@m4trix/evals 0.25.1 → 0.26.0

package/dist/cli.js

@@ -3,16 +3,16 @@ import { withFullScreen, useScreenSize } from 'fullscreen-ink';
 import React, { useState, useRef, useReducer, useEffect, useMemo } from 'react';
 import { useApp, useInput, Box, Text } from 'ink';
 import { jsx, jsxs, Fragment } from 'react/jsx-runtime';
-import { resolve, relative, join, dirname } from 'path';
-import { LineGraph } from '@pppp606/ink-chart';
+import { Schema, Effect, PubSub, Queue, Ref, Fiber, Either, ParseResult } from 'effect';
 import { randomUUID } from 'crypto';
-import { Effect, PubSub, Queue, Ref, Fiber } from 'effect';
+import { readdir, readFile, mkdir, appendFile } from 'fs/promises';
+import { resolve, join, relative, dirname } from 'path';
 import { existsSync } from 'fs';
 import * as jitiModule from 'jiti';
-import { readdir, readFile, mkdir, appendFile } from 'fs/promises';
 import { pathToFileURL } from 'url';
 import { diffLines } from 'diff';
 import stringify from 'fast-json-stable-stringify';
+import { LineGraph } from '@pppp606/ink-chart';
 
 var SEP = "  ";
 var ARROW = "\u203A";
@@ -237,6 +237,50 @@ function isPrintableCharacter(input) {
 function isBackKey(key) {
   return key.backspace || key.delete;
 }
+var ENTITY_ID_PATTERN = /^[a-zA-Z0-9_-]+$/;
+function makeEntityIdSchema(brand, label) {
+  return Schema.String.pipe(
+    Schema.trimmed(),
+    Schema.minLength(1, {
+      message: () => `${label} must be non-empty.`
+    }),
+    Schema.pattern(ENTITY_ID_PATTERN, {
+      message: () => `${label} may only contain letters, digits, underscores, and hyphens (no spaces). Examples: "my-nightly", "my_nightly", "myNightly".`
+    }),
+    Schema.brand(brand)
+  );
+}
+var RunConfigNameSchema = makeEntityIdSchema("RunConfigName", "RunConfig name");
+makeEntityIdSchema("EvaluatorName", "Evaluator name");
+makeEntityIdSchema("TestCaseName", "Test case name");
+function validateWithSchema(schema, raw, context) {
+  const trimmed = raw.trim();
+  const decode = Schema.decodeUnknownEither(
+    schema
+  );
+  const result = decode(trimmed);
+  if (Either.isLeft(result)) {
+    throw new Error(`${context}: ${ParseResult.TreeFormatter.formatErrorSync(result.left)}`);
+  }
+  return result.right;
+}
+function validateRunConfigName(raw, context) {
+  return validateWithSchema(RunConfigNameSchema, raw, context);
+}
+
+// src/evals/evaluator.ts
+function getEvaluatorDisplayLabel(evaluator) {
+  if (typeof evaluator.getDisplayLabel === "function") {
+    const label = evaluator.getDisplayLabel();
+    if (label !== void 0) {
+      return label;
+    }
+  }
+  return typeof evaluator.getName === "function" ? evaluator.getName() : void 0;
+}
+function getEvaluatorTagList(evaluator) {
+  return typeof evaluator.getTags === "function" ? [...evaluator.getTags()] : [];
+}
 
 // src/cli/data.mock.json
 var data_mock_default = {
@@ -492,7 +536,7 @@ function toEvalDataset(item, snapshots) {
 function toEvaluatorOption(item) {
   return {
     id: item.id,
-    name: item.evaluator.getName() ?? toSlug(item.id),
+    name: getEvaluatorDisplayLabel(item.evaluator) ?? toSlug(item.id),
     configPreview: `Source: ${item.filePath}`
   };
 }
@@ -735,6 +779,159 @@ function reduceCliState(state, action) {
   }
   return state;
 }
+async function loadRunSnapshotsFromArtifacts(config) {
+  const baseDir = resolve(config.artifactDirectory);
+  let entries;
+  try {
+    entries = await readdir(baseDir);
+  } catch {
+    return [];
+  }
+  const jsonlFiles = entries.filter((name) => name.endsWith(".jsonl"));
+  const snapshots = [];
+  for (const fileName of jsonlFiles) {
+    const filePath = join(baseDir, fileName);
+    try {
+      const snapshot = await parseArtifactToSnapshot(filePath, config);
+      if (snapshot) {
+        snapshots.push(snapshot);
+      }
+    } catch {
+    }
+  }
+  return snapshots.sort((a, b) => b.queuedAt - a.queuedAt);
+}
+async function parseArtifactToSnapshot(filePath, _config) {
+  const content = await readFile(filePath, "utf8");
+  const lines = content.split("\n").filter((line) => line.trim().length > 0);
+  if (lines.length === 0) {
+    return null;
+  }
+  let runQueued = null;
+  let runCompleted = null;
+  let runFailed = null;
+  let runStarted = null;
+  for (const line of lines) {
+    try {
+      const event = JSON.parse(line);
+      const type = event.type;
+      if (type === "RunQueued") {
+        runQueued = {
+          runId: event.runId,
+          datasetId: event.datasetId,
+          datasetName: event.datasetName,
+          evaluatorIds: event.evaluatorIds,
+          totalTestCases: event.totalTestCases ?? 0,
+          artifactPath: event.artifactPath ?? filePath,
+          ts: event.ts
+        };
+      }
+      if (type === "RunStarted") {
+        runStarted = { startedAt: event.startedAt };
+      }
+      if (type === "RunCompleted") {
+        runCompleted = {
+          passedTestCases: event.passedTestCases,
+          failedTestCases: event.failedTestCases,
