niceeval 0.6.1 → 0.7.0

package/src/report/react/GroupSummary.tsx

@@ -1,66 +0,0 @@
-// GroupSummary:一组 experiment(如自定义报告里同一 <Section> 内的全部 experiment)的
-// 紧凑摘要块——恢复旧 GroupSelector 卡片的信息密度(通过率、experiment/eval 数、failed、
-// errored、总成本、最后运行时间),但只渲染 GroupSummaryData 算好的 MetricCell / 计数,
-// 不现场重算比例(docs/feature/reports/library.md「GroupSummary」)。errored 为 0 时省略该片段(旧卡片同一姿态),
-// 但这只是渲染取舍——数据侧 verdicts.errored 字段本身永不省略。
-
-import type { ReactElement } from "react";
-import type { GroupSummaryData } from "../types.ts";
-import { DEFAULT_REPORT_LOCALE, localeText, type ReportLocale } from "../locale.ts";
-import { MetricCellView } from "./cell.tsx";
-import { cx, formatUSD } from "./format.ts";
-
-export function GroupSummary({
-  data,
-  className,
-  locale = DEFAULT_REPORT_LOCALE,
-}: {
-  data: GroupSummaryData;
-  className?: string;
-  locale?: ReportLocale;
-}): ReactElement {
-  const missing = <span className="nre-missing">{localeText(locale, "cell.missing")}</span>;
-  return (
-    <div className={cx("nre", "nre-group-summary", className)}>
-      <dl className="nre-group-kpis">
-        <div className="nre-group-kpi nre-group-kpi-rate">
-          <dt>{localeText(locale, "overview.passRate")}</dt>
-          <dd>
-            <MetricCellView cell={data.passRate} locale={locale} />
-          </dd>
-        </div>
-        <div className="nre-group-kpi">
-          <dt>{localeText(locale, "groupSummary.experiments")}</dt>
-          <dd>{data.experiments}</dd>
-        </div>
-        <div className="nre-group-kpi">
-          <dt>{localeText(locale, "overview.evals")}</dt>
-          <dd>{data.evals}</dd>
-        </div>
-        <div className="nre-group-kpi">
-          <dt>{localeText(locale, "overview.attempts")}</dt>
-          <dd>{data.attempts}</dd>
-        </div>
-        <div className="nre-group-kpi">
-          <dt>{localeText(locale, "verdict.failed")}</dt>
-          <dd className="nre-verdict-failed">{data.verdicts.failed}</dd>
-        </div>
-        {/* errored 为 0 省略这一片段(旧 GroupSelector 卡片同一姿态);数据字段本身不受影响 */}
-        {data.verdicts.errored > 0 && (
-          <div className="nre-group-kpi">
-            <dt>{localeText(locale, "verdict.errored")}</dt>
-            <dd className="nre-verdict-errored">{data.verdicts.errored}</dd>
-          </div>
-        )}
-        <div className="nre-group-kpi">
-          <dt>{localeText(locale, "overview.totalCost")}</dt>
-          {/* totalCostUSD 全缺 = null:显示缺数据,不编 $0 */}
-          <dd>{data.totalCostUSD === null ? missing : formatUSD(data.totalCostUSD)}</dd>
-        </div>
-      </dl>
-      {data.lastRunAt && (
-        <p className="nre-group-summary-time">{localeText(locale, "latestRun", { run: data.lastRunAt })}</p>
-      )}
-    </div>
-  );
-}

