oh-my-knowledge 0.31.0 → 0.32.0

package/dist/eval-workflows/evaluation-pipeline.js

@@ -0,0 +1,179 @@
+/**
+ * evaluation pipeline orchestrator —— 单次 evaluation run 的「执行+收尾」编排。
+ *
+ * 主入口 `executeEvaluationPipeline` 把以下几段串成线性时序:
+ *   1. `initializeEvaluationRunState` 建 run / job 元数据并写 jobStore
+ *   2. resolve judge models + executor map(必须先于 preflight,见函数内注释)
+ *   3. `preflight` + `preflightAllJudges` LLM 连通性
+ *   4. `emitPowerWarnings` + `emitIsolationWarnings` 结构性 / 隔离性预警
+ *   5. `executeTasks` 真正跑用例
+ *   6. `finalizeSuccessfulRun` + `finalizeEvaluationReport` 收尾 + 后置分析
+ *   7. `persistReport` + `persistSuccessfulJob` 落盘
+ *   失败路径走 `persistFailedJob`, finally 关 MCP server。
+ *
+ * 4 块被拆出去的 helper 都在 `./evaluation-pipeline/` 目录下,与本文件 1:1 协作:
+ *   - run-state.ts: EvaluationRunState 生命周期(init / finalizeSuccess / persistSuccess / persistFailed)
+ *   - test-set-hash.ts: 测试集水印 hash(spec §7.1)
+ *   - preflight-warnings.ts: power + isolation 预警
+ *   - report-finalize.ts: report 后置分析装配
+ *
+ * 兼容 re-export: `buildPowerWarnings` / `buildIsolationWarnings` /
+ * `_computeTestSetHashForTest` 保留在本模块的 export surface,避免 test 与
+ * `run-evaluation.ts:275` 动态 import 路径改动。
+ */
+import { aggregateReport, DEFAULT_OUTPUT_DIR, generateRunId, persistReport } from '../eval-core/evaluation-reporting.js';
+import { executeTasks, preflight, preflightAllJudges } from '../eval-core/evaluation-execution.js';
+import { stopAllServers } from '../inputs/mcp-resolver.js';
+import { finalizeSuccessfulRun, initializeEvaluationRunState, persistFailedJob, persistSuccessfulJob, } from './evaluation-pipeline/run-state.js';
+import { finalizeEvaluationReport } from './evaluation-pipeline/report-finalize.js';
+import { emitIsolationWarnings, emitPowerWarnings } from './evaluation-pipeline/preflight-warnings.js';
+// 兼容 re-export:测试与 run-evaluation.ts 动态 import 仍打 evaluation-pipeline.js
+export { buildPowerWarnings, buildIsolationWarnings } from './evaluation-pipeline/preflight-warnings.js';
+export { _computeTestSetHashForTest } from './evaluation-pipeline/test-set-hash.js';
+export async function executeEvaluationPipeline({ samplesPath, samplesBaseDir, samplesSourceFiles, skillDir, samples, tasks, artifacts, model, judgeModel, noJudge, executorName, judgeExecutorName, executor, judgeExecutor, outputDir = DEFAULT_OUTPUT_DIR, project, owner, tags, blind = false, concurrency = 1, timeoutMs, noCache = false, jobStore = null, persistJob = true, onProgress = null, skipConnectivity = false, verbose = false, retry = 0, existingResults, 
+// requires 现在由 runEvaluation 上游传给 doctor 处理; eval-pipeline 不再用
+// 但保留接口字段,避免破 programmatic API (类型层面接收, 内部忽略)
+requires: _requires, layeredStats = false, repeat, batch, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias = true, budget, strictBaseline, runId, lang = 'zh', effort, noDiagnostic, }) {
+    const variantNames = artifacts.map((artifact) => artifact.name);
+    const runState = await initializeEvaluationRunState({
+        samplesPath,
+        skillDir,
+        artifacts,
+        model,
+        judgeModel,
+        noJudge,
+        executorName,
+        judgeExecutorName,
+        concurrency,
+        timeoutMs,
+        noCache,