+          totalTestCases: event.totalTestCases,
+          finishedAt: event.finishedAt
+        };
+      }
+      if (type === "RunFailed") {
+        runFailed = {
+          finishedAt: event.finishedAt,
+          errorMessage: event.errorMessage
+        };
+      }
+    } catch {
+    }
+  }
+  if (!runQueued) {
+    return null;
+  }
+  const artifactPath = filePath;
+  const status = runFailed ? "failed" : runCompleted ? "completed" : runStarted ? "running" : "queued";
+  const progress = aggregateTestCaseProgress(lines);
+  const completedTestCases = runCompleted ? runQueued.totalTestCases : progress.completedTestCases;
+  const passedTestCases = runCompleted?.passedTestCases ?? progress.passedTestCases;
+  const failedTestCases = runCompleted?.failedTestCases ?? progress.failedTestCases;
+  return {
+    runId: runQueued.runId,
+    datasetId: runQueued.datasetId,
+    datasetName: runQueued.datasetName,
+    evaluatorIds: runQueued.evaluatorIds,
+    queuedAt: runQueued.ts ?? 0,
+    startedAt: runStarted?.startedAt,
+    finishedAt: runCompleted?.finishedAt ?? runFailed?.finishedAt,
+    totalTestCases: runQueued.totalTestCases,
+    completedTestCases,
+    passedTestCases,
+    failedTestCases,
+    status,
+    artifactPath,
+    errorMessage: runFailed?.errorMessage
+  };
+}
+function aggregateTestCaseProgress(lines) {
+  let completedTestCases = 0;
+  const testCasePassedBy = /* @__PURE__ */ new Map();
+  for (const line of lines) {
+    try {
+      const event = JSON.parse(line);
+      if (event.type === "TestCaseProgress") {
+        const ev = event;
+        completedTestCases = ev.completedTestCases ?? completedTestCases;
+        const id = ev.testCaseId;
+        const current = testCasePassedBy.get(id);
+        testCasePassedBy.set(id, current === void 0 ? ev.passed : current && ev.passed);
+      }
+    } catch {
+    }
+  }
+  let passedTestCases = 0;
+  let failedTestCases = 0;
+  for (const passed of testCasePassedBy.values()) {
+    if (passed) {
+      passedTestCases += 1;
+    } else {
+      failedTestCases += 1;
+    }
+  }
+  return { completedTestCases, passedTestCases, failedTestCases };
+}
+async function parseArtifactFile(artifactPath) {
+  try {
+    const content = await readFile(artifactPath, "utf8");
+    const lines = content.split("\n").filter((line) => line.trim().length > 0);
+    const results = [];
+    for (const line of lines) {
+      try {
+        const event = JSON.parse(line);
+        if (event.type === "TestCaseProgress") {
+          const ev = event;
+          const repetitionIndex = ev.repetitionIndex ?? ev.rerunIndex;
+          const repetitionCount = ev.repetitionCount ?? ev.rerunTotal;
+          results.push({
+            testCaseId: ev.testCaseId,
+            testCaseName: ev.testCaseName,
+            completedTestCases: ev.completedTestCases,
+            totalTestCases: ev.totalTestCases,
+            repetitionId: ev.repetitionId,
+            repetitionIndex,
+            repetitionCount,
+            passed: ev.passed,
+            durationMs: ev.durationMs,
+            evaluatorScores: ev.evaluatorScores ?? []
+          });
+        }
+      } catch {
+      }
+    }
+    return results;
+  } catch {
+    return [];
+  }
+}
 
 // src/runner/config.ts
 var defaultRunnerConfig = {
@@ -742,6 +939,7 @@ var defaultRunnerConfig = {
     rootDir: process.cwd(),
     datasetSuffixes: [".dataset.ts", ".dataset.tsx", ".dataset.js", ".dataset.mjs"],
     evaluatorSuffixes: [".evaluator.ts", ".evaluator.tsx", ".evaluator.js", ".evaluator.mjs"],
+    runConfigSuffixes: [".run-config.ts", ".run-config.tsx", ".run-config.js", ".run-config.mjs"],
     testCaseSuffixes: [".test-case.ts", ".test-case.tsx", ".test-case.js", ".test-case.mjs"],
     excludeDirectories: ["node_modules", "dist", ".next", ".git", ".pnpm-store"]
   },
@@ -767,6 +965,11 @@ function toRunnerConfigOverrides(config) {
   } else if (rawDiscovery?.evaluatorSuffixes !== void 0) {
     discovery.evaluatorSuffixes = rawDiscovery.evaluatorSuffixes;
   }
+  if (rawDiscovery?.runConfigFilePatterns !== void 0) {
+    discovery.runConfigSuffixes = rawDiscovery.runConfigFilePatterns;
+  } else if (rawDiscovery?.runConfigSuffixes !== void 0) {
+    discovery.runConfigSuffixes = rawDiscovery.runConfigSuffixes;
+  }
   if (rawDiscovery?.testCaseFilePatterns !== void 0) {
     discovery.testCaseSuffixes = rawDiscovery.testCaseFilePatterns;
   } else if (rawDiscovery?.testCaseSuffixes !== void 0) {
@@ -865,6 +1068,9 @@ function isDatasetLike(value) {
 function isEvaluatorLike(value) {
   return hasMethod(value, "getName") && hasMethod(value, "resolveContext") && hasMethod(value, "getEvaluateFn");
 }
+function isRunConfigLike(value) {
+  return hasMethod(value, "getName") && hasMethod(value, "getRuns") && typeof value.getRuns === "function";
+}
 function isTestCaseLike(value) {
   return hasMethod(value, "getName") && hasMethod(value, "getTags") && hasMethod(value, "getInput");
