oh-my-knowledge 0.26.0 → 0.27.0

package/dist/src/cli/commands/eval-gold.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"eval-gold.js","sourceRoot":"","sources":["../../../../src/cli/commands/eval-gold.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,OAAO,EAAE,MAAM,gBAAgB,CAAC;AACzC,OAAO,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAC1C,OAAO,EAAE,cAAc,EAAE,mBAAmB,EAAE,MAAM,wBAAwB,CAAC;AAC7E,OAAO,EAAE,qBAAqB,EAAE,MAAM,oBAAoB,CAAC;AAE3D,OAAO,EAAE,uBAAuB,EAAE,MAAM,cAAc,CAAC;AAEvD,MAAM,CAAC,KAAK,UAAU,OAAO,CAAC,IAAc;IAC1C,MAAM,IAAI,GAAG,YAAY,CAAC,IAAI,CAAC,CAAC;IAChC,MAAM,CAAC,GAAG,EAAE,GAAG,IAAI,CAAC,GAAG,IAAI,CAAC;IAC5B,IAAI,CAAC,GAAG,EAAE,CAAC;QACT,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;QACzB,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;IACvB,CAAC;IAED,IAAI,GAAG,KAAK,MAAM,EAAE,CAAC;QACnB,MAAM,EAAE,MAAM,EAAE,GAAG,qBAAqB,CAAC;YACvC,IAAI,EAAE,IAAI;YACV,OAAO,EAAE;gBACP,GAAG,cAAc;gBACjB,GAAG,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,gBAAgB,EAAE;gBAClD,SAAS,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;aAC9B;SACF,CAAC,CAAC;QACH,MAAM,EAAE,eAAe,EAAE,GAAG,MAAM,MAAM,CAAC,2BAA2B,CAAC,CAAC;QACtE,IAAI,CAAC;YACH,MAAM,OAAO,GAAG,eAAe,CAAC,MAAM,CAAC,GAAa,EAAE;gBACpD,SAAS,EAAE,MAAM,CAAC,SAA+B;aAClD,CAAC,CAAC;YACH,OAAO,CAAC,GAAG,CAAC,IAAI,KAAK,IAAI;gBACvB,CAAC,CAAC,MAAM,MAAM,CAAC,GAAa,OAAO,OAAO,CAAC,MAAM,OAAO;gBACxD,CAAC,CAAC,WAAW,OAAO,CAAC,MAAM,aAAa,MAAM,CAAC,GAAa,GAAG,CAAC,CAAC;YACnE,KAAK,MAAM,CAAC,IAAI,OAAO;gBAAE,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;YAC/C,OAAO,CAAC,GAAG,CAAC,IAAI,KAAK,IAAI;gBACvB,CAAC,CAAC,8DAA8D;gBAChE,CAAC,CAAC,2FAA2F,CAAC,CAAC;QACnG,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,OAAO,CAAC,KAAK,CAAE,GAAa,CAAC,OAAO,CAAC,CAAC;YACtC,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;QACvB,CAAC;QACD,OAAO;IACT,CAAC;IAED,IAAI,GAAG,KAAK,UAAU,EAAE,CAAC;QACvB,MAAM,EAAE,WAAW,EAAE,GAAG,qBAAqB,CAAC;YAC5C,IAAI,EAAE,IAAI;YACV,gBAAgB,EAAE,IAAI;YACtB,OAAO,EAAE,EAAE,GAAG,cAAc,EAAE;SAC/B,CAAC,CAAC;QACH,MAAM,GAAG,GAAG,WAAW,CAAC,CAAC,CAAC,CAAC;QAC3B,IAAI,CAAC,GAAG,EAAE,CAAC;YACT,OAAO,CAAC,KAAK,CAAC,qCAAqC,CAAC,CAAC;YACrD,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;QACvB,CAAC;QACD,MAAM,EAAE,mBAAmB,EAAE,GAAG,MAAM,MAAM,CAAC,2BAA2B,CAAC,CAAC;QAC1E,MAAM,MAAM,GAAG,mBAAmB,CAAC,GAAG,CAAC,CAAC;QACxC,IAAI,MAAM,CAAC,EAAE,EAAE,CAAC;YACd,OAAO,CAAC,GAAG,CAAC,IAAI,KAAK,IAAI;gBACvB,CAAC,CAAC,uBAAuB,MAAM,CAAC,WAAW,MAAM;gBACjD,CAAC,CAAC,uBAAuB,MAAM,CAAC,WAAW,cAAc,CAAC,CAAC;YAC7D,OAAO;QACT,CAAC;QACD,OAAO,CAAC,KAAK,CAAC,sBAAsB,MAAM,CAAC,MAAM,CAAC,MAAM,YAAY,CAAC,CAAC;QACtE,KAAK,MAAM,GAAG,IAAI,MAAM,CAAC,MAAM;YAAE,OAAO,CAAC,KAAK,CAAC,OAAO,GAAG,EAAE,CAAC,CAAC;QAC7D,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;IACvB,CAAC;IAED,IAAI,GAAG,KAAK,SAAS,EAAE,CAAC;QACtB,MAAM,QAAQ,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC;QACzB,IAAI,CAAC,QAAQ,EAAE,CAAC;YACd,OAAO,CAAC,KAAK,CAAC,0DAA0D,CAAC,CAAC;YAC1E,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;QACvB,CAAC;QACD,MAAM,EAAE,MAAM,EAAE,GAAG,qBAAqB,CAAC;YACvC,IAAI,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC;YACnB,OAAO,EAAE;gBACP,GAAG,cAAc;gBACjB,UAAU,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;gBAC9B,OAAO,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;gBAC3B,aAAa,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,mBAAmB,EAAE;gBAC/D,mBAAmB,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE;gBACxD,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;aACzB;SACF,CAAC,CAAC;QACH,MAAM,OAAO,GAAG,MAAM,CAAC,UAAU,CAAuB,CAAC;QACzD,IAAI,CAAC,OAAO,EAAE,CAAC;YACb,OAAO,CAAC,KAAK,CAAC,wBAAwB,CAAC,CAAC;YACxC,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;QACvB,CAAC;QACD,MAAM,EAAE,eAAe,EAAE,GAAG,MAAM,MAAM,CAAC,+BAA+B,CAAC,CAAC;QAC1E,MAAM,EAAE,mBAAmB,EAAE,iBAAiB,EAAE,GAAG,MAAM,MAAM,CAAC,2BAA2B,CAAC,CAAC;QAC7F,MAAM,EAAE,eAAe,EAAE,GAAG,MAAM,MAAM,CAAC,8BAA8B,CAAC,CAAC;QAEzE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,GAAG,eAAe,CAAC,OAAO,CAAC,CAAC;QACrD,IAAI,CAAC,OAAO,EAAE,CAAC;YACb,OAAO,CAAC,KAAK,CAAC,2BAA2B,CAAC,CAAC;YAC3C,KAAK,MAAM,CAAC,IAAI,MAAM;gBAAE,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC;YAC1D,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;QACvB,CAAC;QACD,KAAK,MAAM,CAAC,IAAI,MAAM;YAAE,OAAO,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC;QAE5D,MAAM,KAAK,GAAgB,eAAe,CAAC,OAAO,CAAC,MAAM,CAAC,aAAa,CAAW,CAAC,CAAC,CAAC;QACrF,MAAM,MAAM,GAAG,uBAAuB,CAAC,MAAM,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,QAAQ,EAAE,IAAI,CAAC,CAAC;QAClF,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,MAAM,CAAC,MAAM,CAAC,mBAAmB,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC;QAC3E,MAAM,OAAO,GAAG,MAAM,CAAC,IAAI,IAAI,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;QACtE,MAAM,MAAM,GAAG,mBAAmB,CAAC;YACjC,MAAM;YACN,IAAI,EAAE,OAAO;YACb,OAAO,EAAE,MAAM,CAAC,OAA6B;YAC7C,OAAO;YACP,IAAI,EAAE,MAAM,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS;SACrD,CAAC,CAAC;QACH,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC;QAChD,OAAO;IACT,CAAC;IAED,OAAO,CAAC,KAAK,CAAC,iCAAiC,GAAG,kCAAkC,CAAC,CAAC;IACtF,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;AACvB,CAAC;AAED,SAAS,KAAK,CAAC,IAAiB;IAC9B,OAAO,IAAI,KAAK,IAAI;QAClB,CAAC,CAAC;YACE,mCAAmC;YACnC,EAAE;YACF,KAAK;YACL,yDAAyD;YACzD,gCAAgC;YAChC,qDAAqD;SACtD,CAAC,IAAI,CAAC,IAAI,CAAC;QACd,CAAC,CAAC;YACE,uDAAuD;YACvD,EAAE;YACF,QAAQ;YACR,yDAAyD;YACzD,gCAAgC;YAChC,qDAAqD;SACtD,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACnB,CAAC"}

package/dist/src/cli/commands/eval-runner.d.ts

@@ -0,0 +1,2 @@
+export declare function execute(argv: string[]): Promise<void>;
+//# sourceMappingURL=eval-runner.d.ts.map

package/dist/src/cli/commands/eval-runner.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"eval-runner.d.ts","sourceRoot":"","sources":["../../../../src/cli/commands/eval-runner.ts"],"names":[],"mappings":"AA4KA,wBAAsB,OAAO,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC,CA2L3D"}