package/src/report/react/RunOverview.tsx

@@ -1,109 +0,0 @@
-// RunOverview:页头 KPI 条——「这批数据是什么」。
-// 数字下面标注数据来源(几个快照、何时跑的);warnings 有内容时直接显示在条内,
-// 诚实不靠使用者记得渲染(docs/feature/reports/library.md「RunOverview」行为清单)。
-
-import type { ReactElement } from "react";
-import type { OverviewData } from "../types.ts";
-import { DEFAULT_REPORT_LOCALE, localeText, type ReportLocale } from "../locale.ts";
-import { cx, formatDurationMs, formatUSD } from "./format.ts";
-
-export function RunOverview({
-  data,
-  className,
-  locale = DEFAULT_REPORT_LOCALE,
-}: {
-  data: OverviewData;
-  className?: string;
-  locale?: ReportLocale;
-}): ReactElement {
-  const { totals } = data;
-  const missing = <span className="nre-missing">{localeText(locale, "cell.missing")}</span>;
-  // 通过率只渲染 computeCell 算好的 MetricCell,不现场重算——覆盖率角标复用
-  // MetricCellView 的缺数据/coverage 语义(缺数据 = 显式文案,绝不画 0%)
-  const { passRate } = totals;
-  const passRateNode =
-    passRate.value === null ? (
-      missing
-    ) : (
-      <>
-        {passRate.display}
-        {passRate.samples < passRate.total && (
-          <sup
-            className="nre-coverage"
-            title={localeText(locale, "cell.coverageTitle", { samples: passRate.samples, total: passRate.total })}
-          >
-            {passRate.samples}/{passRate.total}
-          </sup>
-        )}
-      </>
-    );
-
-  return (
-    <header className={cx("nre", "nre-overview", className)}>
-      <dl className="nre-kpis">
-        <div className="nre-kpi">
-          <dt>{localeText(locale, "overview.snapshots")}</dt>
-          <dd>{data.snapshots.length}</dd>
-        </div>
-        <div className="nre-kpi">
-          <dt>{localeText(locale, "overview.evals")}</dt>
-          <dd>{totals.evals}</dd>
-        </div>
-        <div className="nre-kpi">
-          <dt>{localeText(locale, "overview.attempts")}</dt>
-          <dd>{totals.attempts}</dd>
-        </div>
-        <div className="nre-kpi">
-          <dt>{localeText(locale, "overview.passRate")}</dt>
-          <dd>{passRateNode}</dd>
-        </div>
-        <div className="nre-kpi">
-          <dt>{localeText(locale, "overview.totalCost")}</dt>
-          {/* costUSD 全缺 = null:显示缺数据,不编 $0 */}
-          <dd>{totals.costUSD === null ? missing : formatUSD(totals.costUSD)}</dd>
-        </div>
-        <div className="nre-kpi">
-          <dt>{localeText(locale, "overview.totalDuration")}</dt>
-          <dd>{formatDurationMs(totals.durationMs)}</dd>
-        </div>
-      </dl>
-
-      <p className="nre-verdicts">
-        <span className="nre-verdict nre-verdict-passed">
-          {localeText(locale, "verdict.passed")} {totals.passed}
-        </span>
-        <span className="nre-verdict nre-verdict-failed">
-          {localeText(locale, "verdict.failed")} {totals.failed}
-        </span>
-        <span className="nre-verdict nre-verdict-errored">
-          {localeText(locale, "verdict.errored")} {totals.errored}
-        </span>
-        <span className="nre-verdict nre-verdict-skipped">
-          {localeText(locale, "verdict.skipped")} {totals.skipped}
-        </span>
-      </p>
-
-      {/* 数据来源:哪些快照、何时跑的——报告的数字都从这里来 */}
-      <p className="nre-source">
-        {localeText(locale, "overview.source", { n: data.snapshots.length })}
-        {data.snapshots.map((s) => (
-          <span key={`${s.experimentId}@${s.startedAt}`} className="nre-source-snapshot">
-            {s.experimentId}({s.agent}
-            {s.model ? ` · ${s.model}` : ""})@ {s.startedAt}
-          </span>
-        ))}
-      </p>
-
-      {/* Selection 的警告(残缺快照等)直接渲染在条内,不静默;结构化字段供程序判断,这里打 message */}
-      {data.warnings.length > 0 && (
-        <ul className="nre-warnings">
-          {data.warnings.map((w, i) => (
-            <li key={i} className="nre-warning" data-kind={w.kind}>
-              {w.message}
-            </li>
-          ))}
-        </ul>
-      )}
-    </header>
-  );
-}