 }
@@ -953,6 +1159,23 @@ async function collectEvaluatorsFromFiles(config) {
   );
   return found.flat();
 }
+async function collectRunConfigsFromFiles(config) {
+  const files = await walkDirectory(config.rootDir, config.excludeDirectories);
+  const matched = files.filter((filePath) => hasOneSuffix(filePath, config.runConfigSuffixes));
+  const found = await Promise.all(
+    matched.map(async (absolutePath) => {
+      const exports = await loadModuleExports(absolutePath);
+      const runConfigs = exports.filter(isRunConfigLike);
+      const relPath = relative(config.rootDir, absolutePath);
+      return runConfigs.map((runConfig) => ({
+        id: runConfig.getName(),
+        filePath: relPath,
+        runConfig
+      }));
+    })
+  );
+  return found.flat();
+}
 async function collectTestCasesFromFiles(config) {
   const files = await walkDirectory(config.rootDir, config.excludeDirectories);
   const matched = files.filter((filePath) => hasOneSuffix(filePath, config.testCaseSuffixes));
@@ -1106,6 +1329,17 @@ function getDiffLines(entry) {
   });
 }
 
+// src/evals/test-case.ts
+function getTestCaseDisplayLabel(testCase) {
+  if (typeof testCase.getDisplayLabel === "function") {
+    return testCase.getDisplayLabel();
+  }
+  return typeof testCase.getName === "function" ? testCase.getName() : "";
+}
+function getTestCaseTagList(testCase) {
+  return typeof testCase.getTags === "function" ? [...testCase.getTags()] : [];
+}
+
 // src/evals/metric.ts
 var registry = /* @__PURE__ */ new Map();
 var Metric = {
@@ -1129,6 +1363,54 @@ function getMetricById(id) {
   return registry.get(id);
 }
 
+// src/evals/aggregators.ts
+function aggregateTokenCountSum(values) {
+  const initial = {
+    input: 0,
+    output: 0,
+    inputCached: 0,
+    outputCached: 0
+  };
+  return values.reduce(
+    (acc, v) => ({
+      input: acc.input + (v.input ?? 0),
+      output: acc.output + (v.output ?? 0),
+      inputCached: acc.inputCached + (v.inputCached ?? 0),
+      outputCached: acc.outputCached + (v.outputCached ?? 0)
+    }),
+    initial
+  );
+}
+function aggregateLatencyAverage(values) {
+  if (values.length === 0) {
+    return { ms: 0 };
+  }
+  const sum = values.reduce((s, v) => s + v.ms, 0);
+  return { ms: sum / values.length };
+}
+
+// src/evals/metrics/standard.ts
+Metric.of({
+  id: "token-count",
+  name: "Tokens",
+  aggregate: aggregateTokenCountSum,
+  format: (data, options) => {
+    const input = data.input ?? 0;
+    const output = data.output ?? 0;
+    const inputCached = data.inputCached ?? 0;
+    const outputCached = data.outputCached ?? 0;
+    const cached = inputCached + outputCached;
+    const base = `in:${input} out:${output} cached:${cached}`;
+    return options?.isAggregated ? `Total: ${base}` : base;
+  }
+});
+Metric.of({
+  id: "latency",
+  name: "Latency",
+  aggregate: aggregateLatencyAverage,
+  format: (data, options) => options?.isAggregated ? `Avg: ${data.ms}ms` : `${data.ms}ms`
+});
+
 // src/evals/score.ts
 var registry2 = /* @__PURE__ */ new Map();
 function formatScoreData(def, data, options) {
@@ -1237,54 +1519,6 @@ function getScoreById(id) {
   return registry2.get(id);
 }
 
-// src/evals/aggregators.ts
-function aggregateTokenCountSum(values) {
-  const initial = {
-    input: 0,
-    output: 0,
-    inputCached: 0,
-    outputCached: 0
-  };
-  return values.reduce(
-    (acc, v) => ({
-      input: acc.input + (v.input ?? 0),
-      output: acc.output + (v.output ?? 0),
-      inputCached: acc.inputCached + (v.inputCached ?? 0),
-      outputCached: acc.outputCached + (v.outputCached ?? 0)
-    }),
-    initial
-  );
-}
-function aggregateLatencyAverage(values) {
-  if (values.length === 0) {
-    return { ms: 0 };
-  }
-  const sum = values.reduce((s, v) => s + v.ms, 0);
-  return { ms: sum / values.length };
-}
-
-// src/evals/metrics/standard.ts
-Metric.of({
-  id: "token-count",
-  name: "Tokens",
-  aggregate: aggregateTokenCountSum,
-  format: (data, options) => {
-    const input = data.input ?? 0;
-    const output = data.output ?? 0;
-    const inputCached = data.inputCached ?? 0;
-    const outputCached = data.outputCached ?? 0;
-    const cached = inputCached + outputCached;
-    const base = `in:${input} out:${output} cached:${cached}`;
-    return options?.isAggregated ? `Total: ${base}` : base;
-  }
-});
-Metric.of({
-  id: "latency",
-  name: "Latency",
-  aggregate: aggregateLatencyAverage,
-  format: (data, options) => options?.isAggregated ? `Avg: ${data.ms}ms` : `${data.ms}ms`
-});
-
 // src/evals/scores/standard.ts
 Score.of({
   id: "percent",
@@ -1391,15 +1625,17 @@ function readOutput(testCase) {
   }
   return candidate.getOutput();
 }
-function buildEvaluationUnits(testCases) {
+function buildEvaluationUnits(testCases, repetitionCount) {
+  const count = Math.max(1, repetitionCount);
   const units = [];
   for (const testCaseItem of testCases) {
-    const rerunTotal = typeof testCaseItem.testCase.getReruns === "function" ? testCaseItem.testCase.getReruns() : 1;