package/dist/src/cli/commands/eval-runner.js

@@ -0,0 +1,299 @@
+import { CliExit } from '../cli-exit.js';
+import { tCli, langFromArgv } from '../i18n.js';
+import { parseRunConfig } from '../parse-run-config.js';
+import { makeOnProgress } from '../progress.js';
+import { computeRunTally } from '../run-tally.js';
+function isDryRunReport(report) {
+    return Boolean(report && typeof report === 'object' && report.dryRun === true);
+}
+function isDryRunBatchReport(report) {
+    return isDryRunReport(report) && report.batch === true;
+}
+function verdictOptions(values) {
+    const rawThreshold = values.threshold;
+    const gateThreshold = rawThreshold !== undefined && Number.isFinite(Number(rawThreshold))
+        ? Number(rawThreshold)
+        : 3.5;
+    const rawTrivial = values['trivial-diff'];
+    const triviallySmallDiff = rawTrivial !== undefined && Number.isFinite(Number(rawTrivial))
+        ? Number(rawTrivial)
+        : undefined;
+    return { gateThreshold, triviallySmallDiff };
+}
+function verdictPasses(level, headline) {
+    return level === 'PROGRESS' || (level === 'SOLO' && headline.includes('PASS'));
+}
+function reportOnlyMode(values) {
+    return values['report-only'] === true || values['no-gate'] === true;
+}
+function applyGateExitCode(code, values, lang) {
+    if (!reportOnlyMode(values))
+        return code;
+    process.stderr.write(tCli('cli.run.report_only_gate_skipped', lang));
+    return 0;
+}
+async function emitEvaluationVerdict(report, values) {
+    const { computeVerdict, formatVerdictText } = await import('../../eval-core/verdict.js');
+    const result = computeVerdict(report, verdictOptions(values));
+    console.log(formatVerdictText(result, { verbose: true }));
+    return verdictPasses(result.level, result.headline) ? 0 : 1;
+}
+function batchItemFallbackReport(batch, item) {
+    return {
+        kind: 'evaluation',
+        id: item.reportId,
+        meta: {
+            ...batch.meta,
+            variants: ['baseline', item.name],
+            sampleCount: item.sampleCount,
+            totalCostUSD: item.totalCostUSD,
+            artifactHashes: item.artifactHash ? { [item.name]: item.artifactHash } : {},
+        },
+        summary: item.summary,
+        results: [],
+        ...(item.variance ? { variance: item.variance } : {}),
+    };
+}
+async function loadBatchChildReports(batch, reportsDir, lang) {
+    const { createFileStore } = await import('../../server/report-store.js');
+    const store = createFileStore(reportsDir);
+    const reports = [];
+    for (const item of batch.items) {
+        const loaded = await store.get(item.reportId);
+        if (loaded?.kind === 'evaluation') {
+            reports.push(loaded);
+        }
+        else {
+            process.stderr.write(tCli('cli.run.batch_child_report_missing', lang, { id: item.reportId }));
+            reports.push(batchItemFallbackReport(batch, item));
+        }
+    }
+    return reports;
+}
+async function emitBatchVerdict(report, reportsDir, values, lang) {
+    const { computeVerdict } = await import('../../eval-core/verdict.js');
+    const childReports = await loadBatchChildReports(report, reportsDir, lang);
+    const results = childReports.map((child) => ({
+        id: child.id,
+        treatment: child.meta.variants[1] ?? child.id,
+        verdict: computeVerdict(child, verdictOptions(values)),
+    }));
+    const passed = results.filter((r) => verdictPasses(r.verdict.level, r.verdict.headline)).length;
+    const failed = results.length - passed;
+    const status = lang === 'zh'
+        ? (failed === 0 ? '通过' : '未通过')
+        : (failed === 0 ? 'PASS' : 'FAIL');
+    console.log(tCli('cli.run.batch_verdict_header', lang, {
+        status,
+        passed,
+        total: results.length,
+    }));
+    for (const result of results) {
+        console.log(`  ${result.verdict.level}: ${result.treatment} — ${result.verdict.headline}`);
+    }
+    return failed === 0 ? 0 : 1;
+}
+async function announceSavedReport({ report, filePath, reportsDir, values, lang, }) {
+    const tally = computeRunTally(report);
+    process.stderr.write(tCli(report.kind === 'batch-evaluation' ? 'cli.run.batch_complete' : 'cli.run.eval_complete', lang));
+    process.stderr.write(tCli('cli.run.tally', lang, tally));
+    process.stderr.write(tCli('cli.run.report_saved', lang, { path: filePath }));
+    if (!values['no-serve'] && process.stdout.isTTY) {
+        const { createReportServer } = await import('../../server/report-server.js');
+        const server = createReportServer({ reportsDir });
+        const serverUrl = await server.start();
+        const reportUrl = `${serverUrl}/reports/${report.id}`;
+        process.stderr.write(tCli('cli.run.report_server_running', lang, { url: serverUrl }));
+        process.stderr.write(tCli('cli.run.report_server_view', lang, { url: reportUrl }));
+        process.stderr.write(tCli('cli.run.report_server_stop', lang));
+        const { platform } = await import('node:os');
+        const openCmd = platform() === 'darwin' ? 'open' : platform() === 'win32' ? 'start' : 'xdg-open';
+        const { execFile: execFileCb } = await import('node:child_process');
+        execFileCb(openCmd, [reportUrl], () => undefined);
+    }
+    else if (!values['no-serve']) {
+        process.stderr.write(tCli('cli.run.no_serve_in_non_tty', lang));
+        process.stderr.write(tCli('cli.run.no_serve_view_hint', lang, { id: report.id, dir: reportsDir }));
+    }
+}
+export async function execute(argv) {
+    const lang = langFromArgv(argv);
+    // 注: 这里**不**给 parseArgs default 值, 否则 values.xxx 永远不为 undefined,
+    // CLI > eval.yaml > hardcoded-default 三级 fallback 区分不开 ("用户没传" vs "用户传了等于 default 值")。
+    // hardcoded default 在下面处理 undefined 时显式给。
+    const { values, config, evalConfig } = parseRunConfig(argv, {
+        blind: { type: 'boolean' },
+        repeat: { type: 'string' },
+        'judge-repeat': { type: 'string' },
+        bootstrap: { type: 'boolean' },
+        'bootstrap-samples': { type: 'string' },
+        'gold-dir': { type: 'string' },
+        'no-debias-length': { type: 'boolean' },
+        'budget-usd': { type: 'string' },
+        'budget-per-sample-usd': { type: 'string' },
+        'budget-per-sample-ms': { type: 'string' },
+        threshold: { type: 'string', default: '3.5' },
+        'trivial-diff': { type: 'string' },
+        'report-only': { type: 'boolean' },
+        'no-gate': { type: 'boolean' },
+    });
+    const { runEvaluation, runMultiple, runBatchEvaluation } = await import('../../eval-workflows/run-evaluation.js');
+    if (values.blind !== undefined) {
+        config.blind = values.blind;
+    }
+    config.onProgress = makeOnProgress(lang);
+    // --repeat: CLI > eval.yaml > 1. 非 ≥1 整数时提示并钳到 1。
+    const repeatRaw = values.repeat;
+    const parsedRepeat = repeatRaw !== undefined ? Number(repeatRaw) : (evalConfig?.repeat ?? 1);
+    if (repeatRaw !== undefined && (!Number.isFinite(parsedRepeat) || parsedRepeat < 1)) {
+        process.stderr.write(tCli('cli.run.invalid_repeat', lang, { value: repeatRaw }));
+    }
+    const repeatCount = Math.max(1, Math.floor(parsedRepeat) || 1);
+    // --judge-repeat: CLI > eval.yaml > 1.
+    const judgeRepeatRaw = values['judge-repeat'];
+    const parsedJudgeRepeat = judgeRepeatRaw !== undefined ? Number(judgeRepeatRaw) : (evalConfig?.judgeRepeat ?? 1);
+    if (judgeRepeatRaw !== undefined && (!Number.isFinite(parsedJudgeRepeat) || parsedJudgeRepeat < 1)) {
