                       <td>{r.rowAgent}</td>
                       <td>{r.rowModel || t("config.default")}</td>
                       <td className="num">{formatDuration(r.durationMs)}</td>

package/src/view/app/pages/TracesPage.tsx

@@ -1,5 +1,5 @@
 import type { RowRun, T } from "../shared.ts";
-import { outcomeClass, outcomeLabel } from "../lib/outcome.ts";
+import { verdictClass, verdictLabel } from "../lib/verdict.ts";
 import { formatDuration } from "../lib/format.ts";
 import { LazyArtifact } from "../components/LazyArtifact.tsx";
 
@@ -15,11 +15,11 @@ export function TracesView({ attempts, t }: { attempts: RowRun[]; t: T }) {
         <div className="empty">{t("empty.traces")}</div>
       ) : (
         traceable.map((r: RowRun) => {
-          const outcome = r.outcome;
+          const verdict = r.verdict;
           return (
             <div className="traces-entry" key={`${r.id}-${r.rowLabel}-${r.attempt}`}>
               <div className="traces-entry-head">
-                <span className={`${outcomeClass(outcome)} traces-verdict`}>{outcomeLabel(outcome, t)}</span>
+                <span className={`${verdictClass(verdict)} traces-verdict`}>{verdictLabel(verdict, t)}</span>
                 <span className="eval-id">{r.id}</span>
                 <span className="traces-exp">{r.rowLabel}</span>
                 <span className="num traces-dur">{formatDuration(r.durationMs)}</span>

package/src/view/app/shared.ts

@@ -3,7 +3,6 @@ import type { MessageKey } from "./i18n.ts";
 import type { CodeSource, TranscriptEvent, ViewJson, ViewResult } from "./types.ts";
 
 export type T = (key: MessageKey, vars?: Vars) => string;
-export type OpenModal = (result: ViewResult) => void;
 export type ArtifactLoadState =
   | { sources: CodeSource[] | null; events: TranscriptEvent[] | null; status: "loading" | "ready" | "none" };
 export type RowRun = ViewResult & { rowLabel: string; rowAgent: string; rowModel?: string };

package/src/view/app/types.ts

@@ -1,6 +1,5 @@
 import type {
   AssertionResult,
-  ExperimentRunInfo,
   JsonValue,
   LocalizedText,
   SourceArtifact,
@@ -9,71 +8,27 @@ import type {
   TraceSpan,
   Usage,
 } from "../../types.ts";
-import type { MetricCell, ViewData, ViewEvalResult } from "../shared/types.ts";
+import type { ViewData, ViewEvalResult } from "../shared/types.ts";
 
 export type { LocalizedText };
 // Locale 只在 i18n 内核声明一次;页面数据形状与 server 共用 shared/types.ts 的声明。
 export type { Locale } from "../../i18n/core.ts";
 export type {
-  AttemptRef,
-  MetricCell,
-  MetricColumn,
-  OverviewData,
-  SelectionWarning,
+  AttemptLocator,
+  ReportSlotHtml,
   SkippedRunNotice,
-  TableData,
   ViewData,
   ViewSnapshot,
 } from "../shared/types.ts";
 
-export type Tab = "report" | "experiments" | "runs" | "traces";
-export type SortKey = "experiment" | "model" | "agent" | "duration" | "passRate" | "tokens" | "cost";
-export type SortDir = 1 | -1;
+export type Tab = "report" | "runs" | "traces";
 
-/**
- * Experiments 榜单一行 = 一个实验的最新快照(latest 口径)。
- * 统计数字(cells)是官方 MetricTable.data 的产物,前端只渲染;
- * 明细(results / 判决计数)是该快照的证据室数据,由 buildRows 从 viewData 拼接。
- */
-export interface ViewRow {
-  /** = experimentId(官方 TableData 行键)。 */
-  key: string;
-  experimentId: string;
-  /** experimentId 是 "<agent>/<model>" 合成键时为 true(落盘缺 experimentId)。 */
-  synthetic?: boolean;
-  experiment?: ExperimentRunInfo;
-  group?: string;
-  label: string;
-  agent: string;
-  model?: string;
-  /** 该快照的 startedAt(ISO)—— 榜单「每行带判决产生的时间」。 */
-  lastRunAt: string;
-  /** 默认报告格子(pass-rate / duration / tokens / cost),display 已格式化。 */
-  cells: Partial<Record<string, MetricCell>>;
-  /** 快照内 attempt 总数。 */
-  runs: number;
-  /** 折叠后的 eval 判决计数(钻取明细的展示口径,与 CLI 共用同一份折叠实现)。 */
-  evals: number;
-  passed: number;
-  failed: number;
-  errored: number;
-  skipped: number;
-  /** 快照内成本合计(证据室口径的总花费;榜单列是官方的「平均每 eval 成本」格子)。 */
-  totalCostUSD?: number;
-  results: ViewEvalResult[];
-}
-
-export interface SortState {
-  key: SortKey;
-  dir: SortDir;
-}
-
-/** 前端拿到的单条 attempt 结果就是瘦身后的 EvalResult(attemptRef / artifactBase 由 loader 注入)。 */
+/** 前端拿到的单条 attempt 结果就是瘦身后的 EvalResult(locator / artifactBase 由 loader 注入)。 */
 export type ViewResult = ViewEvalResult;
 