-    for (let r = 0; r < rerunTotal; r++) {
+    const repetitionId = `rep-${randomUUID()}`;
+    for (let r = 0; r < count; r++) {
       units.push({
         testCaseItem,
-        rerunIndex: r + 1,
-        rerunTotal
+        repetitionId,
+        repetitionIndex: r + 1,
+        repetitionCount: count
       });
     }
   }
@@ -1412,7 +1648,7 @@ function createArtifactPath(artifactDirectory, datasetId, runId) {
   return join(artifactDirectory, `${datasetId}_${runId}_${nowIsoForFile()}.jsonl`);
 }
 function processOneEvaluation(task, unit, totalEvaluations, publishEvent, persistenceQueue, updateSnapshot, startedRef, completedRef, passedRef, failedRef, testCaseResultsRef) {
-  const { testCaseItem, rerunIndex, rerunTotal } = unit;
+  const { testCaseItem, repetitionId, repetitionIndex, repetitionCount } = unit;
   return Effect.gen(function* () {
     const evaluatorRunId = `run-${randomUUID()}`;
     const started = Date.now();
@@ -1421,11 +1657,12 @@ function processOneEvaluation(task, unit, totalEvaluations, publishEvent, persis
       type: "TestCaseStarted",
       runId: task.runId,
       testCaseId: testCaseItem.id,
-      testCaseName: testCaseItem.testCase.getName(),
+      testCaseName: getTestCaseDisplayLabel(testCaseItem.testCase),
       startedTestCases: startedEvaluations,
       totalTestCases: totalEvaluations,
-      rerunIndex,
-      rerunTotal
+      repetitionId,
+      repetitionIndex,
+      repetitionCount
     });
     const evaluatorScores = [];
     let testCaseError;
@@ -1459,8 +1696,15 @@ function processOneEvaluation(task, unit, totalEvaluations, publishEvent, persis
               meta: {
                 triggerId: task.triggerId,
                 runId: evaluatorRunId,
-                datasetId: task.datasetId
+                datasetId: task.datasetId,
+                repetitionId,
+                repetitionIndex,
+                repetitionCount,
+                runConfigName: task.runConfigName
               },
+              testCaseTags: getTestCaseTagList(testCaseItem.testCase),
+              runConfigTags: task.runConfigTags,
+              evaluatorTags: getEvaluatorTagList(evaluator),
               logDiff,
               log,
               createError
@@ -1503,18 +1747,19 @@ function processOneEvaluation(task, unit, totalEvaluations, publishEvent, persis
         });
       }
     }
-    const rerunPassedThis = evaluatorScores.every((s) => s.passed);
+    const repetitionPassedThis = evaluatorScores.every((s) => s.passed);
     const completedEvaluations = yield* Ref.modify(completedRef, (n) => [n + 1, n + 1]);
     const progressEvent = {
       type: "TestCaseProgress",
       runId: task.runId,
       testCaseId: testCaseItem.id,
-      testCaseName: testCaseItem.testCase.getName(),
+      testCaseName: getTestCaseDisplayLabel(testCaseItem.testCase),
       completedTestCases: completedEvaluations,
       totalTestCases: totalEvaluations,
-      rerunIndex,
-      rerunTotal,
-      passed: rerunPassedThis,
+      repetitionId,
+      repetitionIndex,
+      repetitionCount,
+      passed: repetitionPassedThis,
       durationMs: Date.now() - started,
       evaluatorScores,
       output,
@@ -1535,9 +1780,9 @@ function processOneEvaluation(task, unit, totalEvaluations, publishEvent, persis
       (map) => {
         const key = testCaseItem.id;
         const existing = map.get(key) ?? { completedCount: 0, results: [] };
-        const newResults = [...existing.results, rerunPassedThis];
+        const newResults = [...existing.results, repetitionPassedThis];
         const newCompletedCount = existing.completedCount + 1;
-        const isLast = newCompletedCount === rerunTotal;
+        const isLast = newCompletedCount === repetitionCount;
         const newMap = new Map(map);
         newMap.set(key, {
           completedCount: newCompletedCount,
@@ -1574,10 +1819,7 @@ var executeRunTask = (task, publishEvent, persistenceQueue, updateSnapshot) => E
     runId: task.runId,
     startedAt
   });
-  const totalEvaluations = task.testCases.reduce(
-    (sum, tc) => sum + (typeof tc.testCase.getReruns === "function" ? tc.testCase.getReruns() : 1),
-    0
-  );
+  const totalEvaluations = task.testCases.length * Math.max(1, task.repetitions);
   const maxConcurrency = Math.max(1, task.maxConcurrency ?? 1);
   const completedRef = yield* Ref.make(0);
   const startedRef = yield* Ref.make(0);
@@ -1586,7 +1828,7 @@ var executeRunTask = (task, publishEvent, persistenceQueue, updateSnapshot) => E
   const testCaseResultsRef = yield* Ref.make(
     /* @__PURE__ */ new Map()
   );
-  const evaluationUnits = buildEvaluationUnits(task.testCases);
+  const evaluationUnits = buildEvaluationUnits(task.testCases, task.repetitions);
   const processEvaluation = (unit) => processOneEvaluation(
     task,
     unit,
@@ -1600,11 +1842,20 @@ var executeRunTask = (task, publishEvent, persistenceQueue, updateSnapshot) => E