package/src/runner/reporters/console.ts

@@ -1,70 +0,0 @@
-// 控制台报告器:流式逐行输出,失败断言内联展开,末尾出效率三件套(时间 / token / $)。
-
-import type { EvalResult, Reporter, RunSummary } from "../../types.ts";
-import { t } from "../../i18n/index.ts";
-import { formatCost, formatDuration, formatTokens, renderRunReport } from "./table.ts";
-import { verdictSymbol } from "./shared.ts";
-
-export function Console(): Reporter {
-  return {
-    onRunStart(evals, _agent, shape) {
-      // compare(多 agent / 多 model)或 runs>1 时,实际 attempt 数 > eval 数。头部如实报清,
-      // 否则「本次运行 5 个 eval」会和末尾「5 passed, 5 failed」(按 attempt 计 10)对不上。
-      const n = shape?.evals ?? evals.length;
-      const extra =
-        shape && shape.totalRuns > n
-          ? t("report.runStartExtra", { configs: shape.configs, totalRuns: shape.totalRuns })
-          : "";
-      process.stdout.write(t("report.runStart", { count: n, extra, concurrency: shape?.maxConcurrency ?? "?" }));
-    },
-    onEvalComplete(result: EvalResult) {
-      const sym = verdictSymbol(result.verdict);
-      const tok = (result.usage?.inputTokens ?? 0) + (result.usage?.outputTokens ?? 0);
-      // requests > 0 但 tokens = 0 → agent 跑了但不上报用量(如 bub);显示 — 而非误导性的 0
-      const tokStr = tok > 0 ? `${formatTokens(tok)} tok` : (result.usage?.requests ?? 0) > 0 ? `— tok` : `0 tok`;
-      const cost = result.estimatedCostUSD !== undefined ? `  ${formatCost(result.estimatedCostUSD)}` : "";
-      const who = result.model ? `${result.agent}/${result.model}` : result.agent;
-      const meta = `(${formatDuration(result.durationMs)}  ${tokStr}${cost})`;
-      const label = result.verdict === "passed" ? "" : ` ${formatVerdict(result.verdict)}`;
-      process.stdout.write(`  ${sym} ${result.id}${label}  [${who}]  ${meta}\n`);
-
-      if (result.skipReason) {
-        process.stdout.write(`      ○ ${t("report.skipped")}: ${result.skipReason}\n`);
-      }
-      if (result.error) {
-        process.stdout.write(`      ! ${t("report.error")}: ${truncate(result.error, 400)}\n`);
-      }
-      let lastGroup: string | undefined;
-      for (const a of result.assertions) {
-        if (a.passed) continue;
-        if (a.group !== undefined && a.group !== lastGroup) {
-          process.stdout.write(`      ▸ ${a.group}\n`);
-        }
-        lastGroup = a.group;
-        const sev = a.severity === "gate" ? t("report.gate") : t("report.soft");
-        const thr = a.threshold !== undefined
-          ? t("report.assertionThreshold", { score: a.score.toFixed(2), threshold: a.threshold })
-          : "";
-        const indent = a.group !== undefined ? "        " : "      ";
-        process.stdout.write(`${indent}- ${sev}: ${a.name}${thr}${a.detail ? ` — ${truncate(a.detail, 300)}` : ""}\n`);
-      }
-    },
-    onRunComplete(summary: RunSummary) {
-      process.stdout.write(renderRunReport(summary));
-    },
-  };
-}
-
-function formatVerdict(verdict: string): string {
-  switch (verdict) {
-    case "passed": return t("report.passed");
-    case "failed": return t("report.failed");
-    case "errored": return t("report.errored");
-    case "skipped": return t("report.skipped");
-    default: return verdict;
-  }
-}
-
-function truncate(s: string, n: number): string {
-  return s.length > n ? s.slice(0, n) + "…" : s;
-}