+        process.stderr.write(tCli('cli.run.invalid_judge_repeat', lang, { value: judgeRepeatRaw }));
+    }
+    const judgeRepeatCount = Math.max(1, Math.floor(parsedJudgeRepeat) || 1);
+    if (judgeRepeatCount > 1)
+        config.judgeRepeat = judgeRepeatCount;
+    // --budget-usd / --budget-per-sample-usd / --budget-per-sample-ms:
+    //  hard budget caps. CLI flags override config-file values. When the
+    // total-USD cap is exceeded mid-run, remaining tasks are skipped and a
+    // partial report is persisted with meta.budgetExhausted=true.
+    const budgetUSD = values['budget-usd'] != null ? Number(values['budget-usd']) : undefined;
+    const budgetPerSampleUSD = values['budget-per-sample-usd'] != null ? Number(values['budget-per-sample-usd']) : undefined;
+    const budgetPerSampleMs = values['budget-per-sample-ms'] != null ? Number(values['budget-per-sample-ms']) : undefined;
+    if (budgetUSD !== undefined || budgetPerSampleUSD !== undefined || budgetPerSampleMs !== undefined) {
+        config.budget = {
+            ...(budgetUSD !== undefined && Number.isFinite(budgetUSD) && budgetUSD >= 0 ? { totalUSD: budgetUSD } : {}),
+            ...(budgetPerSampleUSD !== undefined && Number.isFinite(budgetPerSampleUSD) && budgetPerSampleUSD >= 0 ? { perSampleUSD: budgetPerSampleUSD } : {}),
+            ...(budgetPerSampleMs !== undefined && Number.isFinite(budgetPerSampleMs) && budgetPerSampleMs >= 0 ? { perSampleMs: budgetPerSampleMs } : {}),
+        };
+    }
+    // --no-debias-length / eval.yaml `lengthDebias: false`: opt out of length-controlled prompt。
+    // Default debias-on (judge prompt v3-cot-length); flip off only to reproduce historical reports。
+    // CLI 显式 --no-debias-length > eval.yaml lengthDebias > 默认 true。
+    const lengthDebiasOff = values['no-debias-length'] === true
+        || (values['no-debias-length'] === undefined && evalConfig?.lengthDebias === false);
+    if (lengthDebiasOff) {
+        config.lengthDebias = false;
+        process.stderr.write(tCli('cli.run.no_debias_length_active', lang));
+    }
+    // --bootstrap / --bootstrap-samples: CLI > eval.yaml > default(on / 1000)。
+    // `omk eval` owns the product default here instead of wrapper-injecting argv
+    // in commands/eval.ts, so config-file opt-out stays possible.
+    const bootstrapEnabled = values.bootstrap === true
+        || (values.bootstrap === undefined && evalConfig?.bootstrap !== false);
+    if (bootstrapEnabled) {
+        config.bootstrap = true;
+        const bsRaw = values['bootstrap-samples'];
+        const parsedBs = bsRaw !== undefined ? Number(bsRaw) : (evalConfig?.bootstrapSamples ?? 1000);
+        if (bsRaw !== undefined && (!Number.isFinite(parsedBs) || parsedBs < 100)) {
+            process.stderr.write(tCli('cli.run.invalid_bootstrap_samples', lang, { value: bsRaw }));
+        }
+        const bsCount = Math.max(100, Math.floor(parsedBs) || 1000);
+        if (bsCount > 10000) {
+            process.stderr.write(tCli('cli.run.bootstrap_samples_too_large', lang, { n: bsCount }));
+        }
+        config.bootstrapSamples = bsCount;
+    }
+    // 注入 lang 让 evaluation pipeline 能渲染 doctor 报告(失败时)。
+    config.lang = lang;
+    if (values['skip-connectivity']) {
+        process.stderr.write(tCli('cli.run.skip_connectivity_warning', lang) + '\n');
+    }
+    try {
+        // --batch mode: evaluate each skill independently
+        if (values.batch) {
+            const { report, filePath } = await runBatchEvaluation({
+                ...config,
+                repeat: repeatCount,
+                onSkillProgress({ phase, skill, current, total }) {
+                    if (phase === 'start') {
+                        process.stderr.write(tCli('cli.run.skill_section', lang, {
+                            i: current ?? '', n: total ?? '', skill: skill ?? '',
+                        }));
+                    }
+                },
+            });
+            console.log(JSON.stringify(report, null, 2));
+            if (isDryRunBatchReport(report)) {
+                console.log(tCli('cli.run.dry_run_no_scores', lang));
+                throw new CliExit(0);
+            }
+            if (filePath) {
+                await announceSavedReport({ report, filePath, reportsDir: config.outputDir, values, lang });
+            }
+            const exitCode = await emitBatchVerdict(report, config.outputDir, values, lang);
+            throw new CliExit(applyGateExitCode(exitCode, values, lang));
+        }
+        let report;
+        let filePath;
+        if (repeatCount > 1 && !config.dryRun) {
+            const result = await runMultiple({
+                ...config,
+                repeat: repeatCount,
+                onRepeatProgress({ run, total }) {
+                    process.stderr.write(tCli('cli.run.run_section', lang, { i: run, n: total }));
+                },
+            });
+            report = result.report;
+            filePath = null;
+        }
+        else {
+            const result = (await runEvaluation({
+                ...config,
+                ...(repeatCount > 1 ? { repeat: repeatCount } : {}),
+            }));
+            if (isDryRunReport(result.report)) {
+                console.log(JSON.stringify(result.report, null, 2));
+                console.log(tCli('cli.run.dry_run_no_scores', lang));
+                throw new CliExit(0);
+            }
+            report = result.report;
+            filePath = result.filePath;
+        }
+        // --gold-dir / eval.yaml goldDir: compute α/κ/Pearson against gold annotations and re-persist.
+        const goldDir = values['gold-dir'] ?? evalConfig?.goldDir;
+        if (goldDir && filePath) {
+            const { attachGoldAgreementToReport, formatGoldCompare } = await import('../../grading/gold-cli.js');
+            const out = attachGoldAgreementToReport({
+                report,
+                goldDir,
+                outputDir: config.outputDir,
+                samples: config.bootstrapSamples,
+            });
+            if (out.result && out.gold) {
+                process.stderr.write(formatGoldCompare(out.result, out.gold));
+                if (out.result.contaminationWarning) {
+                    process.stderr.write(tCli('cli.run.contamination_warning', lang, {
+                        warning: out.result.contaminationWarning,
+                    }));
+                }
+            }
+            else {
+                process.stderr.write(tCli('cli.run.gold_load_failed', lang, { dir: goldDir }));
+                for (const m of out.loadIssues) {
+                    process.stderr.write(tCli('cli.run.gold_load_issue', lang, { message: m }));
+                }
+            }
+        }
+        console.log(JSON.stringify(report, null, 2));
+        if (filePath) {
+            await announceSavedReport({ report, filePath, reportsDir: config.outputDir, values, lang });
+        }
+        const exitCode = await emitEvaluationVerdict(report, values);
+        throw new CliExit(applyGateExitCode(exitCode, values, lang));
+    }
+    catch (err) {
+        // CliExit 是显式 exit 信号(从 requireEvaluationReport 等子调用冒上来),
+        // 保持原 code 透传;只有真正运行时错误才包装成 CliExit(1)。
+        if (err instanceof CliExit)
+            throw err;
+        console.error(tCli('cli.common.error_prefix', lang, { message: err.message }));
+        throw new CliExit(1);
+    }
+}
+//# sourceMappingURL=eval-runner.js.map