+        blind,
+        project,
+        owner,
+        tags,
+        runId: runId ?? generateRunId(variantNames),
+        jobStore,
+        persistJob,
+        repeat,
+        batch,
+        judgeRepeat,
+        judgeModels,
+        bootstrap,
+        bootstrapSamples,
+        lengthDebias,
+        budget,
+        effort,
+    });
+    try {
+        // 解析 judge config + 配套 executor map 一次,后续 preflight 与 executeTasks 共用。
+        // production 路径 judgeModels 始终非空(RunConfig 必填);fallback 仅兼容 internal
+        // legacy caller 还在传 single judgeModel + judgeExecutorName 的场景。
+        //
+        // executor map 必须先于 preflight build,否则 ensemble 中第二、第三个 judge 配错
+        // (404 model / executor 不存在 / auth fail)只会在 grading 半途才暴露 — 那时已经
+        // 浪费 task execution 成本 + 报告里也不会有 judgeAgreement。
+        const resolvedJudgeModels = judgeModels && judgeModels.length > 0
+            ? judgeModels
+            : [{ executor: executorName, model: judgeModel }];
+        let resolvedJudgeExecutors;
+        if (judgeModels && judgeModels.length > 0) {
+            const { createExecutor } = await import('../executors/index.js');
+            resolvedJudgeExecutors = {};
+            for (const jc of resolvedJudgeModels) {
+                if (!resolvedJudgeExecutors[jc.executor]) {
+                    resolvedJudgeExecutors[jc.executor] = createExecutor(jc.executor);
+                }
+            }
+        }
+        else {
+            resolvedJudgeExecutors = { [executorName]: judgeExecutor };
+        }
+        if (!skipConnectivity) {
+            if (onProgress)
+                onProgress({ phase: 'preflight', jobId: runState.jobId });
+            // LLM 连通性: eval 唯一职责。doctor 在 runEvaluation 上游已跑,
+            // 包含 dep / 结构 / 元数据 / 契约检查。这里只剩 executor + 所有 judge。
+            await preflight(executor, model);
+            if (!noJudge) {
+                await preflightAllJudges(resolvedJudgeModels, resolvedJudgeExecutors);
+            }
+        }
+        // Structural power warnings — print to stderr after preflight passes, before
+        // tasks start. These are *not* MDE / power-analysis predictions (we don't have
+        // σ before the run); they're hard-floor + experience-based thresholds. Verdict
+        // gate (computeVerdict) handles real power claims post-hoc.
+        emitPowerWarnings(samples.length, repeat ?? 1, lang);
+        // Isolation pre-flight warning (--no-strict-baseline + ~/.claude/skills/ non-empty)
+        emitIsolationWarnings(artifacts, strictBaseline);
+        const { results, totalCostUSD, skipped, budgetExhausted } = await executeTasks({
+            tasks,
+            executor,
+            executorName,
+            model,
+            noJudge,
+            samplesPath,
+            samplesBaseDir,
+            concurrency,
+            timeoutMs,
+            noCache,
+            verbose,
+            onProgress,
+            retry,
+            existingResults,
+            judgeRepeat,
+            judgeModels: resolvedJudgeModels,
+            judgeExecutors: resolvedJudgeExecutors,
+            lengthDebias,
+            budget,
+            effort,
+            noDiagnostic,
+        });
