niceeval 0.6.0 → 0.6.2

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);
-    },
-  };
-}

package/src/runner/reporters/table.ts

@@ -1,277 +0,0 @@
-import type { EvalResult, RunSummary, Usage } from "../../types.ts";
-import { t } from "../../i18n/index.ts";
-import { verdictSymbol } from "./shared.ts";
-import { foldEvalVerdict, evalLevelStats as sharedEvalLevelStats } from "../../shared/verdict.ts";
-// 展示口径(耗时/成本/百分比)与聚合小工具(标签/求和/排序)与 view 共用一份实现;
-// console.ts 经这里 re-export formatDuration。
-import { formatCost, formatDuration, formatPercent } from "../../shared/format.ts";
-import { VERDICT_ORDER, avg, displayExperimentName, fallbackExperimentLabel, sumMaybe, totalTokens } from "../../shared/aggregate.ts";
-
-export { formatCost, formatDuration } from "../../shared/format.ts";
-
-interface ExperimentRow {
-  key: string;
-  label: string;
-  model: string;
-  agent: string;
-  avgDurationMs: number;
-  passRate: number;
-  tokens: number;
-  cost?: number;
-  evals: number;
-  passed: number;
-  failed: number;
-  errored: number;
-  skipped: number;
-  attempts: number;
-  results: EvalResult[];
-}
-
-interface EvalRow {
-  id: string;
-  verdict: EvalResult["verdict"];
-  reason: string;
-  durationMs: number;
-  tokens: number;
-  cost?: number;
-  passedAttempts: number;
-  attempts: number;
-}
-
-export function renderRunReport(summary: RunSummary): string {
-  const rows = aggregateExperimentRows(summary.results);
-  const lines: string[] = ["", t("report.table.experimentsTitle")];
-  lines.push(renderTable(
-    [
-      t("report.table.experiment"),
-      t("report.table.model"),
-      t("report.table.agent"),
-      t("report.table.avgDuration"),
-      t("report.table.successRate"),
-      t("report.table.tokens"),
-      t("report.table.cost"),
-      t("report.table.result"),
-    ],
-    rows.map((row) => [
-      row.label,
-      row.model,
-      row.agent,
-      formatDuration(row.avgDurationMs),
-      formatPercent(row.passRate),
-      formatTokens(row.tokens),
-      formatCost(row.cost),
-      formatResult(row),
-    ]),
-  ));
-
-  for (const row of rows) {
-    lines.push("", `${t("report.table.evalTitle")} ${row.label}`);
-    lines.push(renderTable(
-      [
-        t("report.table.status"),
-        t("report.table.eval"),
-        t("report.table.reason"),
-        t("report.table.duration"),
-        t("report.table.tokens"),
-        t("report.table.cost"),
-        t("report.table.runs"),
-      ],
-      aggregateEvalRows(row.results).map((evalRow) => [
-        formatVerdict(evalRow.verdict),
-        evalRow.id,
-        evalRow.reason,
-        formatDuration(evalRow.durationMs),
-        formatTokens(evalRow.tokens),
-        formatCost(evalRow.cost),
-        `${evalRow.passedAttempts}/${evalRow.attempts}`,
-      ]),
-      { maxWidth: 120, flexibleColumn: 2 },
-    ));
-  }
-
-  const tok = (summary.usage?.inputTokens ?? 0) + (summary.usage?.outputTokens ?? 0);
-  const tokStr = tok > 0 ? `${formatTokens(tok)} tok` : "— tok";