package/src/runner/reporters/live.test.ts

@@ -1,56 +0,0 @@
-import { afterEach, describe, expect, it, vi } from "vitest";
-import { Live, type LiveRow } from "./live.ts";
-
-function withMockTty<T>(fn: () => T): { writes: string[]; result: T } {
-  const writes: string[] = [];
-  vi.spyOn(process.stderr, "write").mockImplementation((chunk: unknown) => {
-    writes.push(String(chunk));
-    return true;
-  });
-  Object.defineProperty(process.stderr, "isTTY", { value: true, configurable: true });
-  Object.defineProperty(process.stderr, "columns", { value: 120, configurable: true });
-  Object.defineProperty(process.stderr, "rows", { value: 40, configurable: true });
-  return { writes, result: fn() };
-}
-
-describe("Live carried rows", () => {
-  afterEach(() => {
-    vi.restoreAllMocks();
-  });
-
-  it("renders a carried row as already-done from the first frame, not waiting for a slot", () => {
-    const rows: LiveRow[] = [
-      { evalId: "memory/carried-fail", who: "codex-e2b", total: 1, carriedVerdict: "failed" },
-      { evalId: "memory/fresh", who: "codex-e2b", total: 1 },
-    ];
-    const live = Live(rows, 2);
-
-    const { writes } = withMockTty(() => {
-      live.onRunStart?.([], { name: "codex", kind: "sandbox" } as never, {
-        evals: 2,
-        configs: 1,
-        totalRuns: 1,
-        maxConcurrency: 5,
-      });
-    });
-    vi.spyOn(process.stdout, "write").mockImplementation(() => true);
-    live.onRunComplete?.({
-      startedAt: "",
-      finishedAt: "",
-      durationMs: 0,
-      results: [],
-      totalRuns: 1,
-    } as never);
-
-    const out = writes.join("");
-    const carriedLine = out.split("\n").find((l) => l.includes("memory/carried-fail"));
-    const freshLine = out.split("\n").find((l) => l.includes("memory/fresh"));
-
-    expect(carriedLine).toBeDefined();
-    expect(carriedLine).toContain("✗");
-    expect(carriedLine).not.toContain("waiting");
-
-    expect(freshLine).toBeDefined();
-    expect(freshLine).toContain("waiting");
-  });
-});