package/dist/src/cli/commands/eval-runner.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"eval-runner.js","sourceRoot":"","sources":["../../../../src/cli/commands/eval-runner.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,gBAAgB,CAAC;AACzC,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAChD,OAAO,EAAE,cAAc,EAAE,MAAM,wBAAwB,CAAC;AACxD,OAAO,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AAChD,OAAO,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAoBlD,SAAS,cAAc,CAAC,MAAe;IACrC,OAAO,OAAO,CAAC,MAAM,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAK,MAA+B,CAAC,MAAM,KAAK,IAAI,CAAC,CAAC;AAC3G,CAAC;AAED,SAAS,mBAAmB,CAAC,MAAe;IAC1C,OAAO,cAAc,CAAC,MAAM,CAAC,IAAK,MAA8B,CAAC,KAAK,KAAK,IAAI,CAAC;AAClF,CAAC;AAED,SAAS,cAAc,CAAC,MAAoB;IAC1C,MAAM,YAAY,GAAG,MAAM,CAAC,SAA+B,CAAC;IAC5D,MAAM,aAAa,GAAG,YAAY,KAAK,SAAS,IAAI,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,YAAY,CAAC,CAAC;QACvF,CAAC,CAAC,MAAM,CAAC,YAAY,CAAC;QACtB,CAAC,CAAC,GAAG,CAAC;IACR,MAAM,UAAU,GAAG,MAAM,CAAC,cAAc,CAAuB,CAAC;IAChE,MAAM,kBAAkB,GAAG,UAAU,KAAK,SAAS,IAAI,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC;QACxF,CAAC,CAAC,MAAM,CAAC,UAAU,CAAC;QACpB,CAAC,CAAC,SAAS,CAAC;IACd,OAAO,EAAE,aAAa,EAAE,kBAAkB,EAAE,CAAC;AAC/C,CAAC;AAED,SAAS,aAAa,CAAC,KAAa,EAAE,QAAgB;IACpD,OAAO,KAAK,KAAK,UAAU,IAAI,CAAC,KAAK,KAAK,MAAM,IAAI,QAAQ,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC;AACjF,CAAC;AAED,SAAS,cAAc,CAAC,MAAoB;IAC1C,OAAO,MAAM,CAAC,aAAa,CAAC,KAAK,IAAI,IAAI,MAAM,CAAC,SAAS,CAAC,KAAK,IAAI,CAAC;AACtE,CAAC;AAED,SAAS,iBAAiB,CAAC,IAAY,EAAE,MAAoB,EAAE,IAAa;IAC1E,IAAI,CAAC,cAAc,CAAC,MAAM,CAAC;QAAE,OAAO,IAAI,CAAC;IACzC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,kCAAkC,EAAE,IAAI,CAAC,CAAC,CAAC;IACrE,OAAO,CAAC,CAAC;AACX,CAAC;AAED,KAAK,UAAU,qBAAqB,CAAC,MAAwB,EAAE,MAAoB;IACjF,MAAM,EAAE,cAAc,EAAE,iBAAiB,EAAE,GAAG,MAAM,MAAM,CAAC,4BAA4B,CAAC,CAAC;IACzF,MAAM,MAAM,GAAG,cAAc,CAAC,MAAM,EAAE,cAAc,CAAC,MAAM,CAAC,CAAC,CAAC;IAC9D,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC,MAAM,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC;IAC1D,OAAO,aAAa,CAAC,MAAM,CAAC,KAAK,EAAE,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;AAC9D,CAAC;AAED,SAAS,uBAAuB,CAC9B,KAA4B,EAC5B,IAA4C;IAE5C,OAAO;QACL,IAAI,EAAE,YAAY;QAClB,EAAE,EAAE,IAAI,CAAC,QAAQ;QACjB,IAAI,EAAE;YACJ,GAAG,KAAK,CAAC,IAAI;YACb,QAAQ,EAAE,CAAC,UAAU,EAAE,IAAI,CAAC,IAAI,CAAC;YACjC,WAAW,EAAE,IAAI,CAAC,WAAW;YAC7B,YAAY,EAAE,IAAI,CAAC,YAAY;YAC/B,cAAc,EAAE,IAAI,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,IAAI,CAAC,YAAY,EAAE,CAAC,CAAC,CAAC,EAAE;SAC5E;QACD,OAAO,EAAE,IAAI,CAAC,OAAO;QACrB,OAAO,EAAE,EAAE;QACX,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,IAAI,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAClC,CAAC;AACxB,CAAC;AAED,KAAK,UAAU,qBAAqB,CAClC,KAA4B,EAC5B,UAAkB,EAClB,IAAa;IAEb,MAAM,EAAE,eAAe,EAAE,GAAG,MAAM,MAAM,CAAC,8BAA8B,CAAC,CAAC;IACzE,MAAM,KAAK,GAAG,eAAe,CAAC,UAAU,CAAC,CAAC;IAC1C,MAAM,OAAO,GAAuB,EAAE,CAAC;IACvC,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,KAAK,EAAE,CAAC;QAC/B,MAAM,MAAM,GAAG,MAAM,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QAC9C,IAAI,MAAM,EAAE,IAAI,KAAK,YAAY,EAAE,CAAC;YAClC,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACvB,CAAC;aAAM,CAAC;YACN,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,oCAAoC,EAAE,IAAI,EAAE,EAAE,EAAE,EAAE,IAAI,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC;YAC9F,OAAO,CAAC,IAAI,CAAC,uBAAuB,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC,CAAC;QACrD,CAAC;IACH,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,KAAK,UAAU,gBAAgB,CAC7B,MAA6B,EAC7B,UAAkB,EAClB,MAAoB,EACpB,IAAa;IAEb,MAAM,EAAE,cAAc,EAAE,GAAG,MAAM,MAAM,CAAC,4BAA4B,CAAC,CAAC;IACtE,MAAM,YAAY,GAAG,MAAM,qBAAqB,CAAC,MAAM,EAAE,UAAU,EAAE,IAAI,CAAC,CAAC;IAC3E,MAAM,OAAO,GAAG,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;QAC3C,EAAE,EAAE,KAAK,CAAC,EAAE;QACZ,SAAS,EAAE,KAAK,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,KAAK,CAAC,EAAE;QAC7C,OAAO,EAAE,cAAc,CAAC,KAAK,EAAE,cAAc,CAAC,MAAM,CAAC,CAAC;KACvD,CAAC,CAAC,CAAC;IACJ,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,aAAa,CAAC,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC;IAChG,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,GAAG,MAAM,CAAC;IAEvC,MAAM,MAAM,GAAG,IAAI,KAAK,IAAI;QAC1B,CAAC,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC;QAC/B,CAAC,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;IACrC,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,8BAA8B,EAAE,IAAI,EAAE;QACrD,MAAM;QACN,MAAM;QACN,KAAK,EAAE,OAAO,CAAC,MAAM;KACtB,CAAC,CAAC,CAAC;IACJ,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;QAC7B,OAAO,CAAC,GAAG,CAAC,KAAK,MAAM,CAAC,OAAO,CAAC,KAAK,KAAK,MAAM,CAAC,SAAS,MAAM,MAAM,CAAC,OAAO,CAAC,QAAQ,EAAE,CAAC,CAAC;IAC7F,CAAC;IACD,OAAO,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;AAC9B,CAAC;AAED,KAAK,UAAU,mBAAmB,CAAC,EACjC,MAAM,EACN,QAAQ,EACR,UAAU,EACV,MAAM,EACN,IAAI,GAOL;IACC,MAAM,KAAK,GAAG,eAAe,CAAC,MAAM,CAAC,CAAC;IACtC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,IAAI,KAAK,kBAAkB,CAAC,CAAC,CAAC,wBAAwB,CAAC,CAAC,CAAC,uBAAuB,EAAE,IAAI,CAAC,CAAC,CAAC;IAC1H,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,eAAe,EAAE,IAAI,EAAE,KAAK,CAAC,CAAC,CAAC;IACzD,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,sBAAsB,EAAE,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,CAAC,CAAC,CAAC;IAE7E,IAAI,CAAC,MAAM,CAAC,UAAU,CAAC,IAAI,OAAO,CAAC,MAAM,CAAC,KAAK,EAAE,CAAC;QAChD,MAAM,EAAE,kBAAkB,EAAE,GAAG,MAAM,MAAM,CAAC,+BAA+B,CAAC,CAAC;QAC7E,MAAM,MAAM,GAAiB,kBAAkB,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC;QAChE,MAAM,SAAS,GAAW,MAAM,MAAM,CAAC,KAAK,EAAE,CAAC;QAC/C,MAAM,SAAS,GAAW,GAAG,SAAS,YAAY,MAAM,CAAC,EAAE,EAAE,CAAC;QAC9D,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,+BAA+B,EAAE,IAAI,EAAE,EAAE,GAAG,EAAE,SAAS,EAAE,CAAC,CAAC,CAAC;QACtF,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,4BAA4B,EAAE,IAAI,EAAE,EAAE,GAAG,EAAE,SAAS,EAAE,CAAC,CAAC,CAAC;QACnF,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,4BAA4B,EAAE,IAAI,CAAC,CAAC,CAAC;QAE/D,MAAM,EAAE,QAAQ,EAAE,GAAG,MAAM,MAAM,CAAC,SAAS,CAAC,CAAC;QAC7C,MAAM,OAAO,GAAW,QAAQ,EAAE,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,QAAQ,EAAE,KAAK,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,UAAU,CAAC;QACzG,MAAM,EAAE,QAAQ,EAAE,UAAU,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;QACpE,UAAU,CAAC,OAAO,EAAE,CAAC,SAAS,CAAC,EAAE,GAAG,EAAE,CAAC,SAAS,CAAC,CAAC;IACpD,CAAC;SAAM,IAAI,CAAC,MAAM,CAAC,UAAU,CAAC,EAAE,CAAC;QAC/B,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,6BAA6B,EAAE,IAAI,CAAC,CAAC,CAAC;QAChE,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,4BAA4B,EAAE,IAAI,EAAE,EAAE,EAAE,EAAE,MAAM,CAAC,EAAE,EAAE,GAAG,EAAE,UAAU,EAAE,CAAC,CAAC,CAAC;IACrG,CAAC;AACH,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,OAAO,CAAC,IAAc;IAC1C,MAAM,IAAI,GAAG,YAAY,CAAC,IAAI,CAAC,CAAC;IAChC,iEAAiE;IACjE,uFAAuF;IACvF,0CAA0C;IAC1C,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,UAAU,EAAE,GAAG,cAAc,CAAC,IAAI,EAAE;QAC1D,KAAK,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE;QAC1B,MAAM,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QAC1B,cAAc,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QAClC,SAAS,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE;QAC9B,mBAAmB,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QACvC,UAAU,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QAC9B,kBAAkB,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE;QACvC,YAAY,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QAChC,uBAAuB,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QAC3C,sBAAsB,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QAC1C,SAAS,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,KAAK,EAAE;QAC7C,cAAc,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QAClC,aAAa,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE;QAClC,SAAS,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE;KAC/B,CAAC,CAAC;IAEH,MAAM,EAAE,aAAa,EAAE,WAAW,EAAE,kBAAkB,EAAE,GAAG,MAAM,MAAM,CAAC,wCAAwC,CAAC,CAAC;IAElH,IAAI,MAAM,CAAC,KAAK,KAAK,SAAS,EAAE,CAAC;QAC/B,MAAM,CAAC,KAAK,GAAG,MAAM,CAAC,KAA4B,CAAC;IACrD,CAAC;IACD,MAAM,CAAC,UAAU,GAAG,cAAc,CAAC,IAAI,CAAgC,CAAC;IAExE,kDAAkD;IAClD,MAAM,SAAS,GAAG,MAAM,CAAC,MAA4B,CAAC;IACtD,MAAM,YAAY,GAAG,SAAS,KAAK,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,UAAU,EAAE,MAAM,IAAI,CAAC,CAAC,CAAC;IAC7F,IAAI,SAAS,KAAK,SAAS,IAAI,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,YAAY,CAAC,IAAI,YAAY,GAAG,CAAC,CAAC,EAAE,CAAC;QACpF,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,wBAAwB,EAAE,IAAI,EAAE,EAAE,KAAK,EAAE,SAAS,EAAE,CAAC,CAAC,CAAC;IACnF,CAAC;IACD,MAAM,WAAW,GAAW,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,CAAC,CAAC,CAAC;IAEvE,uCAAuC;IACvC,MAAM,cAAc,GAAG,MAAM,CAAC,cAAc,CAAuB,CAAC;IACpE,MAAM,iBAAiB,GAAG,cAAc,KAAK,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,cAAc,CAAC,CAAC,CAAC,CAAC,CAAC,UAAU,EAAE,WAAW,IAAI,CAAC,CAAC,CAAC;IACjH,IAAI,cAAc,KAAK,SAAS,IAAI,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,iBAAiB,CAAC,IAAI,iBAAiB,GAAG,CAAC,CAAC,EAAE,CAAC;QACnG,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,8BAA8B,EAAE,IAAI,EAAE,EAAE,KAAK,EAAE,cAAc,EAAE,CAAC,CAAC,CAAC;IAC9F,CAAC;IACD,MAAM,gBAAgB,GAAW,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,iBAAiB,CAAC,IAAI,CAAC,CAAC,CAAC;IACjF,IAAI,gBAAgB,GAAG,CAAC;QAAE,MAAM,CAAC,WAAW,GAAG,gBAAgB,CAAC;IAEhE,mEAAmE;IACnE,qEAAqE;IACrE,uEAAuE;IACvE,8DAA8D;IAC9D,MAAM,SAAS,GAAG,MAAM,CAAC,YAAY,CAAC,IAAI,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IAC1F,MAAM,kBAAkB,GAAG,MAAM,CAAC,uBAAuB,CAAC,IAAI,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,uBAAuB,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACzH,MAAM,iBAAiB,GAAG,MAAM,CAAC,sBAAsB,CAAC,IAAI,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,sBAAsB,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACtH,IAAI,SAAS,KAAK,SAAS,IAAI,kBAAkB,KAAK,SAAS,IAAI,iBAAiB,KAAK,SAAS,EAAE,CAAC;QACnG,MAAM,CAAC,MAAM,GAAG;YACd,GAAG,CAAC,SAAS,KAAK,SAAS,IAAI,MAAM,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,SAAS,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YAC3G,GAAG,CAAC,kBAAkB,KAAK,SAAS,IAAI,MAAM,CAAC,QAAQ,CAAC,kBAAkB,CAAC,IAAI,kBAAkB,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,kBAAkB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACnJ,GAAG,CAAC,iBAAiB,KAAK,SAAS,IAAI,MAAM,CAAC,QAAQ,CAAC,iBAAiB,CAAC,IAAI,iBAAiB,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,iBAAiB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SAC/I,CAAC;IACJ,CAAC;IAED,6FAA6F;IAC7F,iGAAiG;IACjG,gEAAgE;IAChE,MAAM,eAAe,GAAI,MAAM,CAAC,kBAAkB,CAAyB,KAAK,IAAI;WAC/E,CAAC,MAAM,CAAC,kBAAkB,CAAC,KAAK,SAAS,IAAI,UAAU,EAAE,YAAY,KAAK,KAAK,CAAC,CAAC;IACtF,IAAI,eAAe,EAAE,CAAC;QACpB,MAAM,CAAC,YAAY,GAAG,KAAK,CAAC;QAC5B,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,iCAAiC,EAAE,IAAI,CAAC,CAAC,CAAC;IACtE,CAAC;IAED,2EAA2E;IAC3E,6EAA6E;IAC7E,8DAA8D;IAC9D,MAAM,gBAAgB,GAAI,MAAM,CAAC,SAAiC,KAAK,IAAI;WACtE,CAAC,MAAM,CAAC,SAAS,KAAK,SAAS,IAAI,UAAU,EAAE,SAAS,KAAK,KAAK,CAAC,CAAC;IACzE,IAAI,gBAAgB,EAAE,CAAC;QACrB,MAAM,CAAC,SAAS,GAAG,IAAI,CAAC;QACxB,MAAM,KAAK,GAAG,MAAM,CAAC,mBAAmB,CAAuB,CAAC;QAChE,MAAM,QAAQ,GAAG,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,UAAU,EAAE,gBAAgB,IAAI,IAAI,CAAC,CAAC;QAC9F,IAAI,KAAK,KAAK,SAAS,IAAI,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,QAAQ,GAAG,GAAG,CAAC,EAAE,CAAC;YAC1E,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,mCAAmC,EAAE,IAAI,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC;QAC1F,CAAC;QACD,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,IAAI,CAAC,CAAC;QAC5D,IAAI,OAAO,GAAG,KAAK,EAAE,CAAC;YACpB,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,qCAAqC,EAAE,IAAI,EAAE,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC;QAC1F,CAAC;QACD,MAAM,CAAC,gBAAgB,GAAG,OAAO,CAAC;IACpC,CAAC;IAED,oDAAoD;IACpD,MAAM,CAAC,IAAI,GAAG,IAAI,CAAC;IACnB,IAAI,MAAM,CAAC,mBAAmB,CAAY,EAAE,CAAC;QAC3C,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,mCAAmC,EAAE,IAAI,CAAC,GAAG,IAAI,CAAC,CAAC;IAC/E,CAAC;IAED,IAAI,CAAC;QACH,kDAAkD;QAClD,IAAI,MAAM,CAAC,KAAK,EAAE,CAAC;YACjB,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,GAAG,MAAM,kBAAkB,CAAC;gBACpD,GAAG,MAAM;gBACT,MAAM,EAAE,WAAW;gBACnB,eAAe,CAAC,EAAE,KAAK,EAAE,KAAK,EAAE,OAAO,EAAE,KAAK,EAAqB;oBACjE,IAAI,KAAK,KAAK,OAAO,EAAE,CAAC;wBACtB,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,uBAAuB,EAAE,IAAI,EAAE;4BACvD,CAAC,EAAE,OAAO,IAAI,EAAE,EAAE,CAAC,EAAE,KAAK,IAAI,EAAE,EAAE,KAAK,EAAE,KAAK,IAAI,EAAE;yBACrD,CAAC,CAAC,CAAC;oBACN,CAAC;gBACH,CAAC;aACF,CAAmF,CAAC;YACrF,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;YAC7C,IAAI,mBAAmB,CAAC,MAAM,CAAC,EAAE,CAAC;gBAChC,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,2BAA2B,EAAE,IAAI,CAAC,CAAC,CAAC;gBACrD,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;YACvB,CAAC;YACD,IAAI,QAAQ,EAAE,CAAC;gBACb,MAAM,mBAAmB,CAAC,EAAE,MAAM,EAAE,QAAQ,EAAE,UAAU,EAAE,MAAM,CAAC,SAAS,EAAE,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC;YAC9F,CAAC;YACD,MAAM,QAAQ,GAAG,MAAM,gBAAgB,CAAC,MAAM,EAAE,MAAM,CAAC,SAAS,EAAE,MAAM,EAAE,IAAI,CAAC,CAAC;YAChF,MAAM,IAAI,OAAO,CAAC,iBAAiB,CAAC,QAAQ,EAAE,MAAM,EAAE,IAAI,CAAC,CAAC,CAAC;QAC/D,CAAC;QAED,IAAI,MAAc,CAAC;QACnB,IAAI,QAAuB,CAAC;QAE5B,IAAI,WAAW,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC;YACtC,MAAM,MAAM,GAAG,MAAM,WAAW,CAAC;gBAC/B,GAAG,MAAM;gBACT,MAAM,EAAE,WAAW;gBACnB,gBAAgB,CAAC,EAAE,GAAG,EAAE,KAAK,EAAsB;oBACjD,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,qBAAqB,EAAE,IAAI,EAAE,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC;gBAChF,CAAC;aACF,CAAuB,CAAC;YACzB,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC;YACvB,QAAQ,GAAG,IAAI,CAAC;QAClB,CAAC;aAAM,CAAC;YACN,MAAM,MAAM,GAAG,CAAC,MAAM,aAAa,CAAC;gBAClC,GAAG,MAAM;gBACT,GAAG,CAAC,WAAW,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aACpD,CAAC,CAAe,CAAC;YAClB,IAAI,cAAc,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC;gBAClC,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;gBACpD,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,2BAA2B,EAAE,IAAI,CAAC,CAAC,CAAC;gBACrD,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;YACvB,CAAC;YACD,MAAM,GAAG,MAAM,CAAC,MAAgB,CAAC;YACjC,QAAQ,GAAG,MAAM,CAAC,QAAQ,CAAC;QAC7B,CAAC;QAED,+FAA+F;QAC/F,MAAM,OAAO,GAAI,MAAM,CAAC,UAAU,CAAwB,IAAI,UAAU,EAAE,OAAO,CAAC;QAClF,IAAI,OAAO,IAAI,QAAQ,EAAE,CAAC;YACxB,MAAM,EAAE,2BAA2B,EAAE,iBAAiB,EAAE,GAAG,MAAM,MAAM,CAAC,2BAA2B,CAAC,CAAC;YACrG,MAAM,GAAG,GAAG,2BAA2B,CAAC;gBACtC,MAAM;gBACN,OAAO;gBACP,SAAS,EAAE,MAAM,CAAC,SAAS;gBAC3B,OAAO,EAAE,MAAM,CAAC,gBAAgB;aACjC,CAAC,CAAC;YACH,IAAI,GAAG,CAAC,MAAM,IAAI,GAAG,CAAC,IAAI,EAAE,CAAC;gBAC3B,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,iBAAiB,CAAC,GAAG,CAAC,MAAM,EAAE,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC;gBAC9D,IAAI,GAAG,CAAC,MAAM,CAAC,oBAAoB,EAAE,CAAC;oBACpC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,+BAA+B,EAAE,IAAI,EAAE;wBAC/D,OAAO,EAAE,GAAG,CAAC,MAAM,CAAC,oBAAoB;qBACzC,CAAC,CAAC,CAAC;gBACN,CAAC;YACH,CAAC;iBAAM,CAAC;gBACN,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,0BAA0B,EAAE,IAAI,EAAE,EAAE,GAAG,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC;gBAC/E,KAAK,MAAM,CAAC,IAAI,GAAG,CAAC,UAAU,EAAE,CAAC;oBAC/B,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,yBAAyB,EAAE,IAAI,EAAE,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC;gBAC9E,CAAC;YACH,CAAC;QACH,CAAC;QAED,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;QAC7C,IAAI,QAAQ,EAAE,CAAC;YACb,MAAM,mBAAmB,CAAC,EAAE,MAAM,EAAE,QAAQ,EAAE,UAAU,EAAE,MAAM,CAAC,SAAS,EAAE,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC;QAC9F,CAAC;QACD,MAAM,QAAQ,GAAG,MAAM,qBAAqB,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;QAC7D,MAAM,IAAI,OAAO,CAAC,iBAAiB,CAAC,QAAQ,EAAE,MAAM,EAAE,IAAI,CAAC,CAAC,CAAC;IAC/D,CAAC;IAAC,OAAO,GAAY,EAAE,CAAC;QACtB,0DAA0D;QAC1D,wCAAwC;QACxC,IAAI,GAAG,YAAY,OAAO;YAAE,MAAM,GAAG,CAAC;QACtC,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,yBAAyB,EAAE,IAAI,EAAE,EAAE,OAAO,EAAG,GAAa,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QAC1F,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;IACvB,CAAC;AACH,CAAC"}