-  const cost = summary.estimatedCostUSD !== undefined ? ` · ${formatCost(summary.estimatedCostUSD)}` : "";
-  const parts = [
-    t("report.summary.passed", { count: summary.passed }),
-    t("report.summary.failed", { count: summary.failed }),
-    ...(summary.errored > 0 ? [t("report.summary.errored", { count: summary.errored })] : []),
-    t("report.summary.skipped", { count: summary.skipped }),
-  ];
-  lines.push(t("report.result", {
-    parts: parts.join(", "),
-    duration: formatDuration(summary.durationMs),
-    tokens: tokStr,
-    cost,
-  }).trimEnd());
-  lines.push(t("report.viewHint").trimEnd());
-
-  return `${lines.join("\n")}\n\n`;
-}
-
-export function formatTokens(n: number): string {
-  if (n >= 1000) return `${(n / 1000).toFixed(1)}k`;
-  return String(n);
-}
-
-export function formatVerdict(verdict: string): string {
-  const sym = verdictSymbol(verdict);
-  switch (verdict) {
-    case "passed": return `${sym} ${t("report.passed")}`;
-    case "failed": return `${sym} ${t("report.failed")}`;
-    case "errored": return `${sym} ${t("report.errored")}`;
-    case "skipped": return `${sym} ${t("report.skipped")}`;
-    default: return `${sym} ${verdict}`;
-  }
-}
-
-function aggregateExperimentRows(results: EvalResult[]): ExperimentRow[] {
-  const groups = new Map<string, EvalResult[]>();
-  for (const result of results) {
-    const key = result.experimentId ? `exp|||${result.experimentId}` : `legacy|||${result.agent}|||${result.model ?? ""}`;
-    groups.set(key, [...(groups.get(key) ?? []), result]);
-  }
-
-  return [...groups.entries()]
-    .map(([key, groupResults]) => {
-      const first = groupResults[0]!;
-      const stats = evalLevelStats(groupResults);
-      return {
-        key,
-        label: displayExperimentName(first.experimentId) ?? fallbackExperimentLabel(first),
-        model: first.model ?? t("report.table.default"),
-        agent: first.agent,
-        avgDurationMs: avg(groupResults.map((r) => r.durationMs)),
-        passRate: stats.passRate,
-        tokens: totalTokens(groupResults.map((r) => r.usage)),
-        cost: sumMaybe(groupResults.map((r) => r.estimatedCostUSD)),
-        evals: stats.evals,
-        passed: stats.passed,
-        failed: stats.failed,
-        errored: stats.errored,
-        skipped: stats.skipped,
-        attempts: groupResults.length,
-        results: groupResults.slice().sort((a, b) => a.id.localeCompare(b.id) || a.attempt - b.attempt),
-      };
-    })
-    .sort((a, b) => b.passRate - a.passRate || a.label.localeCompare(b.label));
-}
-
-function aggregateEvalRows(results: EvalResult[]): EvalRow[] {
-  const byEval = new Map<string, EvalResult[]>();
-  for (const result of results) byEval.set(result.id, [...(byEval.get(result.id) ?? []), result]);
-  return [...byEval.entries()]
-    .map(([id, evalResults]) => {
-      const verdict = foldEvalVerdict(evalResults);
-      const representative =
-        evalResults.find((r) => r.verdict === verdict) ??
-        evalResults.find((r) => r.verdict !== "passed") ??