+        if (skipped > 0 && onProgress) {
+            onProgress({ phase: 'done', completed: tasks.length, total: tasks.length, sample_id: '', variant: '', skipped: true });
+        }
+        const { run, job } = finalizeSuccessfulRun(runState);
+        const report = finalizeEvaluationReport({
+            report: aggregateReport({
+                runId: runState.runId,
+                variants: variantNames,
+                model,
+                judgeModel,
+                noJudge,
+                executorName,
+                samples,
+                tasks,
+                results,
+                totalCostUSD,
+                artifacts,
+                request: runState.request,
+                run,
+                job,
+                layeredStats,
+            }),
+            results,
+            artifacts,
+            variantNames,
+            blind,
+            samplesPath,
+            samplesSourceFiles,
+            samples,
+        });
+        if (budgetExhausted) {
+            report.meta.budgetExhausted = true;
+        }
+        if (budget) {
+            report.meta.budget = budget;
+        }
+        const filePath = persistReport(report, outputDir);
+        await persistSuccessfulJob(runState, job);
+        return { report, filePath };
+    }
+    catch (err) {
+        await persistFailedJob(runState, err);
+        throw err;
+    }
+    finally {
+        await stopAllServers();
+    }
+}
+//# sourceMappingURL=evaluation-pipeline.js.map

package/dist/eval-workflows/evaluation-pipeline.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluation-pipeline.js","sourceRoot":"","sources":["../../src/eval-workflows/evaluation-pipeline.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;GAsBG;AAEH,OAAO,EAAE,eAAe,EAAE,kBAAkB,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM,sCAAsC,CAAC;AACzH,OAAO,EAAE,YAAY,EAAE,SAAS,EAAE,kBAAkB,EAAE,MAAM,sCAAsC,CAAC;AAEnG,OAAO,EAAE,cAAc,EAAE,MAAM,2BAA2B,CAAC;AAW3D,OAAO,EACL,qBAAqB,EACrB,4BAA4B,EAC5B,gBAAgB,EAChB,oBAAoB,GACrB,MAAM,oCAAoC,CAAC;AAC5C,OAAO,EAAE,wBAAwB,EAAE,MAAM,0CAA0C,CAAC;AACpF,OAAO,EAAE,qBAAqB,EAAE,iBAAiB,EAAE,MAAM,6CAA6C,CAAC;AAEvG,yEAAyE;AACzE,OAAO,EAAE,kBAAkB,EAAE,sBAAsB,EAAE,MAAM,6CAA6C,CAAC;AACzG,OAAO,EAAE,0BAA0B,EAAE,MAAM,wCAAwC,CAAC;AAoEpF,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAAC,EAC9C,WAAW,EACX,cAAc,EACd,kBAAkB,EAClB,QAAQ,EACR,OAAO,EACP,KAAK,EACL,SAAS,EACT,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,QAAQ,EACR,aAAa,EACb,SAAS,GAAG,kBAAkB,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,KAAK,GAAG,KAAK,EACb,WAAW,GAAG,CAAC,EACf,SAAS,EACT,OAAO,GAAG,KAAK,EACf,QAAQ,GAAG,IAAI,EACf,UAAU,GAAG,IAAI,EACjB,UAAU,GAAG,IAAI,EACjB,gBAAgB,GAAG,KAAK,EACxB,OAAO,GAAG,KAAK,EACf,KAAK,GAAG,CAAC,EACT,eAAe;AACf,+DAA+D;AAC/D,8CAA8C;AAC9C,QAAQ,EAAE,SAAS,EACnB,YAAY,GAAG,KAAK,EACpB,MAAM,EACN,KAAK,EACL,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,GAAG,IAAI,EACnB,MAAM,EACN,cAAc,EACd,KAAK,EACL,IAAI,GAAG,IAAI,EACX,MAAM,EACN,YAAY,GACc;IAC1B,MAAM,YAAY,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;IAChE,MAAM,QAAQ,GAAG,MAAM,4BAA4B,CAAC;QAClD,WAAW;QACX,QAAQ;QACR,SAAS;QACT,KAAK;QACL,UAAU;QACV,OAAO;QACP,YAAY;QACZ,iBAAiB;QACjB,WAAW;QACX,SAAS;QACT,OAAO;QACP,KAAK;QACL,OAAO;QACP,KAAK;QACL,IAAI;QACJ,KAAK,EAAE,KAAK,IAAI,aAAa,CAAC,YAAY,CAAC;QAC3C,QAAQ;QACR,UAAU;QACV,MAAM;QACN,KAAK;QACL,WAAW;QACX,WAAW;QACX,SAAS;QACT,gBAAgB;QAChB,YAAY;QACZ,MAAM;QACN,MAAM;KACP,CAAC,CAAC;IAEH,IAAI,CAAC;QACH,uEAAuE;QACvE,qEAAqE;QACrE,+DAA+D;QAC/D,EAAE;QACF,iEAAiE;QACjE,iEAAiE;QACjE,iDAAiD;QACjD,MAAM,mBAAmB,GAAG,WAAW,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC;YAC/D,CAAC,CAAC,WAAW;YACb,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,YAAY,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,CAAC;QACpD,IAAI,sBAAkD,CAAC;QACvD,IAAI,WAAW,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC1C,MAAM,EAAE,cAAc,EAAE,GAAG,MAAM,MAAM,CAAC,uBAAuB,CAAC,CAAC;YACjE,sBAAsB,GAAG,EAAE,CAAC;YAC5B,KAAK,MAAM,EAAE,IAAI,mBAAmB,EAAE,CAAC;gBACrC,IAAI,CAAC,sBAAsB,CAAC,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC;oBACzC,sBAAsB,CAAC,EAAE,CAAC,QAAQ,CAAC,GAAG,cAAc,CAAC,EAAE,CAAC,QAAQ,CAAC,CAAC;gBACpE,CAAC;YACH,CAAC;QACH,CAAC;aAAM,CAAC;YACN,sBAAsB,GAAG,EAAE,CAAC,YAAY,CAAC,EAAE,aAAa,EAAE,CAAC;QAC7D,CAAC;QAED,IAAI,CAAC,gBAAgB,EAAE,CAAC;YACtB,IAAI,UAAU;gBAAE,UAAU,CAAC,EAAE,KAAK,EAAE,WAAW,EAAE,KAAK,EAAE,QAAQ,CAAC,KAAK,EAAE,CAAC,CAAC;YAC1E,kDAAkD;YAClD,qDAAqD;YACrD,MAAM,SAAS,CAAC,QAAQ,EAAE,KAAK,CAAC,CAAC;YACjC,IAAI,CAAC,OAAO,EAAE,CAAC;gBACb,MAAM,kBAAkB,CAAC,mBAAmB,EAAE,sBAAsB,CAAC,CAAC;YACxE,CAAC;QACH,CAAC;QAED,6EAA6E;QAC7E,+EAA+E;QAC/E,+EAA+E;QAC/E,4DAA4D;QAC5D,iBAAiB,CAAC,OAAO,CAAC,MAAM,EAAE,MAAM,IAAI,CAAC,EAAE,IAAI,CAAC,CAAC;QACrD,oFAAoF;QACpF,qBAAqB,CAAC,SAAS,EAAE,cAAc,CAAC,CAAC;QAEjD,MAAM,EAAE,OAAO,EAAE,YAAY,EAAE,OAAO,EAAE,eAAe,EAAE,GAAG,MAAM,YAAY,CAAC;YAC7E,KAAK;YACL,QAAQ;YACR,YAAY;YACZ,KAAK;YACL,OAAO;YACP,WAAW;YACX,cAAc;YACd,WAAW;YACX,SAAS;YACT,OAAO;YACP,OAAO;YACP,UAAU;YACV,KAAK;YACL,eAAe;YACf,WAAW;YACX,WAAW,EAAE,mBAAmB;YAChC,cAAc,EAAE,sBAAsB;YACtC,YAAY;YACZ,MAAM;YACN,MAAM;YACN,YAAY;SACb,CAAC,CAAC;QACH,IAAI,OAAO,GAAG,CAAC,IAAI,UAAU,EAAE,CAAC;YAC9B,UAAU,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,KAAK,CAAC,MAAM,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,EAAE,SAAS,EAAE,EAAE,EAAE,OAAO,EAAE,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;QACzH,CAAC;QAED,MAAM,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,qBAAqB,CAAC,QAAQ,CAAC,CAAC;QACrD,MAAM,MAAM,GAAG,wBAAwB,CAAC;YACtC,MAAM,EAAE,eAAe,CAAC;gBACtB,KAAK,EAAE,QAAQ,CAAC,KAAK;gBACrB,QAAQ,EAAE,YAAY;gBACtB,KAAK;gBACL,UAAU;gBACV,OAAO;gBACP,YAAY;gBACZ,OAAO;gBACP,KAAK;gBACL,OAAO;gBACP,YAAY;gBACZ,SAAS;gBACT,OAAO,EAAE,QAAQ,CAAC,OAAO;gBACzB,GAAG;gBACH,GAAG;gBACH,YAAY;aACb,CAAC;YACF,OAAO;YACP,SAAS;YACT,YAAY;YACZ,KAAK;YACL,WAAW;YACX,kBAAkB;YAClB,OAAO;SACR,CAAC,CAAC;QACH,IAAI,eAAe,EAAE,CAAC;YACpB,MAAM,CAAC,IAAI,CAAC,eAAe,GAAG,IAAI,CAAC;QACrC,CAAC;QACD,IAAI,MAAM,EAAE,CAAC;YACX,MAAM,CAAC,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;QAC9B,CAAC;QACD,MAAM,QAAQ,GAAG,aAAa,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;QAClD,MAAM,oBAAoB,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;QAC1C,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,CAAC;IAC9B,CAAC;IAAC,OAAO,GAAY,EAAE,CAAC;QACtB,MAAM,gBAAgB,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;QACtC,MAAM,GAAG,CAAC;IACZ,CAAC;YAAS,CAAC;QACT,MAAM,cAAc,EAAE,CAAC;IACzB,CAAC;AACH,CAAC"}