 export type Assertion = AssertionResult;
 
-export type Outcome = "passed" | "failed" | "errored" | "skipped" | string;
+export type Verdict = "passed" | "failed" | "errored" | "skipped" | string;
 
 export interface SourceTurn {
   loc?: SourceLoc;

package/src/view/artifact-serving.test.ts

@@ -0,0 +1,136 @@
+// 回归覆盖:src/results/open.ts 把 sources.json 从「全量 SourceArtifact[]」改成两层去重存储
+// (attempt 级引用 `{path, sha256}[]` + 快照根 `sources/<sha256>.json` 内容仓库,见 memory/
+// attempt-locator-and-source-dedup.md)之后,view 的两条 artifact 出口(server.ts 的
+// serveArtifact、index.ts 的 copyFetchedArtifacts)曾各自继续把落盘的引用格式原样吐给浏览器
+// (readFile 直传 / copyFile 直拷),而不是走 AttemptHandle.sources() 解引用。浏览器端
+// isCodeSource 守卫要求 `content: string` 字段,引用格式没有,代码视图因此静默判空——
+// 即便 sources 明明被捕获了(memory/static-site-export-drops-sources.md 记录的是同一症状的
+// 更早一次事故,根因不同:那次是文件压根没导出,这次是导出了但格式不对)。
+//
+// 两个测试都特意让两个 attempt 共享字节相同的 eval 源码,确保真的走了去重路径(sha256 引用 +
+// 共享 blob),而不是巧合地测到「本来就没有第二层」的场景。
+
+import { mkdtemp, readFile, readdir, rm } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { afterEach, describe, expect, it } from "vitest";
+import { createResultsWriter } from "../results/index.ts";
+import { loadViewScan } from "./data.ts";
+import { buildView } from "./index.ts";
+import { startViewServer, type ViewServer } from "./server.ts";
+import { artifactUrl } from "./app/lib/artifact-url.ts";
+
+const SHARED_CONTENT = "export default { test() {} };\n";
+const SHARED_PATH = "evals/shared.eval.ts";
+
+const roots: string[] = [];
+async function makeRoot(): Promise<string> {
+  const root = await mkdtemp(join(tmpdir(), "niceeval-artifact-serving-"));
+  roots.push(root);
+  return root;
+}
+afterEach(async () => {
+  await Promise.all(roots.splice(0).map((r) => rm(r, { recursive: true, force: true })));
+});
+
+/**
+ * 写一份真实快照:两个 attempt(q1/q2)共享字节相同的 eval 源码触发去重仓库,
+ * 外加一个 attempt(q3)带 events.json(用于确认 events/trace 这类 artifact 不受影响)。
+ * 走真实 writer(createResultsWriter),不手写落盘格式——这正是 sources 去重两层存储
+ * 唯一的生产写入路径。
+ */
+async function seedDedupedSnapshot(root: string): Promise<{ snapDir: string }> {
+  const writer = createResultsWriter(root, { producer: { name: "niceeval", version: "1.0.0" } });
+  const snap = await writer.snapshot({ experimentId: "e", agent: "bub", startedAt: "2026-07-01T08:00:00.000Z" });
+  await snap.writeAttempt(
+    { id: "q1", verdict: "passed", attempt: 0, durationMs: 1, assertions: [] },
+    { sources: [{ path: SHARED_PATH, content: SHARED_CONTENT }] },
+  );
+  await snap.writeAttempt(
+    { id: "q2", verdict: "passed", attempt: 0, durationMs: 1, assertions: [] },
+    { sources: [{ path: SHARED_PATH, content: SHARED_CONTENT }] },
+  );
+  await snap.writeAttempt(
+    { id: "q3", verdict: "passed", attempt: 0, durationMs: 1, assertions: [] },
+    { events: [{ type: "message", role: "user", text: "hi" }] },
+  );
+  await writer.finish();
+
+  // 前提断言:去重真的生效了(否则这份 fixture 没有覆盖到引用/仓库两层结构,回归测试会
+  // 在「碰巧没有第二层」的场景下通过,失去意义)。
+  const storeFiles = await readdir(join(snap.dir, "sources"));
+  expect(storeFiles).toHaveLength(1);