-        evalResults[0]!;
-      return {
-        id,
-        verdict,
-        reason: verdict === "passed" ? t("report.passed") : reasonFor(representative),
-        durationMs: avg(evalResults.map((r) => r.durationMs)),
-        tokens: totalTokens(evalResults.map((r) => r.usage)),
-        cost: sumMaybe(evalResults.map((r) => r.estimatedCostUSD)),
-        passedAttempts: evalResults.filter((r) => r.verdict === "passed").length,
-        attempts: evalResults.length,
-      };
-    })
-    .sort((a, b) => VERDICT_ORDER[a.verdict] - VERDICT_ORDER[b.verdict] || a.id.localeCompare(b.id));
-}
-
-// 折叠 / 计票口径与 view 同一份实现(src/shared/verdict.ts),CLI 表格和网页榜单永不打架。
-function evalLevelStats(results: EvalResult[]) {
-  return sharedEvalLevelStats(results, (r) => r.id);
-}
-
-function reasonFor(result: EvalResult): string {
-  if (result.skipReason) return result.skipReason;
-  if (result.error) return truncateOneLine(result.error, 120);
-  const failed = result.assertions.filter((a) => !a.passed && a.severity === "gate");
-  const assertions = failed.length ? failed : result.assertions.slice(0, 2);
-  if (!assertions.length) return "";
-  return truncateOneLine(
-    assertions
-      .map((a) => {
-        const threshold = a.threshold !== undefined ? ` ${t("report.assertionThreshold", { score: a.score.toFixed(2), threshold: a.threshold }).trim()}` : "";
-        return `${a.severity}: ${a.name}${threshold}`;
-      })
-      .join(" · "),
-    120,
-  );
-}
-
-function formatResult(row: ExperimentRow): string {
-  const failures = row.failed + row.errored;
-  const suffix = row.skipped > 0 ? ` · ${row.skipped} ${t("report.skipped")}` : "";
-  return `${row.passed}/${row.evals} ${t("report.passed")} · ${failures} ${t("report.failed")}${suffix}`;
-}
-
-
-
-
-
-
-
-function renderTable(headers: string[], rows: string[][], opts: { maxWidth?: number; flexibleColumn?: number } = {}): string {
-  const rawRows = [headers, ...rows];
-  const widths = headers.map((_, col) => Math.max(...rawRows.map((row) => displayWidth(row[col] ?? ""))));
-  const maxWidth = opts.maxWidth ?? 140;
-  const tableWidth = widths.reduce((sum, w) => sum + w, 0) + (headers.length - 1) * 3;
-  if (tableWidth > maxWidth && opts.flexibleColumn !== undefined) {
-    const over = tableWidth - maxWidth;
-    widths[opts.flexibleColumn] = Math.max(18, widths[opts.flexibleColumn]! - over);
-  }
-
-  const lines = [formatRow(headers, widths), widths.map((w) => "-".repeat(w)).join("-+-")];
-  for (const row of rows) lines.push(formatRow(row, widths));
-  return lines.join("\n");
-}
-
-function formatRow(row: string[], widths: number[]): string {
-  return row.map((cell, i) => padEnd(truncateToWidth(cell ?? "", widths[i]!), widths[i]!)).join(" | ");
-}
-
-function truncateOneLine(s: string, width: number): string {
-  return truncateToWidth(s.replace(/\s+/g, " ").trim(), width);
-}
-
-function truncateToWidth(s: string, width: number): string {
-  if (displayWidth(s) <= width) return s;
-  const ellipsis = "…";
-  let out = "";
-  for (const char of s) {
-    if (displayWidth(out + char + ellipsis) > width) break;
-    out += char;
-  }
-  return out + ellipsis;
-}
-
-function padEnd(s: string, width: number): string {
-  return s + " ".repeat(Math.max(0, width - displayWidth(s)));
-}
-
-function displayWidth(s: string): number {
-  let width = 0;
-  for (const char of s) {
-    const code = char.codePointAt(0) ?? 0;
-    width += isWide(code) ? 2 : 1;
-  }
-  return width;
-}
-
-function isWide(code: number): boolean {
-  return (
-    (code >= 0x1100 && code <= 0x115f) ||
-    (code >= 0x2e80 && code <= 0xa4cf) ||
-    (code >= 0xac00 && code <= 0xd7a3) ||
-    (code >= 0xf900 && code <= 0xfaff) ||
-    (code >= 0xfe10 && code <= 0xfe19) ||
-    (code >= 0xfe30 && code <= 0xfe6f) ||
-    (code >= 0xff00 && code <= 0xff60) ||
-    (code >= 0xffe0 && code <= 0xffe6) ||
-    (code >= 0x1f300 && code <= 0x1faff)
-  );
-}