@axlsdk/eval 0.13.4 → 0.13.5

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/dist/cli.cjs CHANGED
@@ -65,6 +65,7 @@ function evalCompare(baseline, candidate) {
65
65
  const delta = round(cScore - bScore);
66
66
  if (delta < -0.1)
67
67
  regressions.push({
68
+ itemIndex: i,
68
69
  input: bItem.input,
69
70
  scorer: name,
70
71
  baselineScore: bScore,
@@ -73,6 +74,7 @@ function evalCompare(baseline, candidate) {
73
74
  });
74
75
  else if (delta > 0.1)
75
76
  improvements.push({
77
+ itemIndex: i,
76
78
  input: bItem.input,
77
79
  scorer: name,
78
80
  baselineScore: bScore,
@@ -81,6 +83,25 @@ function evalCompare(baseline, candidate) {
81
83
  });
82
84
  }
83
85
  }
86
+ let timing;
87
+ if (baseline.summary.timing && candidate.summary.timing) {
88
+ const bMean = baseline.summary.timing.mean;
89
+ const cMean = candidate.summary.timing.mean;
90
+ const delta = round(cMean - bMean);
91
+ const deltaPercent = bMean > 0 ? round(delta / bMean * 100) : 0;
92
+ timing = { baselineMean: bMean, candidateMean: cMean, delta, deltaPercent };
93
+ }
94
+ let cost;
95
+ if (baseline.totalCost > 0 || candidate.totalCost > 0) {
96
+ const delta = round(candidate.totalCost - baseline.totalCost);
97
+ const deltaPercent = baseline.totalCost > 0 ? round(delta / baseline.totalCost * 100) : 0;
98
+ cost = {
99
+ baselineTotal: baseline.totalCost,
100
+ candidateTotal: candidate.totalCost,
101
+ delta,
102
+ deltaPercent
103
+ };
104
+ }
84
105
  const parts = [];
85
106
  for (const name of baselineScorerNames) {
86
107
  const s = scorers[name];
@@ -92,17 +113,32 @@ function evalCompare(baseline, candidate) {
92
113
  );
93
114
  }
94
115
  }
116
+ if (timing && Math.abs(timing.deltaPercent) > 1) {
117
+ const dir = timing.delta > 0 ? "slower" : "faster";
118
+ parts.push(`${Math.abs(timing.deltaPercent).toFixed(0)}% ${dir}`);
119
+ }
120
+ if (cost && Math.abs(cost.deltaPercent) > 1) {
121
+ const dir = cost.delta > 0 ? "more expensive" : "cheaper";
122
+ parts.push(`${Math.abs(cost.deltaPercent).toFixed(0)}% ${dir}`);
123
+ }
95
124
  const summaryStr = `candidate ${parts.length > 0 ? parts.join(", ") : "no significant changes"} with ${regressions.length} regressions and ${improvements.length} improvements`;
96
125
  return {
97
126
  baseline: { id: baseline.id, metadata: baseline.metadata },
98
127
  candidate: { id: candidate.id, metadata: candidate.metadata },
99
128
  scorers,
129
+ timing,
130
+ cost,
100
131
  regressions,
101
132
  improvements,
102
133
  summary: summaryStr
103
134
  };
104
135
  }
105
136
 
137
+ // src/scorer.ts
138
+ function normalizeScorerResult(result) {
139
+ return typeof result === "number" ? { score: result } : result;
140
+ }
141
+
106
142
  // src/runner.ts
107
143
  var import_node_crypto = require("crypto");
108
144
  function computeStats(scores) {
@@ -139,18 +175,18 @@ async function runEval(config, executeWorkflow, runtime) {
139
175
  return runtime.resolveProvider(uri);
140
176
  }
141
177
  };
142
- const evalItems = [];
178
+ const evalItems = new Array(items.length);
143
179
  let totalCost = 0;
144
180
  let budgetExceeded = false;
145
- async function processItem(item) {
181
+ async function processItem(item, itemIndex) {
146
182
  if (budgetExceeded) {
147
- evalItems.push({
183
+ evalItems[itemIndex] = {
148
184
  input: item.input,
149
185
  annotations: item.annotations,
150
186
  output: null,
151
187
  error: "Budget exceeded",
152
188
  scores: {}
153
- });
189
+ };
154
190
  return;
155
191
  }
156
192
  const evalItem = {
@@ -159,64 +195,88 @@ async function runEval(config, executeWorkflow, runtime) {
159
195
  output: null,
160
196
  scores: {}
161
197