     failedRef,
     testCaseResultsRef
   );
-  yield* Effect.forEach(
-    evaluationUnits,
-    processEvaluation,
-    maxConcurrency > 1 ? { concurrency: maxConcurrency } : void 0
-  );
+  const globalSem = task.globalEvaluationSemaphore;
+  if (globalSem !== void 0) {
+    yield* Effect.forEach(
+      evaluationUnits,
+      (unit) => globalSem.withPermits(1)(processEvaluation(unit)),
+      { concurrency: "unbounded", discard: true }
+    );
+  } else {
+    yield* Effect.forEach(
+      evaluationUnits,
+      processEvaluation,
+      maxConcurrency > 1 ? { concurrency: maxConcurrency } : void 0
+    );
+  }
   const [completedEvaluations, passedUniqueTestCases, failedUniqueTestCases] = yield* Effect.all([
     Ref.get(completedRef),
     Ref.get(passedRef),
@@ -1640,155 +1891,34 @@ var executeRunTask = (task, publishEvent, persistenceQueue, updateSnapshot) => E
     artifactPath: task.snapshot.artifactPath
   });
 });
-async function loadRunSnapshotsFromArtifacts(config) {
-  const baseDir = resolve(config.artifactDirectory);
-  let entries;
-  try {
-    entries = await readdir(baseDir);
-  } catch {
-    return [];
-  }
-  const jsonlFiles = entries.filter((name) => name.endsWith(".jsonl"));
-  const snapshots = [];
-  for (const fileName of jsonlFiles) {
-    const filePath = join(baseDir, fileName);
-    try {
-      const snapshot = await parseArtifactToSnapshot(filePath, config);
-      if (snapshot) {
-        snapshots.push(snapshot);
-      }
-    } catch {
-    }
-  }
-  return snapshots.sort((a, b) => b.queuedAt - a.queuedAt);
-}
-async function parseArtifactToSnapshot(filePath, _config) {
-  const content = await readFile(filePath, "utf8");
-  const lines = content.split("\n").filter((line) => line.trim().length > 0);
-  if (lines.length === 0) {
-    return null;
-  }
-  let runQueued = null;
-  let runCompleted = null;
-  let runFailed = null;
-  let runStarted = null;
-  for (const line of lines) {
-    try {
-      const event = JSON.parse(line);
-      const type = event.type;
-      if (type === "RunQueued") {
-        runQueued = {
-          runId: event.runId,
-          datasetId: event.datasetId,
-          datasetName: event.datasetName,
-          evaluatorIds: event.evaluatorIds,
-          totalTestCases: event.totalTestCases ?? 0,
-          artifactPath: event.artifactPath ?? filePath,
-          ts: event.ts
-        };
-      }
-      if (type === "RunStarted") {
-        runStarted = { startedAt: event.startedAt };
-      }
-      if (type === "RunCompleted") {
-        runCompleted = {
-          passedTestCases: event.passedTestCases,
-          failedTestCases: event.failedTestCases,
-          totalTestCases: event.totalTestCases,
-          finishedAt: event.finishedAt
-        };
-      }
-      if (type === "RunFailed") {
-        runFailed = {
-          finishedAt: event.finishedAt,
-          errorMessage: event.errorMessage
-        };
-      }
-    } catch {
-    }
+
+// src/runner/name-pattern.ts
+function parseRegexLiteral(pattern) {
+  if (!pattern.startsWith("/")) {
+    return void 0;
   }
-  if (!runQueued) {
-    return null;
+  const lastSlash = pattern.lastIndexOf("/");
+  if (lastSlash <= 0) {
+    return void 0;
   }
-  const artifactPath = filePath;
-  const status = runFailed ? "failed" : runCompleted ? "completed" : runStarted ? "running" : "queued";
-  const progress = aggregateTestCaseProgress(lines);
-  const completedTestCases = runCompleted ? runQueued.totalTestCases : progress.completedTestCases;
-  const passedTestCases = runCompleted?.passedTestCases ?? progress.passedTestCases;
-  const failedTestCases = runCompleted?.failedTestCases ?? progress.failedTestCases;
   return {
-    runId: runQueued.runId,
-    datasetId: runQueued.datasetId,
-    datasetName: runQueued.datasetName,
-    evaluatorIds: runQueued.evaluatorIds,
-    queuedAt: runQueued.ts ?? 0,
-    startedAt: runStarted?.startedAt,
-    finishedAt: runCompleted?.finishedAt ?? runFailed?.finishedAt,
-    totalTestCases: runQueued.totalTestCases,
-    completedTestCases,
-    passedTestCases,
-    failedTestCases,
-    status,
-    artifactPath,
-    errorMessage: runFailed?.errorMessage
+    source: pattern.slice(1, lastSlash),
+    flags: pattern.slice(lastSlash + 1)
   };
 }
-function aggregateTestCaseProgress(lines) {
-  let completedTestCases = 0;
-  const testCasePassedBy = /* @__PURE__ */ new Map();
-  for (const line of lines) {
-    try {
-      const event = JSON.parse(line);
-      if (event.type === "TestCaseProgress") {
-        const ev = event;
-        completedTestCases = ev.completedTestCases ?? completedTestCases;
-        const id = ev.testCaseId;
-        const current = testCasePassedBy.get(id);
-        testCasePassedBy.set(id, current === void 0 ? ev.passed : current && ev.passed);
-      }
-    } catch {
-    }
-  }
-  let passedTestCases = 0;
-  let failedTestCases = 0;
-  for (const passed of testCasePassedBy.values()) {