package/src/runner/reporters/live.ts

@@ -1,247 +0,0 @@
-// Live terminal reporter:在 TTY 终端里渲染实时状态表,每个 (eval, who) 对占一行。
-// spinner 每 80ms 刷新;attempt 完成后行内显示 ✓/✗/~ 符号。
-// onRunComplete 时清除状态表,打印和网页榜单同口径的表格报告。
-
-import type { Reporter, ReporterEvent, RunShape, RunSummary } from "../../types.ts";
-import { t } from "../../i18n/index.ts";
-import { renderRunReport } from "./table.ts";
-import { verdictSymbol, WAITING_SYM } from "./shared.ts";
-import { runWho } from "../types.ts";
-import { onBeforeExternalTerminalWrite } from "../../tty-line.ts";
-
-const SPINNER = ["⠋", "⠙", "⠹", "⠸", "⠼", "⠴", "⠦", "⠧", "⠇", "⠏"];
-
-// 进度日志里的基础设施噪声:OTLP 端口、remote-agent 启动提示、trace span 计数。
-// 这些在行尾显示毫无意义,直接丢弃。
-const NOISE_PATTERNS = [
-  /^OTLP /,
-  /^使用 remote agent/,
-  /^using remote agent/,
-  /^驱动 agent/,
-  /^driving agent/,
-  /^trace:\d/,
-  /^agent tracing/,
-  /^agent setup/,
-];
-
-function isNoise(msg: string): boolean {
-  return NOISE_PATTERNS.some((p) => p.test(msg));
-}
-
-export interface LiveRow {
-  evalId: string;
-  who: string;
-  /** 该 (evalId, who) 对的总 attempt 数(含 earlyExit 估算值)。 */
-  total: number;
-  /**
-   * 设置时表示这一行从第一帧起就是携入(carry)的结果,不会真的调度 attempt——直接按这个
-   * verdict 渲染成已完成,不经过 waiting → spinner 的过程。没有它,carry 掉的行会卡在
-   * "waiting for a slot"直到进程退出:它永远等不到 eval:start,因为 run.ts 压根不会为
-   * 它派发 attempt。
-   */
-  carriedVerdict?: string;
-}
-
-export interface LiveReporter extends Reporter {
-  /** 被 RunOptions.onProgress 调用,更新行尾的 lastMsg。 */
-  progress(evalId: string, who: string, msg: string): void;
-}
-
-interface RowState extends LiveRow {
-  completed: number;
-  lastMsg: string;
-  dominantVerdict: string | undefined;
-  /** true 一旦拿到并发名额、attempt effect 真正开始跑;之前是排队等待,不该转圈误导。 */
-  started: boolean;
-}
-
-export function Live(rows: LiveRow[], totalAttempts: number): LiveReporter {
-  const stateMap = new Map<string, RowState>();
-  const keyOrder: string[] = [];
-  let totalCompleted = 0;
-
-  for (const r of rows) {
-    const k = `${r.evalId}|${r.who}`;
-    if (!stateMap.has(k)) {
-      const carried = r.carriedVerdict !== undefined;
-      stateMap.set(k, {
-        ...r,
-        completed: carried ? r.total : 0,
-        lastMsg: "",
-        dominantVerdict: r.carriedVerdict,
-        started: carried,
-      });
-      keyOrder.push(k);
-      if (carried) totalCompleted += r.total;
-    } else {
-      // 同一 (evalId, who) 可能在多个 agentRun 里出现(不应发生,但做防御)
-      stateMap.get(k)!.total += r.total;
-    }
-  }
-
-  let spinFrame = 0;
-  let drawnLines = 0; // 上次 draw() 写了多少行,用于 \x1B[nA 回跳
-  let intervalId: ReturnType<typeof setInterval> | undefined;
-  let shape: RunShape | undefined;
-  let unsubscribeExternalWrite: (() => void) | undefined;
-
-  const cols = () => process.stderr.columns || 100;
-
-  function renderRow(state: RowState, frame: number): string {
-    const done = state.completed >= state.total;
-    const sym = done
-      ? verdictSymbol(state.dominantVerdict ?? "")
-      : state.started
-        ? SPINNER[frame % SPINNER.length]
-        : WAITING_SYM;
-
-    const evalCol = state.evalId.slice(0, 24).padEnd(24);
-    const whoCol = `[${state.who}]`.slice(0, 26).padEnd(26);
-    const cntCol = `${state.completed}/${state.total}`.padEnd(5);
-
-    const prefix = `  ${sym} ${evalCol} ${whoCol} ${cntCol}  `;
-    const budget = Math.max(0, cols() - prefix.length - 1);
-    const msg = done ? "" : state.started ? state.lastMsg.slice(0, budget) : t("live.waiting").slice(0, budget);
-
-    return `\x1B[2K${prefix}${msg}`;
-  }
-
-  function renderHeader(): string {
-    const hdr = shape
-      ? t("live.running", {
-          totalRuns: shape.totalRuns,
-          evals: shape.evals,
-          configs: shape.configs,
-          concurrency: shape.maxConcurrency,
-          completed: totalCompleted,
-          total: totalAttempts,
-        })
-      : t("live.runningUnknown", { completed: totalCompleted, total: totalAttempts });
-    return `\x1B[2K${hdr}`;
-  }
-
-  // 终端放不下全部行时,光标回跳会被屏幕顶端截断,导致每帧往下追加整表。