package/dist/eval-workflows/evaluation-preparation.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluation-preparation.d.ts","sourceRoot":"","sources":["../../src/eval-workflows/evaluation-preparation.ts"],"names":[],"mappings":"AAOA,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,oCAAoC,CAAC;AACjF,OAAO,KAAK,EAAE,QAAQ,EAAE,WAAW,EAAc,MAAM,EAAE,IAAI,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAEtG,MAAM,WAAW,qBAAqB;IACpC,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,KAAK,EAAE,IAAI,EAAE,CAAC;IACd,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,QAAQ,CAAC,EAAE,sBAAsB,CAAC;IAClC;qDACiD;IACjD,cAAc,EAAE,MAAM,CAAC;IACvB,6DAA6D;IAC7D,kBAAkB,EAAE,MAAM,EAAE,CAAC;CAC9B;AAED,wBAAsB,oBAAoB,CAAC,EACzC,WAAW,EACX,QAAQ,EACR,YAAY,EACZ,SAAS,EACT,MAAM,EACN,SAAS,EACT,cAAc,EACd,oBAAoB,GACrB,EAAE;IACD,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,WAAW,EAAE,CAAC;IAC5B,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;IACvB,MAAM,EAAE,OAAO,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,0DAA0D;IAC1D,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,oFAAoF;IACpF,oBAAoB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;CACjD,GAAG,OAAO,CAAC,qBAAqB,CAAC,CA+CjC;AAED,wBAAgB,qBAAqB,CAAC,EACpC,KAAK,EACL,WAAW,EACX,YAAY,EACZ,WAAW,EACX,QAAQ,EACR,KAAK,EACL,YAAY,GACb,EAAE;IACD,KAAK,EAAE,MAAM,CAAC;IACd,WAAW,EAAE,WAAW,EAAE,CAAC;IAC3B,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,IAAI,EAAE,CAAC;IACd,YAAY,EAAE,MAAM,EAAE,CAAC;CACxB;;;;;;;;;;;;;;;;;;;;;;;;EA6BA"}

package/dist/eval-workflows/evaluation-preparation.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluation-preparation.js","sourceRoot":"","sources":["../../src/eval-workflows/evaluation-preparation.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,uBAAuB,EAAE,MAAM,8BAA8B,CAAC;AACvE,OAAO,EAAE,WAAW,EAAE,MAAM,2BAA2B,CAAC;AACxD,OAAO,EAAE,gBAAgB,EAAE,MAAM,2BAA2B,CAAC;AAC7D,OAAO,EAAE,kBAAkB,EAAE,MAAM,oCAAoC,CAAC;AACxE,OAAO,EAAE,aAAa,EAAE,cAAc,EAAE,MAAM,2BAA2B,CAAC;AAC1E,OAAO,EAAE,WAAW,EAAE,MAAM,0BAA0B,CAAC;AAiBvD,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,EACzC,WAAW,EACX,QAAQ,EACR,YAAY,EACZ,SAAS,EACT,MAAM,EACN,SAAS,EACT,cAAc,EACd,oBAAoB,GAYrB;IACC,MAAM,EAAE,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE,cAAc,EAAE,WAAW,EAAE,kBAAkB,EAAE,GAAG,WAAW,CAAC,WAAW,CAAC,CAAC;IAEjH,4EAA4E;IAC5E,MAAM,kBAAkB,GAAG,YAAY,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACjE,MAAM,iBAAiB,GAAG,SAAS,IAAI,gBAAgB,CACrD,OAAO,CAAC,QAAQ,CAAC,EACjB,kBAAkB,EAClB,EAAE,cAAc,EAAE,oBAAoB,EAAE,CACzC,CAAC;IAEF,iFAAiF;IACjF,MAAM,UAAU,GAAwC,EAAE,CAAC;IAC3D,KAAK,MAAM,IAAI,IAAI,YAAY;QAAE,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC;IACnE,KAAK,MAAM,QAAQ,IAAI,iBAAiB,EAAE,CAAC;QACzC,IAAI,QAAQ,CAAC,cAAc;YAAE,SAAS,CAAE,gDAAgD;QACxF,MAAM,IAAI,GAAG,UAAU,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;QACvC,IAAI,IAAI;YAAE,QAAQ,CAAC,cAAc,GAAG,IAAI,CAAC;IAC3C,CAAC;IAED,IAAI,CAAC,MAAM,EAAE,CAAC;QACZ,MAAM,UAAU,GAAsB,aAAa,CAAC,SAAS,CAAC,CAAC;QAC/D,MAAM,WAAW,GAAG,UAAU,CAAC,CAAC,CAAC,MAAM,cAAc,CAAC,OAAO,EAAE,UAAU,CAAC,CAAC,CAAC,CAAC,IAAI,GAAG,EAAU,CAAC;QAC/F,MAAM,WAAW,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC;IAC1C,CAAC;IAED,IAAI,iBAAiB,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACnC,MAAM,IAAI,KAAK,CACb,sBAAsB;cACpB,qBAAqB,OAAO,CAAC,QAAQ,CAAC,IAAI;cAC1C,sCAAsC;cACtC,wEAAwE,CAC3E,CAAC;IACJ,CAAC;IAED,MAAM,KAAK,GAAG,uBAAuB,CAAC,OAAO,EAAE,iBAAiB,CAAC,CAAC;IAClE,MAAM,YAAY,GAAG,iBAAiB,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;IAExE,OAAO;QACL,OAAO;QACP,SAAS,EAAE,iBAAiB;QAC5B,KAAK;QACL,YAAY;QACZ,QAAQ;QACR,cAAc;QACd,kBAAkB;KACnB,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,qBAAqB,CAAC,EACpC,KAAK,EACL,WAAW,EACX,YAAY,EACZ,WAAW,EACX,QAAQ,EACR,KAAK,EACL,YAAY,GASb;IACC,OAAO;QACL,MAAM,EAAE,IAAa;QACrB,KAAK;QACL,WAAW;QACX,QAAQ,EAAE,YAAY;QACtB,QAAQ,EAAE,YAAY;QACtB,WAAW;QACX,QAAQ;QACR,UAAU,EAAE,KAAK,CAAC,MAAM;QACxB,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE;YACxB,MAAM,MAAM,GAAG,kBAAkB,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;YACjD,OAAO;gBACL,SAAS,EAAE,IAAI,CAAC,SAAS;gBACzB,OAAO,EAAE,IAAI,CAAC,OAAO;gBACrB,YAAY,EAAE,IAAI,CAAC,QAAQ,CAAC,IAAI;gBAChC,cAAc,EAAE,IAAI,CAAC,QAAQ,CAAC,MAAM;gBACpC,iBAAiB,EAAE,MAAM,CAAC,iBAAiB;gBAC3C,cAAc,EAAE,MAAM,CAAC,cAAc;gBACrC,cAAc,EAAE,MAAM,CAAC,cAAc;gBACrC,GAAG,EAAE,IAAI,CAAC,GAAG;gBACb,aAAa,EAAE,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC;gBACxC,SAAS,EAAE,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC;gBAC/B,aAAa,EAAE,OAAO,CAAC,IAAI,CAAC,UAAU,EAAE,MAAM,CAAC;gBAC/C,aAAa,EAAE,OAAO,CAAC,IAAI,CAAC,UAAU,IAAI,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC;gBAC9E,SAAS,EAAE,OAAO,CAAC,IAAI,CAAC,eAAe,CAAC;aACzC,CAAC;QACJ,CAAC,CAAC;KACH,CAAC;AACJ,CAAC"}

package/dist/eval-workflows/messages.d.ts

@@ -0,0 +1,4 @@
+import type { Lang } from '../types/shared.js';
+export type EvalWorkflowMessageCode = 'power_warning_tiny_n' | 'power_warning_small_n' | 'power_warning_repeat_one' | 'doctor_gate_blocked';
+export declare function tEvalWorkflowMessage(code: EvalWorkflowMessageCode, lang?: Lang, params?: Record<string, string | number>): string;
+//# sourceMappingURL=messages.d.ts.map

package/dist/eval-workflows/messages.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"messages.d.ts","sourceRoot":"","sources":["../../src/eval-workflows/messages.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,oBAAoB,CAAC;AAE/C,MAAM,MAAM,uBAAuB,GAC/B,sBAAsB,GACtB,uBAAuB,GACvB,0BAA0B,GAC1B,qBAAqB,CAAC;AA4B1B,wBAAgB,oBAAoB,CAClC,IAAI,EAAE,uBAAuB,EAC7B,IAAI,GAAE,IAAmB,EACzB,MAAM,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,GAAG,MAAM,CAAC,GACvC,MAAM,CASR"}

package/dist/eval-workflows/messages.js

@@ -0,0 +1,30 @@
+const MESSAGES = {
+    power_warning_tiny_n: {
+        zh: '⚠ N={n} < 5：仅适合探索，任何结论都不可靠，CI 会很宽。需要决策时建议 ≥20 条评测用例。',
+        en: '⚠ N={n} < 5 (exploration-only): any conclusion is unreliable, CI will be uselessly wide. Decisions need ≥20 cases.',
+    },
+    power_warning_small_n: {
+        zh: '⚠ N={n} < 20：只能识别很大的效果（Cohen\'s d > 0.8），中等效果（d ≈ 0.5）很难检出。要做可靠决策建议 ≥20 条评测用例。',
+        en: '⚠ N={n} < 20 (large-effect-only, Cohen\'s d > 0.8): medium effects (d ≈ 0.5) hard to detect. For confident decisions consider ≥20 cases.',
+    },
+    power_warning_repeat_one: {
+        zh: '⚠ --repeat=1：单轮评测无法测稳定性（CV 会标记为未测量）。用 --repeat 3+ 检测同一 variant 内部方差。',
+        en: '⚠ --repeat=1: single-run cannot measure stability (CV will be marked "not measured"). Use --repeat 3+ to detect within-variant variance.',