+  const q1Ref = JSON.parse(await readFile(join(snap.dir, "q1", "a0", "sources.json"), "utf-8"));
+  expect(q1Ref).toEqual([{ path: SHARED_PATH, sha256: expect.any(String) }]);
+  expect(q1Ref[0]).not.toHaveProperty("content"); // 落盘就是引用,不含内容——下面断言浏览器出口必须解引用
+
+  return { snapDir: snap.dir };
+}
+
+describe("server.ts · serveArtifact 对 sources.json 解引用", () => {
+  let server: ViewServer | undefined;
+  afterEach(async () => {
+    await server?.close();
+    server = undefined;
+  });
+
+  it("HTTP /artifact/<base>/sources.json 返回解引用后的 {path,content}[],不是落盘的 {path,sha256}[] 引用", async () => {
+    const root = await makeRoot();
+    await seedDedupedSnapshot(root);
+
+    const scan = await loadViewScan(root);
+    const byId = new Map(scan.viewData.snapshots.flatMap((s) => s.results.map((r) => [r.id, r])));
+    const q1 = byId.get("q1")!;
+    const q2 = byId.get("q2")!;
+    expect(q1.hasSources).toBe(true);
+    expect(q2.hasSources).toBe(true);
+
+    server = await startViewServer({ input: root });
+
+    for (const result of [q1, q2]) {
+      const url = new URL(artifactUrl(`${result.artifactBase}/sources.json`), server.url);
+      const res = await fetch(url);
+      expect(res.status).toBe(200);
+      const body = await res.json();
+      expect(body).toEqual([{ path: SHARED_PATH, content: SHARED_CONTENT }]);
+      expect(body[0]).not.toHaveProperty("sha256");
+    }
+  });
+
+  it("events.json 仍走原文件直传,不受 sources 解引用改动影响", async () => {
+    const root = await makeRoot();
+    await seedDedupedSnapshot(root);
+    const scan = await loadViewScan(root);
+    const q3 = scan.viewData.snapshots.flatMap((s) => s.results).find((r) => r.id === "q3")!;
+    expect(q3.hasEvents).toBe(true);
+
+    server = await startViewServer({ input: root });
+    const url = new URL(artifactUrl(`${q3.artifactBase}/events.json`), server.url);
+    const res = await fetch(url);
+    expect(res.status).toBe(200);
+    expect(await res.json()).toEqual([{ type: "message", role: "user", text: "hi" }]);
+  });
+});
+
+describe("index.ts · copyFetchedArtifacts(--out 静态导出)对 sources.json 解引用", () => {
+  it("导出目录里的 sources.json 是解引用后的 {path,content}[],不是落盘引用;events.json 仍原字节复制", async () => {
+    const root = await makeRoot();
+    await seedDedupedSnapshot(root);
+
+    const out = join(root, "site");
+    await buildView({ input: root, out });
+
+    const scan = await loadViewScan(root);
+    const byId = new Map(scan.viewData.snapshots.flatMap((s) => s.results.map((r) => [r.id, r])));
+    for (const id of ["q1", "q2"]) {
+      const base = byId.get(id)!.artifactBase!;
+      const exported = JSON.parse(await readFile(join(out, "artifact", base, "sources.json"), "utf-8"));
+      expect(exported).toEqual([{ path: SHARED_PATH, content: SHARED_CONTENT }]);
+      expect(exported[0]).not.toHaveProperty("sha256");
+    }
+
+    const q3Base = byId.get("q3")!.artifactBase!;
+    const exportedEvents = JSON.parse(await readFile(join(out, "artifact", q3Base, "events.json"), "utf-8"));
+    expect(exportedEvents).toEqual([{ type: "message", role: "user", text: "hi" }]);
+  });
+});