-  // 所以按终端高度截断:优先显示运行中的行,放不下的折叠成一行摘要。
-  function frameLines(frame: number): string[] {
-    const lines = [renderHeader()];
-    const termRows = process.stderr.rows || 30;
-    // 预留:1 行表头 + 1 行防止末尾换行触发滚动
-    const budget = Math.max(1, termRows - 2);
-
-    if (keyOrder.length <= budget) {
-      for (const k of keyOrder) lines.push(renderRow(stateMap.get(k)!, frame));
-      return lines;
-    }
-
-    const running: string[] = [];
-    const waiting: string[] = [];
-    const done: string[] = [];
-    for (const k of keyOrder) {
-      const s = stateMap.get(k)!;
-      if (s.completed >= s.total) done.push(k);
-      else if (s.started) running.push(k);
-      else waiting.push(k);
-    }
-    // 选出要显示的 key(运行中 > 等待 > 已完成),但按原始顺序渲染,避免行来回跳动
-    const shown = new Set([...running, ...waiting, ...done].slice(0, budget - 1));
-    for (const k of keyOrder) {
-      if (shown.has(k)) lines.push(renderRow(stateMap.get(k)!, frame));
-    }
-    lines.push(
-      `\x1B[2K  ${t("live.more", {
-        hidden: keyOrder.length - shown.size,
-        running: running.filter((k) => !shown.has(k)).length,
-        waiting: waiting.filter((k) => !shown.has(k)).length,
-        done: done.filter((k) => !shown.has(k)).length,
-      })}`,
-    );
-    return lines;
-  }
-
-  function draw(frame: number) {
-    if (!process.stderr.isTTY) return;
-
-    const lines = frameLines(frame);
-    let out = drawnLines > 0 ? `\x1B[${drawnLines}A` : "";
-    out += lines.join("\n") + "\n";
-    // 本帧比上帧短(行完成后折叠、终端拉高)时,清掉下方残留的旧行
-    const extra = drawnLines - lines.length;
-    if (extra > 0) {
-      out += "\x1B[2K\n".repeat(extra) + `\x1B[${extra}A`;
-    }
-    process.stderr.write(out);
-    drawnLines = lines.length;
-  }
-
-  function clearDisplay() {
-    if (!process.stderr.isTTY || drawnLines === 0) return;
-    // 回到起点,逐行清空
-    process.stderr.write(`\x1B[${drawnLines}A`);
-    for (let i = 0; i < drawnLines; i++) {
-      process.stderr.write("\x1B[2K\n");
-    }
-    process.stderr.write(`\x1B[${drawnLines}A`);
-    drawnLines = 0;
-  }
-
-  return {
-    progress(evalId, who, msg) {
-      if (isNoise(msg)) return;
-      const state = stateMap.get(`${evalId}|${who}`);
-      if (state) state.lastMsg = msg;
-      // 不在这里 draw();由 interval 驱动,避免每条日志都刷屏
-    },
-
-    onEvent(event: ReporterEvent) {
-      if (event.type !== "eval:start") return;
-      const who = runWho({ agentName: event.agent.name, model: event.model, experimentId: event.experimentId });
-      const state = stateMap.get(`${event.eval.id}|${who}`);
-      if (state) state.started = true;
-    },
-
-    onRunStart(_evals, _agent, s) {
-      shape = s;
-      // sandbox teardown 失败、budget 不可执行这类独立诊断行可能绕开 progress()/onEvalComplete
-      // 直接落地(见 tty-line.ts)。订阅后先把已画的表格清掉、drawnLines 归零,下一帧从新起点
-      // 重画,不然回跳量和实际光标错位,越滚越多。
-      unsubscribeExternalWrite = onBeforeExternalTerminalWrite(() => clearDisplay());
-      // 初始渲染:让用户看到行表
-      draw(0);
-      intervalId = setInterval(() => {
-        spinFrame = (spinFrame + 1) % SPINNER.length;
-        draw(spinFrame);
-      }, 80);
-    },
-
-    onEvalComplete(result) {
-      const who = runWho({ agentName: result.agent, model: result.model, experimentId: result.experimentId });
-      const state = stateMap.get(`${result.id}|${who}`);
-      if (state) {
-        state.completed += 1;
-        totalCompleted += 1;
-        const prev = state.dominantVerdict;
-        if (!prev || (prev === "passed" && result.verdict !== "passed")) {
-          state.dominantVerdict = result.verdict;
-        }
-      } else {
-        totalCompleted += 1;
-      }
-    },
-
-    onRunComplete(summary) {
-      if (intervalId) {
-        clearInterval(intervalId);
-        intervalId = undefined;
-      }
-      if (unsubscribeExternalWrite) {
-        unsubscribeExternalWrite();
-        unsubscribeExternalWrite = undefined;
-      }
-      // 最后刷一帧(所有行都 done)
-      draw(spinFrame);
-      clearDisplay();
-
-      process.stdout.write(renderRunReport(summary));
-    },
-  };
-}