+    },
+    doctor_gate_blocked: {
+        zh: 'skill 健康检查未通过，评测已中止。doctor 是评测必经环节，无 skip 选项——请修复上述问题后重跑。',
+        en: 'skill health check failed; evaluation aborted. doctor is mandatory and not skippable — fix the issues above and re-run.',
+    },
+};
+const DEFAULT_LANG = 'zh';
+export function tEvalWorkflowMessage(code, lang = DEFAULT_LANG, params) {
+    const entry = MESSAGES[code];
+    let text = entry[lang] ?? entry[DEFAULT_LANG];
+    if (params) {
+        for (const [k, v] of Object.entries(params)) {
+            text = text.replaceAll(`{${k}}`, String(v));
+        }
+    }
+    return text;
+}
+//# sourceMappingURL=messages.js.map

package/dist/eval-workflows/messages.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"messages.js","sourceRoot":"","sources":["../../src/eval-workflows/messages.ts"],"names":[],"mappings":"AAaA,MAAM,QAAQ,GAAkD;IAC9D,oBAAoB,EAAE;QACpB,EAAE,EAAE,sDAAsD;QAC1D,EAAE,EAAE,oHAAoH;KACzH;IACD,qBAAqB,EAAE;QACrB,EAAE,EAAE,gFAAgF;QACpF,EAAE,EAAE,0IAA0I;KAC/I;IACD,wBAAwB,EAAE;QACxB,EAAE,EAAE,sEAAsE;QAC1E,EAAE,EAAE,0IAA0I;KAC/I;IACD,mBAAmB,EAAE;QACnB,EAAE,EAAE,2DAA2D;QAC/D,EAAE,EAAE,yHAAyH;KAC9H;CACF,CAAC;AAEF,MAAM,YAAY,GAAS,IAAI,CAAC;AAEhC,MAAM,UAAU,oBAAoB,CAClC,IAA6B,EAC7B,OAAa,YAAY,EACzB,MAAwC;IAExC,MAAM,KAAK,GAAG,QAAQ,CAAC,IAAI,CAAC,CAAC;IAC7B,IAAI,IAAI,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,KAAK,CAAC,YAAY,CAAC,CAAC;IAC9C,IAAI,MAAM,EAAE,CAAC;QACX,KAAK,MAAM,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC;YAC5C,IAAI,GAAG,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;QAC9C,CAAC;IACH,CAAC;IACD,OAAO,IAAI,CAAC;AACd,CAAC"}

package/dist/eval-workflows/run-evaluation.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"run-evaluation.d.ts","sourceRoot":"","sources":["../../src/eval-workflows/run-evaluation.ts"],"names":[],"mappings":"AAYA,OAAO,KAAK,EACV,QAAQ,EACR,qBAAqB,EACrB,gBAAgB,EAEhB,QAAQ,EACR,gBAAgB,EAChB,MAAM,EAIN,YAAY,EAGZ,WAAW,EAGZ,MAAM,mBAAmB,CAAC;AAI3B,MAAM,WAAW,iBAAiB;IAChC,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,UAAU,uBAAuB;IAC/B,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,EAAE,QAAQ,GAAG,IAAI,CAAC;IAC3B,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,UAAU,CAAC,EAAE,gBAAgB,GAAG,IAAI,CAAC;IACrC,8CAA8C;IAC9C,gBAAgB,CAAC,EAAE,OAAO,CAAC;IAC3B;;wDAEoD;IACpD,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,6BAA6B;IAC7B,IAAI,CAAC,EAAE,IAAI,GAAG,IAAI,CAAC;IACnB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,OAAO,CAAC;IAIlB,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB;yDACqD;IACrD,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB;wCACoC;IACpC,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB;wEACoE;IACpE,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB;;4DAEwD;IACxD,WAAW,CAAC,EAAE,OAAO,mBAAmB,EAAE,WAAW,EAAE,CAAC;IACxD,8EAA8E;IAC9E,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,2CAA2C;IAC3C,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B;2DACuD;IACvD,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,wBAAwB;IACxB,MAAM,CAAC,EAAE,OAAO,mBAAmB,EAAE,UAAU,CAAC;IAChD,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB;;oDAEgD;IAChD,MAAM,CAAC,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,GAAG,KAAK,CAAC;IACrD,2DAA2D;IAC3D,YAAY,CAAC,EAAE,OAAO,CAAC;CACxB;AAED,MAAM,WAAW,oBAAqB,SAAQ,uBAAuB;IACnE,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,CAAC,EAAE,WAAW,EAAE,CAAC;IAC7B,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;IACvB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,qFAAqF;IACrF,KAAK,CAAC,EAAE,MAAM,CAAC;IACf;;6DAEyD;IACzD,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB;0EACsE;IACtE,oBAAoB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;CACjD;AAED,MAAM,WAAW,yBAA0B,SAAQ,uBAAuB;IACxE,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,eAAe,CAAC,EAAE,CAAC,CAAC,IAAI,EAAE,iBAAiB,KAAK,IAAI,CAAC,GAAG,IAAI,CAAC;IAC7D,qDAAqD;IACrD,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,6EAA6E;IAC7E,oBAAoB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;CACjD;AAED,MAAM,WAAW,kBAAmB,SAAQ,oBAAoB;IAC9D,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,gBAAgB,CAAC,EAAE,CAAC,CAAC,IAAI,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,KAAK,IAAI,CAAC,GAAG,IAAI,CAAC;CAC5E;AAED,UAAU,UAAU;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC;IAC/B,cAAc,EAAE,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACnC,iBAAiB,EAAE,MAAM,CAAC;IAC1B,cAAc,EAAE,MAAM,CAAC;IACvB,cAAc,EAAE,MAAM,CAAC;IACvB,GAAG,EAAE,MAAM,GAAG,IAAI,CAAC;IACnB,aAAa,EAAE,MAAM,CAAC;IACtB,SAAS,EAAE,OAAO,CAAC;IACnB,aAAa,EAAE,OAAO,CAAC;IACvB,aAAa,EAAE,OAAO,CAAC;IACvB,SAAS,EAAE,OAAO,CAAC;CACpB;AAED,UAAU,UAAU;IAClB,MAAM,EAAE,IAAI,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;IACd,WAAW,EAAE,OAAO,mBAAmB,EAAE,WAAW,EAAE,CAAC;IACvD,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,YAAa,SAAQ,UAAU;IAC9C,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,UAAU,EAAE,CAAC;CACrB;AAED,wBAAsB,aAAa,CAAC,EAClC,WAAW,EACX,QAAQ,EACR,YAAiB,EACjB,SAAS,EACT,KAAqB,EACrB,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAe,EACf,MAAc,EACd,KAAa,EACb,WAAe,EACf,SAAS,EACT,OAAe,EACf,YAAuB,EACvB,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,gBAAwB,EACxB,UAAkB,EAClB,IAAW,EACX,SAAS,EACT,OAAe,EACf,KAAS,EACT,MAAM,EACN,KAAK,EACL,YAAoB,EACpB,MAAM,EACN,KAAK,EACL,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,EACZ,MAAM,EACN,cAAc,EACd,oBAAoB,EACpB,MAAM,EACN,YAAY,GACb,EAAE,oBAAoB,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,MAAM,GAAG,YAAY,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CAiL5F;AAED,UAAU,gBAAgB;IACxB,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,iBAAkB,SAAQ,UAAU;IACnD,KAAK,EAAE,IAAI,CAAC;IACZ,cAAc,EAAE,MAAM,CAAC;IACvB,SAAS,EAAE,gBAAgB,EAAE,CAAC;CAC/B;AAyHD,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,YAAY,GAAG,IAAI,CA6ErE;AAED,wBAAsB,kBAAkB,CAAC,EACvC,QAAQ,EACR,KAAqB,EACrB,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAe,EACf,MAAc,EACd,WAAe,EACf,SAAS,EACT,YAAuB,EACvB,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,eAAsB,EACtB,gBAAwB,EACxB,UAAkB,EAClB,IAAW,EACX,SAAS,EACT,OAAe,EACf,MAAM,EACN,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,EACZ,MAAM,EACN,OAAe,EACf,cAAc,EACd,oBAAoB,GACrB,EAAE,yBAAyB,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,qBAAqB,GAAG,iBAAiB,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CA4IrH;AAED,wBAAsB,WAAW,CAAC,EAAE,MAAU,EAAE,gBAAgB,EAAE,GAAG,MAAM,EAAE,EAAE,kBAAkB;;;;GAyBhG"}