package/dist/src/cli/commands/eval.d.ts

@@ -0,0 +1,2 @@
+export declare function execute(argv: string[]): Promise<void>;
+//# sourceMappingURL=eval.d.ts.map

package/dist/src/cli/commands/eval.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"eval.d.ts","sourceRoot":"","sources":["../../../../src/cli/commands/eval.ts"],"names":[],"mappings":"AAIA,wBAAsB,OAAO,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC,CAY3D"}

package/dist/src/cli/commands/eval.js

@@ -0,0 +1,16 @@
+import { execute as debias } from './eval-debias.js';
+import { execute as gold } from './eval-gold.js';
+import { execute as run } from './eval-runner.js';
+export async function execute(argv) {
+    const [sub, ...rest] = argv;
+    if (sub === 'gold') {
+        await gold(rest);
+        return;
+    }
+    if (sub === 'debias') {
+        await debias(rest);
+        return;
+    }
+    await run(argv);
+}
+//# sourceMappingURL=eval.js.map

package/dist/src/cli/commands/eval.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"eval.js","sourceRoot":"","sources":["../../../../src/cli/commands/eval.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,IAAI,MAAM,EAAE,MAAM,kBAAkB,CAAC;AACrD,OAAO,EAAE,OAAO,IAAI,IAAI,EAAE,MAAM,gBAAgB,CAAC;AACjD,OAAO,EAAE,OAAO,IAAI,GAAG,EAAE,MAAM,kBAAkB,CAAC;AAElD,MAAM,CAAC,KAAK,UAAU,OAAO,CAAC,IAAc;IAC1C,MAAM,CAAC,GAAG,EAAE,GAAG,IAAI,CAAC,GAAG,IAAI,CAAC;IAC5B,IAAI,GAAG,KAAK,MAAM,EAAE,CAAC;QACnB,MAAM,IAAI,CAAC,IAAI,CAAC,CAAC;QACjB,OAAO;IACT,CAAC;IACD,IAAI,GAAG,KAAK,QAAQ,EAAE,CAAC;QACrB,MAAM,MAAM,CAAC,IAAI,CAAC,CAAC;QACnB,OAAO;IACT,CAAC;IAED,MAAM,GAAG,CAAC,IAAI,CAAC,CAAC;AAClB,CAAC"}

package/dist/src/cli/commands/export-diff.d.ts

@@ -0,0 +1,2 @@
+export declare function execute(argv: string[]): Promise<void>;
+//# sourceMappingURL=export-diff.d.ts.map

package/dist/src/cli/commands/export-diff.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"export-diff.d.ts","sourceRoot":"","sources":["../../../../src/cli/commands/export-diff.ts"],"names":[],"mappings":"AAQA,wBAAsB,OAAO,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC,CAuF3D"}

package/dist/src/cli/commands/export-diff.js

@@ -0,0 +1,177 @@
+import { resolve } from 'node:path';
+import { CliExit } from '../cli-exit.js';
+import { langFromArgv } from '../i18n.js';
+import { COMMON_OPTIONS, DEFAULT_REPORTS_DIR } from '../parse-run-config.js';
+import { parseArgsStrictOrExit } from '../parse-strict.js';
+import { requireEvaluationReport } from './_shared.js';
+export async function execute(argv) {
+    const lang = langFromArgv(argv);
+    const { values, positionals } = parseArgsStrictOrExit({
+        args: argv,
+        allowPositionals: true,
+        options: {
+            ...COMMON_OPTIONS,
+            'reports-dir': { type: 'string', default: DEFAULT_REPORTS_DIR },
+            'regressions-only': { type: 'boolean', default: false },
+            threshold: { type: 'string' },
+            variant: { type: 'string' },
+            top: { type: 'string' },
+        },
+    });
+    if (positionals.length === 0) {
+        console.error(usage(lang));
+        throw new CliExit(1);
+    }
+    const { createFileStore } = await import('../../server/report-store.js');
+    const store = createFileStore(resolve(values['reports-dir']));
+    if (positionals.length === 1) {
+        await runSampleLevelDiff(positionals[0], store, values, lang);
+        return;
+    }
+    if (positionals.length > 2) {
+        console.error(usage(lang));
+        throw new CliExit(1);
+    }
+    const [id1, id2] = positionals;
+    const r1 = requireEvaluationReport(await store.get(id1), id1, lang);
+    const r2 = requireEvaluationReport(await store.get(id2), id2, lang);
+    console.log(`\n  Diff: ${id1} → ${id2}\n`);
+    const g1 = r1.meta?.gitInfo;
+    const g2 = r2.meta?.gitInfo;
+    if (g1 || g2) {
+        console.log(`  Git:  ${g1?.commitShort || '?'}${g1?.dirty ? '*' : ''} (${g1?.branch || '?'}) → ${g2?.commitShort || '?'}${g2?.dirty ? '*' : ''} (${g2?.branch || '?'})`);
+    }
+    const { crossReportComparabilityWarnings, formatComparabilityWarnings } = await import('../../eval-core/comparability.js');
+    const comparability = formatComparabilityWarnings(crossReportComparabilityWarnings(r1, r2), lang);
+    if (comparability)
+        process.stderr.write(`\n${comparability}\n\n`);
+    const variants = [...new Set([...(r1.meta?.variants || []), ...(r2.meta?.variants || [])])];
+    for (const variant of variants) {
+        const s1 = r1.summary?.[variant];
+        const s2 = r2.summary?.[variant];
+        if (!s1 && !s2)
+            continue;
+        console.log(`\n  [${variant}]`);
+        const score1 = s1?.avgCompositeScore ?? '-';
+        const score2 = s2?.avgCompositeScore ?? '-';
+        const scoreDelta = typeof score1 === 'number' && typeof score2 === 'number'
+            ? ` (${score2 > score1 ? '+' : ''}${(score2 - score1).toFixed(2)})`
+            : '';
+        console.log(`    Score:   ${score1} → ${score2}${scoreDelta}`);
+        console.log(`    Turns:   ${s1?.avgNumTurns ?? '-'} → ${s2?.avgNumTurns ?? '-'}`);
+        if (s1?.avgToolCalls != null || s2?.avgToolCalls != null) {
+            console.log(`    Tools:   ${s1?.avgToolCalls ?? '-'} → ${s2?.avgToolCalls ?? '-'}`);
+            const sr1 = s1?.toolSuccessRate != null ? `${(s1.toolSuccessRate * 100).toFixed(0)}%` : '-';
+            const sr2 = s2?.toolSuccessRate != null ? `${(s2.toolSuccessRate * 100).toFixed(0)}%` : '-';
+            console.log(`    ToolOK:  ${sr1} → ${sr2}`);
+        }
+        const cost1 = s1?.avgCostPerSample ?? 0;
+        const cost2 = s2?.avgCostPerSample ?? 0;
+        const reported1 = s1?.execCostReported !== false;
+        const reported2 = s2?.execCostReported !== false;
+        const fmtCost = (cost, reported) => reported ? `$${cost.toFixed(4)}` : '—';
+        const costPct = reported1 && reported2 && cost1 > 0
+            ? ` (${cost2 > cost1 ? '+' : ''}${(((cost2 - cost1) / cost1) * 100).toFixed(0)}%)`
+            : '';
+        console.log(`    Cost:    ${fmtCost(cost1, reported1)} → ${fmtCost(cost2, reported2)}${costPct}`);
+        const h1 = r1.meta?.artifactHashes?.[variant];
+        const h2 = r2.meta?.artifactHashes?.[variant];
+        if (h1 && h2 && h1 !== h2) {
+            console.log(`    Skill:   ${h1.slice(0, 8)} → ${h2.slice(0, 8)} (changed)`);
+        }
+    }
+    console.log('');
+}
+async function runSampleLevelDiff(reportId, store, flags, lang) {
+    const report = requireEvaluationReport(await store.get(reportId), reportId, lang);
+    const { reportComparabilityWarnings, formatComparabilityWarnings } = await import('../../eval-core/comparability.js');
+    const comparability = formatComparabilityWarnings(reportComparabilityWarnings(report), lang);
+    if (comparability)
+        process.stderr.write(`\n${comparability}\n\n`);
+    const variants = report.meta?.variants ?? [];
+    if (variants.length < 2) {
+        console.error('Sample-level diff needs at least 2 variants in the report.');
+        throw new CliExit(1);
+    }
+    const control = variants[0];
+    const treatment = flags.variant ?? variants[1];
+    if (!variants.includes(treatment)) {
+        console.error(`Variant "${treatment}" not in report. Available: ${variants.join(', ')}`);
+        throw new CliExit(1);
+    }
+    const threshold = flags.threshold != null ? Number(flags.threshold) : 0;
+    const regressionsOnly = Boolean(flags['regressions-only']);
+    const topN = flags.top != null ? Math.max(1, Number(flags.top) || 0) : undefined;
+    const rows = [];
+    for (const entry of report.results ?? []) {
+        const c = entry.variants?.[control];
+        const t = entry.variants?.[treatment];
+        if (!c || !t)
+            continue;
+        const cComp = c.compositeScore ?? c.llmScore ?? 0;
+        const tComp = t.compositeScore ?? t.llmScore ?? 0;
+        rows.push({
+            id: entry.sample_id,
+            cFact: c.layeredScores?.factScore,
+            tFact: t.layeredScores?.factScore,
+            cBeh: c.layeredScores?.behaviorScore,
+            tBeh: t.layeredScores?.behaviorScore,
+            cJudge: c.layeredScores?.judgeScore,
+            tJudge: t.layeredScores?.judgeScore,
+            cComp,
+            tComp,
+            delta: Number((tComp - cComp).toFixed(3)),
+        });
+    }
+    rows.sort((a, b) => Math.abs(b.delta) - Math.abs(a.delta));
+    let filtered = rows;
+    if (regressionsOnly)
+        filtered = filtered.filter((r) => r.delta < threshold);
+    if (topN !== undefined)
+        filtered = filtered.slice(0, topN);
+    console.log(`\n  Sample-level diff: ${treatment} vs ${control} (report ${reportId})`);
+    if (regressionsOnly)
+        console.log(`  Filter: regressions only (Δ < ${threshold})`);
+    console.log('');
+    console.log('  sample_id           Δ      composite (c→t)   fact (c→t)     behavior (c→t)   judge (c→t)');
+    console.log('  ' + '-'.repeat(100));
+    if (filtered.length === 0) {
+        console.log(regressionsOnly ? '  (no regressions found)' : '  (no shared samples)');
+        console.log('');
+        return;
+    }
+    const fmtLayer = (a, b) => {
+        const av = typeof a === 'number' ? a.toFixed(2) : '—';
+        const bv = typeof b === 'number' ? b.toFixed(2) : '—';
+        return `${av} → ${bv}`.padEnd(15);
+    };
+    for (const row of filtered) {
+        const sign = row.delta > 0 ? '+' : '';
+        console.log(`  ${row.id.slice(0, 18).padEnd(20)}`
+            + `${`${sign}${row.delta.toFixed(2)}`.padEnd(7)}`
+            + `${`${row.cComp.toFixed(2)} → ${row.tComp.toFixed(2)}`.padEnd(17)}`
+            + `${fmtLayer(row.cFact, row.tFact)} ${fmtLayer(row.cBeh, row.tBeh)} ${fmtLayer(row.cJudge, row.tJudge)}`);
+    }
+    console.log('');
+    console.log(`  Showing ${filtered.length} of ${rows.length} samples · sorted by |Δ|`);
+    if (regressionsOnly) {
+        const total = rows.length;
+        const regressions = rows.filter((r) => r.delta < threshold).length;
+        console.log(`  Regression rate: ${regressions}/${total} samples (${total > 0 ? ((regressions / total) * 100).toFixed(0) : 0}%)`);
+    }
+    console.log('');
+}
+function usage(lang) {
+    return lang === 'zh'
+        ? [
+            '用法：',
+            '  omk export diff <report-id> [--variant <name>] [--regressions-only] [--top <n>]',
+            '  omk export diff <report-id-a> <report-id-b>',
+        ].join('\n')
+        : [
+            'Usage:',
+            '  omk export diff <report-id> [--variant <name>] [--regressions-only] [--top <n>]',
+            '  omk export diff <report-id-a> <report-id-b>',
+        ].join('\n');
+}
+//# sourceMappingURL=export-diff.js.map