package/src/runner/reporters/quiet.test.ts

@@ -1,66 +0,0 @@
-import { describe, expect, it } from "vitest";
-import { quietLine } from "./quiet.ts";
-import type { EvalResult } from "../../types.ts";
-
-function baseResult(overrides: Partial<EvalResult> = {}): EvalResult {
-  return {
-    id: "algebra/quadratic",
-    agent: "codex",
-    verdict: "passed",
-    attempt: 0,
-    durationMs: 42_000,
-    assertions: [],
-    ...overrides,
-  };
-}
-
-describe("quietLine", () => {
-  it("passed / skipped 静默(返回 undefined)", () => {
-    expect(quietLine(baseResult())).toBeUndefined();
-    expect(quietLine(baseResult({ verdict: "skipped", skipReason: "no api key" }))).toBeUndefined();
-  });
-
-  it("errored 带 eval id、[who] 与截断后的 error", () => {
-    const line = quietLine(
-      baseResult({
-        verdict: "errored",
-        model: "gpt-5",
-        error: "sandbox create failed: e2b timeout after 3.9s",
-      }),
-    );
-    expect(line).toContain("algebra/quadratic");
-    expect(line).toContain("[codex/gpt-5]");
-    expect(line).toContain("errored");
-    expect(line).toContain("e2b timeout after 3.9s");
-  });
-
-  it("[who] 与进度行同源:有 experimentId 时用其 basename", () => {
-    const line = quietLine(
-      baseResult({ verdict: "errored", experimentId: "compare/xxx--agents-md", error: "boom" }),
-    );
-    expect(line).toContain("[xxx--agents-md]");
-  });
-
-  it("failed 无 error 时取首个失败断言(severity、阈值、detail)", () => {
-    const line = quietLine(
-      baseResult({
-        verdict: "failed",
-        assertions: [
-          { name: "compiles", severity: "gate", score: 1, passed: true },
-          { name: "closedQA", severity: "gate", score: 0.2, passed: false, threshold: 0.7, detail: "wrong city" },
-        ],
-      }),
-    );
-    expect(line).toContain("failed");
-    expect(line).toContain("closedQA");
-    expect(line).toContain("0.20");
-    expect(line).toContain("wrong city");
-    expect(line).not.toContain("compiles");
-  });
-
-  it("超长 error 截断到 200 字符并加省略号", () => {
-    const line = quietLine(baseResult({ verdict: "errored", error: "x".repeat(500) }))!;
-    expect(line).toContain("…");
-    expect(line.length).toBeLessThan(300);
-  });
-});

package/src/runner/reporters/quiet.ts

@@ -1,49 +0,0 @@
-// Quiet 报告器:--quiet 下的最小结果流。进度流照旧(attempt 无 onProgress 时直写 stderr),
-// 这里只补「坏结果」:verdict 为 errored / failed 的结果各写一行 stderr,passed / skipped
-// 静默 —— 保证 --quiet 下起沙箱失败这类执行错不会全程无声、只能事后读 summary.json。
-// 结果流仍走统一的 Reporter 管线,不在 attempt 里散写。
-
-import type { EvalResult, Reporter } from "../../types.ts";
-import { t } from "../../i18n/index.ts";
-import { runWho } from "../types.ts";
-import { verdictSymbol } from "./shared.ts";
-
-/** error / 断言 detail 的截断上限;比 Console 的 400 更紧,--quiet 只要能定位问题。 */
-const DETAIL_MAX = 200;
-
-/** 纯函数:一条结果 → 该写 stderr 的行;passed / skipped 返回 undefined(静默)。 */
-export function quietLine(result: EvalResult): string | undefined {
-  if (result.verdict !== "errored" && result.verdict !== "failed") return undefined;
-  // [who] 与 attempt 进度行同源(runWho),两条流才能对上同一个运行配置。
-  const who = runWho({ agentName: result.agent, model: result.model, experimentId: result.experimentId });
-  const verdict = result.verdict === "errored" ? t("report.errored") : t("report.failed");
-  const detail =
-    result.error !== undefined
-      ? `${t("report.error")}: ${truncate(result.error, DETAIL_MAX)}`
-      : firstFailedAssertion(result);
-  return `  ${verdictSymbol(result.verdict)} ${result.id} ${verdict}  [${who}]${detail ? `  ${detail}` : ""}\n`;
-}
-
-function firstFailedAssertion(result: EvalResult): string | undefined {
-  const a = result.assertions.find((x) => !x.passed);
-  if (!a) return undefined;
-  const sev = a.severity === "gate" ? t("report.gate") : t("report.soft");
-  const thr =
-    a.threshold !== undefined
-      ? t("report.assertionThreshold", { score: a.score.toFixed(2), threshold: a.threshold })
-      : "";
-  return `${sev}: ${a.name}${thr}${a.detail ? ` — ${truncate(a.detail, DETAIL_MAX)}` : ""}`;
-}
-
-function truncate(s: string, n: number): string {
-  return s.length > n ? s.slice(0, n) + "…" : s;
-}
-
-export function Quiet(): Reporter {
-  return {
-    onEvalComplete(result: EvalResult) {
-      const line = quietLine(result);
-      if (line !== undefined) process.stderr.write(line);
-    },
-  };
-}