  };
198
+ const itemStart = Date.now();
162
199
  try {
163
200
  const result = await executeWorkflow(item.input, runtime);
201
+ evalItem.duration = Date.now() - itemStart;
164
202
  evalItem.output = result.output;
203
+ evalItem.cost = result.cost;
165
204
  if (result.cost != null) {
166
205
  totalCost += result.cost;
167
206
  }
168
207
  } catch (err) {
208
+ evalItem.duration = Date.now() - itemStart;
169
209
  evalItem.error = err instanceof Error ? err.message : String(err);
170
- evalItems.push(evalItem);
210
+ evalItems[itemIndex] = evalItem;
171
211
  return;
172
212
  }
173
213
  if (budgetLimit != null && totalCost > budgetLimit) {
174
214
  budgetExceeded = true;
175
215
  }
216
+ evalItem.scoreDetails = {};
217
+ let itemScorerCost = 0;
176
218
  for (const scorer of config.scorers) {
219
+ const scorerStart = Date.now();
177
220
  try {
178
- const score = await scorer.score(
221
+ const raw = await scorer.score(
179
222
  evalItem.output,
180
223
  item.input,
181
224
  item.annotations,
182
225
  scorerContext
183
226
  );
184
- if (scorer.isLlm) {
185
- const lastCost = scorer._lastCost;
186
- if (lastCost != null) {
187
- totalCost += lastCost;
188
- }
227
+ const scorerResult = normalizeScorerResult(raw);
228
+ if (scorerResult.cost != null) {
229
+ itemScorerCost += scorerResult.cost;
230
+ totalCost += scorerResult.cost;
189
231
  }
190
- if (!Number.isFinite(score) || score < 0 || score > 1) {
232
+ const scorerDuration = Date.now() - scorerStart;
233
+ if (!Number.isFinite(scorerResult.score) || scorerResult.score < 0 || scorerResult.score > 1) {
191
234
  if (!evalItem.scorerErrors) evalItem.scorerErrors = [];
192
235
  evalItem.scorerErrors.push(
193
- `Scorer "${scorer.name}" returned out-of-range score ${score} for input ${JSON.stringify(item.input)}`
236
+ `Scorer "${scorer.name}" returned out-of-range score ${scorerResult.score} for input ${JSON.stringify(item.input)}`
194
237
  );
195
238
  evalItem.scores[scorer.name] = null;
239
+ evalItem.scoreDetails[scorer.name] = {
240
+ score: null,
241
+ metadata: scorerResult.metadata,
242
+ duration: scorerDuration,
243
+ cost: scorerResult.cost
244
+ };
196
245
  } else {
197
- evalItem.scores[scorer.name] = round2(score);
246
+ evalItem.scores[scorer.name] = round2(scorerResult.score);
247
+ evalItem.scoreDetails[scorer.name] = {
248
+ score: round2(scorerResult.score),
249
+ metadata: scorerResult.metadata,
250
+ duration: scorerDuration,
251
+ cost: scorerResult.cost
252
+ };
198
253
  }
199
254
  } catch (err) {
200
- if (scorer.isLlm) {
201
- const lastCost = scorer._lastCost;
202
- if (lastCost != null) {
203
- totalCost += lastCost;
204
- }
255
+ const errCost = typeof err?.cost === "number" ? err.cost : void 0;
256
+ if (errCost != null) {
257
+ itemScorerCost += errCost;
258
+ totalCost += errCost;
205
259
  }
206
260
  if (!evalItem.scorerErrors) evalItem.scorerErrors = [];
207
261
  evalItem.scorerErrors.push(
208
262
  `Scorer "${scorer.name}" threw: ${err instanceof Error ? err.message : String(err)}`
209
263
  );
210
264
  evalItem.scores[scorer.name] = null;
265
+ evalItem.scoreDetails[scorer.name] = {
266
+ score: null,
267
+ duration: Date.now() - scorerStart,
268
+ cost: errCost
269
+ };
211
270
  }
212
271
  }
213
- evalItems.push(evalItem);
272
+ evalItem.scorerCost = itemScorerCost > 0 ? itemScorerCost : void 0;
273
+ evalItems[itemIndex] = evalItem;
214
274
  }
215
275
  let index = 0;
216
276
  async function runNext() {
217
277
  while (index < items.length) {
218
278
  const currentIndex = index++;
219
- await processItem(items[currentIndex]);
279
+ await processItem(items[currentIndex], currentIndex);
220
280