-    if (passed) {
-      passedTestCases += 1;
-    } else {
-      failedTestCases += 1;
-    }
+function createNameMatcher(pattern) {
+  const normalizedPattern = pattern.trim();
+  const regexLiteral = parseRegexLiteral(normalizedPattern);
+  if (regexLiteral) {
+    const regex = new RegExp(regexLiteral.source, regexLiteral.flags);
+    return (value) => regex.test(value);
   }
-  return { completedTestCases, passedTestCases, failedTestCases };
-}
-async function parseArtifactFile(artifactPath) {
-  try {
-    const content = await readFile(artifactPath, "utf8");
-    const lines = content.split("\n").filter((line) => line.trim().length > 0);
-    const results = [];
-    for (const line of lines) {
-      try {
-        const event = JSON.parse(line);
-        if (event.type === "TestCaseProgress") {
-          const ev = event;
-          results.push({
-            testCaseId: ev.testCaseId,
-            testCaseName: ev.testCaseName,
-            completedTestCases: ev.completedTestCases,
-            totalTestCases: ev.totalTestCases,
-            rerunIndex: ev.rerunIndex,
-            rerunTotal: ev.rerunTotal,
-            passed: ev.passed,
-            durationMs: ev.durationMs,
-            evaluatorScores: ev.evaluatorScores ?? []
-          });
-        }
-      } catch {
-      }
-    }
-    return results;
-  } catch {
-    return [];
+  if (normalizedPattern.includes("*")) {
+    const escaped = normalizedPattern.replace(/[.+^${}()|[\]\\]/g, "\\$&").replace(/\*/g, ".*");
+    const regex = new RegExp(`^${escaped}$`, "i");
+    return (value) => regex.test(value);
   }
+  return (value) => value.toLowerCase() === normalizedPattern.toLowerCase();
 }
 async function appendJsonLine(artifactPath, payload) {
   await mkdir(dirname(artifactPath), { recursive: true });
@@ -1847,32 +1977,12 @@ function searchCollectedTestCases(all, query) {
 }
 
 // src/runner/api.ts
-function parseRegexLiteral(pattern) {
-  if (!pattern.startsWith("/")) {
-    return void 0;
-  }
-  const lastSlash = pattern.lastIndexOf("/");
-  if (lastSlash <= 0) {
-    return void 0;
+function normalizeRunRepetitions(value) {
+  const n = value ?? 1;
+  if (!Number.isInteger(n) || n < 1) {
+    throw new Error(`repetitions must be a positive integer, got ${String(value)}`);
   }
-  return {
-    source: pattern.slice(1, lastSlash),
-    flags: pattern.slice(lastSlash + 1)
-  };
-}
-function createNameMatcher(pattern) {
-  const normalizedPattern = pattern.trim();
-  const regexLiteral = parseRegexLiteral(normalizedPattern);
-  if (regexLiteral) {
-    const regex = new RegExp(regexLiteral.source, regexLiteral.flags);
-    return (value) => regex.test(value);
-  }
-  if (normalizedPattern.includes("*")) {
-    const escaped = normalizedPattern.replace(/[.+^${}()|[\]\\]/g, "\\$&").replace(/\*/g, ".*");
-    const regex = new RegExp(`^${escaped}$`, "i");
-    return (value) => regex.test(value);
-  }
-  return (value) => value.toLowerCase() === normalizedPattern.toLowerCase();
+  return n;
 }
 function mergeRunnerOverrides(base, next) {
   if (!base) {
@@ -1907,6 +2017,7 @@ var EffectRunner = class {
     this.listeners = /* @__PURE__ */ new Set();
     this.datasetsById = /* @__PURE__ */ new Map();
     this.evaluatorsById = /* @__PURE__ */ new Map();
+    this.runConfigsById = /* @__PURE__ */ new Map();
     this.schedulerFiber = Effect.runFork(this.createSchedulerEffect());
     this.persistenceFiber = Effect.runFork(
       createPersistenceWorker(this.persistenceQueue)
@@ -1947,6 +2058,137 @@ var EffectRunner = class {
       (item) => matcher(item.evaluator.getName() ?? "")
     );
   }
+  async collectRunConfigs() {
+    const runConfigs = await collectRunConfigsFromFiles(this.config.discovery);
+    this.runConfigsById.clear();
+    const byNameLower = /* @__PURE__ */ new Map();
+    for (const item of runConfigs) {
+      const id = item.runConfig.getName();
+      const lower = id.toLowerCase();
+      const prev = byNameLower.get(lower);
+      if (prev !== void 0 && prev.filePath !== item.filePath) {
+        throw new Error(
+          `Duplicate RunConfig name "${id}" (matches "${prev.runConfig.getName()}" case-insensitively): ${prev.filePath} and ${item.filePath}`
+        );
+      }
+      byNameLower.set(lower, item);
+      this.runConfigsById.set(id, item);
+    }
+    return runConfigs;
+  }
+  async resolveRunConfigByName(name) {
+    if (this.runConfigsById.size === 0) {
+      await this.collectRunConfigs();
+    }
+    const key = validateRunConfigName(name, `RunConfig "${name.trim()}"`);
+    const keyLower = key.toLowerCase();
+    const matches = Array.from(this.runConfigsById.values()).filter(
+      (item) => item.runConfig.getName().toLowerCase() === keyLower
+    );
+    if (matches.length === 0) {
+      return void 0;
+    }
+    if (matches.length > 1) {
+      throw new Error(
+        `Multiple RunConfigs named "${name}": ${matches.map((m) => m.filePath).join(", ")}`
+      );
+    }