package/dist/src/cli/commands/export-diff.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"export-diff.js","sourceRoot":"","sources":["../../../../src/cli/commands/export-diff.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,OAAO,EAAE,MAAM,gBAAgB,CAAC;AACzC,OAAO,EAAE,YAAY,EAAgB,MAAM,YAAY,CAAC;AACxD,OAAO,EAAE,cAAc,EAAE,mBAAmB,EAAE,MAAM,wBAAwB,CAAC;AAC7E,OAAO,EAAE,qBAAqB,EAAE,MAAM,oBAAoB,CAAC;AAE3D,OAAO,EAAE,uBAAuB,EAAE,MAAM,cAAc,CAAC;AAEvD,MAAM,CAAC,KAAK,UAAU,OAAO,CAAC,IAAc;IAC1C,MAAM,IAAI,GAAG,YAAY,CAAC,IAAI,CAAC,CAAC;IAChC,MAAM,EAAE,MAAM,EAAE,WAAW,EAAE,GAAG,qBAAqB,CAAC;QACpD,IAAI,EAAE,IAAI;QACV,gBAAgB,EAAE,IAAI;QACtB,OAAO,EAAE;YACP,GAAG,cAAc;YACjB,aAAa,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,mBAAmB,EAAE;YAC/D,kBAAkB,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,OAAO,EAAE,KAAK,EAAE;YACvD,SAAS,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;YAC7B,OAAO,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;YAC3B,GAAG,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;SACxB;KACF,CAAC,CAAC;IAEH,IAAI,WAAW,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC7B,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;QAC3B,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;IACvB,CAAC;IAED,MAAM,EAAE,eAAe,EAAE,GAAG,MAAM,MAAM,CAAC,8BAA8B,CAAC,CAAC;IACzE,MAAM,KAAK,GAAgB,eAAe,CAAC,OAAO,CAAC,MAAM,CAAC,aAAa,CAAW,CAAC,CAAC,CAAC;IAErF,IAAI,WAAW,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC7B,MAAM,kBAAkB,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,CAAC,CAAC;QAC9D,OAAO;IACT,CAAC;IAED,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC3B,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;QAC3B,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;IACvB,CAAC;IAED,MAAM,CAAC,GAAG,EAAE,GAAG,CAAC,GAAG,WAAW,CAAC;IAC/B,MAAM,EAAE,GAAG,uBAAuB,CAAC,MAAM,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,GAAG,EAAE,IAAI,CAAC,CAAC;IACpE,MAAM,EAAE,GAAG,uBAAuB,CAAC,MAAM,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,GAAG,EAAE,IAAI,CAAC,CAAC;IAEpE,OAAO,CAAC,GAAG,CAAC,aAAa,GAAG,MAAM,GAAG,IAAI,CAAC,CAAC;IAC3C,MAAM,EAAE,GAA+B,EAAE,CAAC,IAAI,EAAE,OAAO,CAAC;IACxD,MAAM,EAAE,GAA+B,EAAE,CAAC,IAAI,EAAE,OAAO,CAAC;IACxD,IAAI,EAAE,IAAI,EAAE,EAAE,CAAC;QACb,OAAO,CAAC,GAAG,CAAC,WAAW,EAAE,EAAE,WAAW,IAAI,GAAG,GAAG,EAAE,EAAE,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,EAAE,MAAM,IAAI,GAAG,OAAO,EAAE,EAAE,WAAW,IAAI,GAAG,GAAG,EAAE,EAAE,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,EAAE,MAAM,IAAI,GAAG,GAAG,CAAC,CAAC;IAC3K,CAAC;IAED,MAAM,EAAE,gCAAgC,EAAE,2BAA2B,EAAE,GAAG,MAAM,MAAM,CAAC,kCAAkC,CAAC,CAAC;IAC3H,MAAM,aAAa,GAAG,2BAA2B,CAAC,gCAAgC,CAAC,EAAE,EAAE,EAAE,CAAC,EAAE,IAAI,CAAC,CAAC;IAClG,IAAI,aAAa;QAAE,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,KAAK,aAAa,MAAM,CAAC,CAAC;IAElE,MAAM,QAAQ,GAAG,CAAC,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,EAAE,QAAQ,IAAI,EAAE,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,IAAI,EAAE,QAAQ,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IAC5F,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,MAAM,EAAE,GAA+B,EAAE,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,CAAC;QAC7D,MAAM,EAAE,GAA+B,EAAE,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,CAAC;QAC7D,IAAI,CAAC,EAAE,IAAI,CAAC,EAAE;YAAE,SAAS;QAEzB,OAAO,CAAC,GAAG,CAAC,QAAQ,OAAO,GAAG,CAAC,CAAC;QAChC,MAAM,MAAM,GAAoB,EAAE,EAAE,iBAAiB,IAAI,GAAG,CAAC;QAC7D,MAAM,MAAM,GAAoB,EAAE,EAAE,iBAAiB,IAAI,GAAG,CAAC;QAC7D,MAAM,UAAU,GAAG,OAAO,MAAM,KAAK,QAAQ,IAAI,OAAO,MAAM,KAAK,QAAQ;YACzE,CAAC,CAAC,KAAK,MAAM,GAAG,MAAM,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,CAAC,MAAM,GAAG,MAAM,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;YACnE,CAAC,CAAC,EAAE,CAAC;QACP,OAAO,CAAC,GAAG,CAAC,gBAAgB,MAAM,MAAM,MAAM,GAAG,UAAU,EAAE,CAAC,CAAC;QAC/D,OAAO,CAAC,GAAG,CAAC,gBAAgB,EAAE,EAAE,WAAW,IAAI,GAAG,MAAM,EAAE,EAAE,WAAW,IAAI,GAAG,EAAE,CAAC,CAAC;QAElF,IAAI,EAAE,EAAE,YAAY,IAAI,IAAI,IAAI,EAAE,EAAE,YAAY,IAAI,IAAI,EAAE,CAAC;YACzD,OAAO,CAAC,GAAG,CAAC,gBAAgB,EAAE,EAAE,YAAY,IAAI,GAAG,MAAM,EAAE,EAAE,YAAY,IAAI,GAAG,EAAE,CAAC,CAAC;YACpF,MAAM,GAAG,GAAG,EAAE,EAAE,eAAe,IAAI,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,eAAe,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC;YAC5F,MAAM,GAAG,GAAG,EAAE,EAAE,eAAe,IAAI,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,eAAe,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC;YAC5F,OAAO,CAAC,GAAG,CAAC,gBAAgB,GAAG,MAAM,GAAG,EAAE,CAAC,CAAC;QAC9C,CAAC;QAED,MAAM,KAAK,GAAG,EAAE,EAAE,gBAAgB,IAAI,CAAC,CAAC;QACxC,MAAM,KAAK,GAAG,EAAE,EAAE,gBAAgB,IAAI,CAAC,CAAC;QACxC,MAAM,SAAS,GAAG,EAAE,EAAE,gBAAgB,KAAK,KAAK,CAAC;QACjD,MAAM,SAAS,GAAG,EAAE,EAAE,gBAAgB,KAAK,KAAK,CAAC;QACjD,MAAM,OAAO,GAAG,CAAC,IAAY,EAAE,QAAiB,EAAU,EAAE,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC;QACpG,MAAM,OAAO,GAAG,SAAS,IAAI,SAAS,IAAI,KAAK,GAAG,CAAC;YACjD,CAAC,CAAC,KAAK,KAAK,GAAG,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,KAAK,GAAG,KAAK,CAAC,GAAG,KAAK,CAAC,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI;YAClF,CAAC,CAAC,EAAE,CAAC;QACP,OAAO,CAAC,GAAG,CAAC,gBAAgB,OAAO,CAAC,KAAK,EAAE,SAAS,CAAC,MAAM,OAAO,CAAC,KAAK,EAAE,SAAS,CAAC,GAAG,OAAO,EAAE,CAAC,CAAC;QAElG,MAAM,EAAE,GAAG,EAAE,CAAC,IAAI,EAAE,cAAc,EAAE,CAAC,OAAO,CAAC,CAAC;QAC9C,MAAM,EAAE,GAAG,EAAE,CAAC,IAAI,EAAE,cAAc,EAAE,CAAC,OAAO,CAAC,CAAC;QAC9C,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC;YAC1B,OAAO,CAAC,GAAG,CAAC,gBAAgB,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,YAAY,CAAC,CAAC;QAC9E,CAAC;IACH,CAAC;IACD,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;AAClB,CAAC;AAED,KAAK,UAAU,kBAAkB,CAC/B,QAAgB,EAChB,KAAkB,EAClB,KAAmD,EACnD,IAAa;IAEb,MAAM,MAAM,GAAG,uBAAuB,CAAC,MAAM,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,QAAQ,EAAE,IAAI,CAAC,CAAC;IAClF,MAAM,EAAE,2BAA2B,EAAE,2BAA2B,EAAE,GAAG,MAAM,MAAM,CAAC,kCAAkC,CAAC,CAAC;IACtH,MAAM,aAAa,GAAG,2BAA2B,CAAC,2BAA2B,CAAC,MAAM,CAAC,EAAE,IAAI,CAAC,CAAC;IAC7F,IAAI,aAAa;QAAE,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,KAAK,aAAa,MAAM,CAAC,CAAC;IAElE,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,EAAE,QAAQ,IAAI,EAAE,CAAC;IAC7C,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACxB,OAAO,CAAC,KAAK,CAAC,4DAA4D,CAAC,CAAC;QAC5E,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;IACvB,CAAC;IACD,MAAM,OAAO,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;IAC5B,MAAM,SAAS,GAAI,KAAK,CAAC,OAA8B,IAAI,QAAQ,CAAC,CAAC,CAAC,CAAC;IACvE,IAAI,CAAC,QAAQ,CAAC,QAAQ,CAAC,SAAS,CAAC,EAAE,CAAC;QAClC,OAAO,CAAC,KAAK,CAAC,YAAY,SAAS,+BAA+B,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACzF,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;IACvB,CAAC;IAED,MAAM,SAAS,GAAG,KAAK,CAAC,SAAS,IAAI,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IACxE,MAAM,eAAe,GAAG,OAAO,CAAC,KAAK,CAAC,kBAAkB,CAAC,CAAC,CAAC;IAC3D,MAAM,IAAI,GAAG,KAAK,CAAC,GAAG,IAAI,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACjF,MAAM,IAAI,GAAuK,EAAE,CAAC;IAEpL,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,OAAO,IAAI,EAAE,EAAE,CAAC;QACzC,MAAM,CAAC,GAAG,KAAK,CAAC,QAAQ,EAAE,CAAC,OAAO,CAAC,CAAC;QACpC,MAAM,CAAC,GAAG,KAAK,CAAC,QAAQ,EAAE,CAAC,SAAS,CAAC,CAAC;QACtC,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC;YAAE,SAAS;QACvB,MAAM,KAAK,GAAG,CAAC,CAAC,cAAc,IAAI,CAAC,CAAC,QAAQ,IAAI,CAAC,CAAC;QAClD,MAAM,KAAK,GAAG,CAAC,CAAC,cAAc,IAAI,CAAC,CAAC,QAAQ,IAAI,CAAC,CAAC;QAClD,IAAI,CAAC,IAAI,CAAC;YACR,EAAE,EAAE,KAAK,CAAC,SAAS;YACnB,KAAK,EAAE,CAAC,CAAC,aAAa,EAAE,SAAS;YACjC,KAAK,EAAE,CAAC,CAAC,aAAa,EAAE,SAAS;YACjC,IAAI,EAAE,CAAC,CAAC,aAAa,EAAE,aAAa;YACpC,IAAI,EAAE,CAAC,CAAC,aAAa,EAAE,aAAa;YACpC,MAAM,EAAE,CAAC,CAAC,aAAa,EAAE,UAAU;YACnC,MAAM,EAAE,CAAC,CAAC,aAAa,EAAE,UAAU;YACnC,KAAK;YACL,KAAK;YACL,KAAK,EAAE,MAAM,CAAC,CAAC,KAAK,GAAG,KAAK,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;SAC1C,CAAC,CAAC;IACL,CAAC;IAED,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC;IAC3D,IAAI,QAAQ,GAAG,IAAI,CAAC;IACpB,IAAI,eAAe;QAAE,QAAQ,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,GAAG,SAAS,CAAC,CAAC;IAC5E,IAAI,IAAI,KAAK,SAAS;QAAE,QAAQ,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC;IAE3D,OAAO,CAAC,GAAG,CAAC,0BAA0B,SAAS,OAAO,OAAO,YAAY,QAAQ,GAAG,CAAC,CAAC;IACtF,IAAI,eAAe;QAAE,OAAO,CAAC,GAAG,CAAC,mCAAmC,SAAS,GAAG,CAAC,CAAC;IAClF,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;IAChB,OAAO,CAAC,GAAG,CAAC,4FAA4F,CAAC,CAAC;IAC1G,OAAO,CAAC,GAAG,CAAC,IAAI,GAAG,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC;IACpC,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC1B,OAAO,CAAC,GAAG,CAAC,eAAe,CAAC,CAAC,CAAC,0BAA0B,CAAC,CAAC,CAAC,uBAAuB,CAAC,CAAC;QACpF,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;QAChB,OAAO;IACT,CAAC;IAED,MAAM,QAAQ,GAAG,CAAC,CAAqB,EAAE,CAAqB,EAAU,EAAE;QACxE,MAAM,EAAE,GAAG,OAAO,CAAC,KAAK,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC;QACtD,MAAM,EAAE,GAAG,OAAO,CAAC,KAAK,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC;QACtD,OAAO,GAAG,EAAE,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC;IACpC,CAAC,CAAC;IACF,KAAK,MAAM,GAAG,IAAI,QAAQ,EAAE,CAAC;QAC3B,MAAM,IAAI,GAAG,GAAG,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;QACtC,OAAO,CAAC,GAAG,CACT,KAAK,GAAG,CAAC,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,MAAM,CAAC,EAAE,CAAC,EAAE;cACnC,GAAG,GAAG,IAAI,GAAG,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE;cAC/C,GAAG,GAAG,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,MAAM,CAAC,EAAE,CAAC,EAAE;cACnE,GAAG,QAAQ,CAAC,GAAG,CAAC,KAAK,EAAE,GAAG,CAAC,KAAK,CAAC,IAAI,QAAQ,CAAC,GAAG,CAAC,IAAI,EAAE,GAAG,CAAC,IAAI,CAAC,IAAI,QAAQ,CAAC,GAAG,CAAC,MAAM,EAAE,GAAG,CAAC,MAAM,CAAC,EAAE,CAC1G,CAAC;IACJ,CAAC;IACD,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;IAChB,OAAO,CAAC,GAAG,CAAC,aAAa,QAAQ,CAAC,MAAM,OAAO,IAAI,CAAC,MAAM,0BAA0B,CAAC,CAAC;IACtF,IAAI,eAAe,EAAE,CAAC;QACpB,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC;QAC1B,MAAM,WAAW,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,GAAG,SAAS,CAAC,CAAC,MAAM,CAAC;QACnE,OAAO,CAAC,GAAG,CAAC,sBAAsB,WAAW,IAAI,KAAK,aAAa,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,WAAW,GAAG,KAAK,CAAC,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;IACnI,CAAC;IACD,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;AAClB,CAAC;AAED,SAAS,KAAK,CAAC,IAAa;IAC1B,OAAO,IAAI,KAAK,IAAI;QAClB,CAAC,CAAC;YACE,KAAK;YACL,mFAAmF;YACnF,+CAA+C;SAChD,CAAC,IAAI,CAAC,IAAI,CAAC;QACd,CAAC,CAAC;YACE,QAAQ;YACR,mFAAmF;YACnF,+CAA+C;SAChD,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACnB,CAAC"}

package/dist/src/cli/commands/export-saturation.d.ts

@@ -0,0 +1,2 @@
+export declare function execute(argv: string[]): Promise<void>;
+//# sourceMappingURL=export-saturation.d.ts.map

package/dist/src/cli/commands/export-saturation.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"export-saturation.d.ts","sourceRoot":"","sources":["../../../../src/cli/commands/export-saturation.ts"],"names":[],"mappings":"AAQA,wBAAsB,OAAO,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC,CAsD3D"}