+    return matches[0];
+  }
+  async expandRunConfigToJobs(collected) {
+    if (this.datasetsById.size === 0) {
+      await this.collectDatasets();
+    }
+    if (this.evaluatorsById.size === 0) {
+      await this.collectEvaluators();
+    }
+    const rcName = collected.runConfig.getName();
+    const jobs = [];
+    const runs = collected.runConfig.getRuns();
+    for (const [i, row] of runs.entries()) {
+      const dsCollected = Array.from(this.datasetsById.values()).find(
+        (d) => d.dataset === row.dataset
+      );
+      if (!dsCollected) {
+        throw new Error(
+          `RunConfig "${rcName}" run[${i}]: dataset "${row.dataset.getName()}" was not found among discovered dataset exports (import the same module instances the scanner loads).`
+        );
+      }
+      let evaluatorIds;
+      if ("evaluatorPattern" in row && typeof row.evaluatorPattern === "string") {
+        const matcher = createNameMatcher(row.evaluatorPattern);
+        const matched = Array.from(this.evaluatorsById.values()).filter(
+          (item) => matcher(item.evaluator.getName() ?? "")
+        );
+        if (matched.length === 0) {
+          throw new Error(
+            `RunConfig "${rcName}" run[${i}]: no evaluator matched pattern "${row.evaluatorPattern}"`
+          );
+        }
+        evaluatorIds = matched.map((item) => item.id);
+      } else {
+        const evaluators = row.evaluators;
+        evaluatorIds = [];
+        for (const ev of evaluators) {
+          const found = Array.from(this.evaluatorsById.values()).find(
+            (item) => item.evaluator === ev
+          );
+          if (!found) {
+            throw new Error(
+              `RunConfig "${rcName}" run[${i}]: evaluator "${getEvaluatorDisplayLabel(ev) ?? "unknown"}" was not found among discovered evaluator exports`
+            );
+          }
+          evaluatorIds.push(found.id);
+        }
+      }
+      const repetitions = "repetitions" in row && row.repetitions !== void 0 ? row.repetitions : 1;
+      jobs.push({
+        datasetId: dsCollected.id,
+        evaluatorIds,
+        runConfigName: rcName,
+        runConfigDisplayLabel: collected.runConfig.getDisplayLabel(),
+        runConfigTags: collected.runConfig.getTags(),
+        repetitions
+      });
+    }
+    return jobs;
+  }
+  async expandRunConfigNamesToJobs(names) {
+    const jobs = [];
+    for (const name of names) {
+      const collected = await this.resolveRunConfigByName(name);
+      if (!collected) {
+        const known = await this.collectRunConfigs();
+        const available = known.map((r) => r.runConfig.getName()).sort();
+        throw new Error(
+          available.length > 0 ? `RunConfig "${name}" not found. Available RunConfigs: ${available.join(", ")}` : `RunConfig "${name}" not found and no RunConfigs were discovered.`
+        );
+      }
+      jobs.push(...await this.expandRunConfigToJobs(collected));
+    }
+    return jobs;
+  }
+  async runDatasetJobsWithSharedConcurrency(request) {
+    const globalConcurrency = Math.max(1, request.globalConcurrency);
+    const sem = Effect.unsafeMakeSemaphore(globalConcurrency);
+    const triggerId = request.triggerId ?? `trg-${randomUUID()}`;
+    const snapshots = [];
+    for (const job of request.jobs) {
+      snapshots.push(
+        await this.startDatasetRun({
+          datasetId: job.datasetId,
+          evaluatorIds: job.evaluatorIds,
+          triggerId,
+          maxConcurrency: this.config.maxConcurrency ?? 1,
+          globalEvaluationSemaphore: sem,
+          runConfigName: job.runConfigName,
+          runConfigTags: job.runConfigTags,
+          repetitions: job.repetitions
+        })
+      );
+    }
+    return snapshots;
+  }
   async searchTestCases(query) {
     const testCases = await collectTestCasesFromFiles(this.config.discovery);
     return searchCollectedTestCases(testCases, query);
@@ -1965,35 +2207,45 @@ var EffectRunner = class {
     );
   }
   async runDatasetWith(request) {
+    const runConfigName = validateRunConfigName(
+      request.runConfigName,
+      "runDatasetWith.runConfigName"
+    );
+    return this.startDatasetRun({
+      datasetId: request.datasetId,
+      evaluatorIds: request.evaluatorIds,
+      triggerId: request.triggerId,
+      maxConcurrency: request.concurrency ?? this.config.maxConcurrency ?? 1,
+      repetitions: request.repetitions,
+      runConfigName,
+      runConfigTags: request.runConfigTags
+    });
+  }
+  async startDatasetRun(params) {
     if (this.datasetsById.size === 0) {
       await this.collectDatasets();
     }
     if (this.evaluatorsById.size === 0) {
       await this.collectEvaluators();
     }
-    const dataset = this.datasetsById.get(request.datasetId);
+    const dataset = this.datasetsById.get(params.datasetId);
     if (!dataset) {
-      throw new Error(`Unknown dataset: ${request.datasetId}`);
+      throw new Error(`Unknown dataset: ${params.datasetId}`);
     }
-    const selectedEvaluators = request.evaluatorIds.map((id) => this.evaluatorsById.get(id)).filter((value) => Boolean(value)).map((value) => ({ id: value.id, evaluator: value.evaluator }));
+    const selectedEvaluators = params.evaluatorIds.map((id) => this.evaluatorsById.get(id)).filter((value) => Boolean(value)).map((value) => ({ id: value.id, evaluator: value.evaluator }));
     if (selectedEvaluators.length === 0) {
       throw new Error("No evaluators selected for run");
     }
-    const selectedTestCases = await this.collectDatasetTestCases(request.datasetId);
-    const totalEvaluations = selectedTestCases.reduce(
-      (sum, tc) => sum + (typeof tc.testCase.getReruns === "function" ? tc.testCase.getReruns() : 1),
-      0
-    );
-    const triggerId = request.triggerId ?? `trg-${randomUUID()}`;
+    const selectedTestCases = await this.collectDatasetTestCases(params.datasetId);
+    const repetitions = normalizeRunRepetitions(params.repetitions);
+    const totalEvaluations = selectedTestCases.length * repetitions;
+    const runConfigTags = [...params.runConfigTags ?? []];
+    const triggerId = params.triggerId ?? `trg-${randomUUID()}`;
     const runId = `run-${randomUUID()}`;
-    const artifactPath = createArtifactPath(
-      this.config.artifactDirectory,
-      request.datasetId,
-      runId
-    );
+    const artifactPath = createArtifactPath(this.config.artifactDirectory, params.datasetId, runId);
     const snapshot = {
       runId,
-      datasetId: request.datasetId,
+      datasetId: params.datasetId,
       datasetName: dataset.dataset.getName(),
       evaluatorIds: selectedEvaluators.map((item) => item.id),
       queuedAt: Date.now(),
@@ -2014,7 +2266,7 @@ var EffectRunner = class {
     const queuedEvent = {
       type: "RunQueued",
       runId,
-      datasetId: request.datasetId,
+      datasetId: params.datasetId,
       datasetName: dataset.dataset.getName(),
       evaluatorIds: selectedEvaluators.map((item) => item.id),
       totalTestCases: totalEvaluations,
@@ -2028,17 +2280,20 @@ var EffectRunner = class {
         payload: queuedEvent
       })
     );
-    const maxConcurrency = request.concurrency ?? this.config.maxConcurrency ?? 1;
     await Effect.runPromise(
       Queue.offer(this.runQueue, {
         runId,
         triggerId,
-        datasetId: request.datasetId,
+        datasetId: params.datasetId,
         dataset: dataset.dataset,
         evaluators: selectedEvaluators,
         testCases: selectedTestCases,
         snapshot,
-        maxConcurrency
+        maxConcurrency: params.maxConcurrency,
+        globalEvaluationSemaphore: params.globalEvaluationSemaphore,
+        runConfigName: params.runConfigName,
+        runConfigTags,
+        repetitions
       })
     );
     return snapshot;
@@ -2109,6 +2364,11 @@ var EffectRunner = class {
     );
   }
 };
+
+// src/runner/events.ts
+var PROGRAMMATIC_RUN_CONFIG = {
+  runConfigName: "programmatic"
+};
 var LEFT_PANE_WIDTH2 = 44;
 var MAX_RUNS_FOR_CHART = 12;
 var MAX_RUNS_FOR_TREND = 20;
@@ -2456,7 +2716,7 @@ function buildDetailRows(run, testCases, evaluatorNameById) {
     rows.push(/* @__PURE__ */ jsx(Text, { children: " " }, "sp6"));
     rows.push(/* @__PURE__ */ jsx(SectionHeader, { children: "Test cases" }, "tc-h"));
     for (const tc of testCases) {
-      const rerunPart = tc.rerunTotal != null && tc.rerunIndex != null ? ` (${tc.rerunIndex}/${tc.rerunTotal})` : "";
+      const repetitionPart = tc.repetitionCount != null && tc.repetitionCount > 1 && tc.repetitionIndex != null ? ` (${tc.repetitionIndex}/${tc.repetitionCount})` : "";
       rows.push(
         /* @__PURE__ */ jsxs(Text, { children: [
           /* @__PURE__ */ jsxs(Text, { color: "cyan", children: [
@@ -2468,13 +2728,13 @@ function buildDetailRows(run, testCases, evaluatorNameById) {
           ] }),
           " ",
           tc.testCaseName,
-          rerunPart ? /* @__PURE__ */ jsx(Text, { color: "cyan", children: rerunPart }) : null,
+          repetitionPart ? /* @__PURE__ */ jsx(Text, { color: "cyan", children: repetitionPart }) : null,
           /* @__PURE__ */ jsxs(Text, { color: "gray", children: [
             " (",
             tc.durationMs,
             "ms)"
           ] })
-        ] }, `tc-${tc.testCaseId}-${tc.rerunIndex ?? 0}`)
+        ] }, `tc-${tc.testCaseId}-${tc.repetitionId ?? "x"}-${tc.repetitionIndex ?? 0}`)
       );
       for (const item of tc.evaluatorScores) {
         const name = evaluatorNameById.get(item.evaluatorId) ?? item.evaluatorId;
@@ -2800,7 +3060,8 @@ function EvalsCliApp({ data, args, runner }) {
       }
       void runner.runDatasetWith({
         datasetId: selectedDataset.id,
-        evaluatorIds: clampedState.selectedEvaluatorIds
+        evaluatorIds: clampedState.selectedEvaluatorIds,
+        ...PROGRAMMATIC_RUN_CONFIG
       }).then((snapshot) => {
         setRuntimeMessage(
           `Started ${snapshot.runId} on ${selectedDataset.name} (${snapshot.totalTestCases} cases).`