  }
221
281
  }
222
282
  const workers = Array.from({ length: Math.min(concurrency, items.length) }, () => runNext());
@@ -228,6 +288,8 @@ async function runEval(config, executeWorkflow, runtime) {
228
288
  const scores = evalItems.filter((i) => !i.error && i.scores[name] != null).map((i) => i.scores[name]);
229
289
  scorerStats[name] = computeStats(scores);
230
290
  }
291
+ const durations = evalItems.filter((i) => !i.error && i.duration != null).map((i) => i.duration);
292
+ const timing = durations.length > 0 ? computeStats(durations) : void 0;
231
293
  return {
232
294
  id,
233
295
  workflow: config.workflow,
@@ -237,7 +299,7 @@ async function runEval(config, executeWorkflow, runtime) {
237
299
  totalCost,
238
300
  duration: Date.now() - startTime,
239
301
  items: evalItems,
240
- summary: { count: items.length, failures, scorers: scorerStats }
302
+ summary: { count: items.length, failures, scorers: scorerStats, timing }
241
303
  };
242
304
  }
243
305
 
@@ -359,6 +421,21 @@ Compare: baseline (${comparison.baseline.id.slice(0, 8)}) -> candidate (${compar
359
421
  ` ${name.padEnd(maxNameLen)} ${s.baselineMean.toFixed(3).padStart(8)} ${s.candidateMean.toFixed(3).padStart(9)} ${(sign + s.delta.toFixed(3)).padStart(8)} ${(sign + s.deltaPercent.toFixed(1) + "%").padStart(6)}`
360
422
  );
361
423
  }
424
+ if (comparison.timing) {
425
+ const t = comparison.timing;
426
+ const sign = t.delta > 0 ? "+" : "";
427
+ console.log(
428
+ `
429
+ Timing: baseline ${(t.baselineMean / 1e3).toFixed(2)}s -> candidate ${(t.candidateMean / 1e3).toFixed(2)}s (${sign}${t.deltaPercent.toFixed(1)}%)`
430
+ );
431
+ }
432
+ if (comparison.cost) {
433
+ const c = comparison.cost;
434
+ const sign = c.delta > 0 ? "+" : "";
435
+ console.log(
436
+ ` Cost: baseline $${c.baselineTotal.toFixed(2)} -> candidate $${c.candidateTotal.toFixed(2)} (${sign}${c.deltaPercent.toFixed(1)}%)`
437
+ );
438
+ }
362
439
  const stable = Math.max(
363
440
  0,
364
441
  baseline.items.length - comparison.regressions.length - comparison.improvements.length
@@ -409,6 +486,12 @@ function formatTable(result) {
409
486
  );
410
487
  }
411
488
  }
489
+ if (result.summary.timing) {
490
+ const t = result.summary.timing;
491
+ lines.push(
492
+ ` ${"Timing".padEnd(maxNameLen)} ${(t.mean / 1e3).toFixed(2).padStart(colWidth)}s ${(t.p50 / 1e3).toFixed(2).padStart(colWidth)}s ${(t.p95 / 1e3).toFixed(2).padStart(colWidth)}s`
493
+ );
494
+ }
412
495
  const durationSec = (result.duration / 1e3).toFixed(1);
413
496
  const costStr = result.totalCost > 0 ? `$${result.totalCost.toFixed(2)}` : "$0.00";
414
497
  lines.push("");
package/dist/cli.cjs.map CHANGED
@@ -1 +1 @@
1
- {"version":3,"sources":["../src/cli.ts","../src/compare.ts","../src/runner.ts","../src/cli-utils.ts"],"sourcesContent":["#!/usr/bin/env node\n\nimport { readdirSync, statSync } from 'node:fs';\nimport { readFile as readFileAsync, writeFile as writeFileAsync, mkdir } from 'node:fs/promises';\nimport * as path from 'node:path';\nimport type { AxlRuntime } from '@axlsdk/axl';\nimport { evalCompare } from './compare.js';\nimport { runEval } from './runner.js';\nimport type { EvalConfig, EvalResult } from './types.js';\nimport {\n findConfig,\n resolveRuntime,\n importModule,\n registerConditions,\n CONFIG_CANDIDATES,\n} from './cli-utils.js';\n\nconst KNOWN_FLAGS = new Set(['--output', '--config', '--conditions', '--fail-on-regression']);\n\nasync function main() {\n const args = process.argv.slice(2);\n\n if (args.length === 0 || args[0] === '--help' || args[0] === '-h') {\n console.log(`\nUsage:\n axl-eval <path> Run eval file(s)\n axl-eval <path> --output <file> Save results to JSON\n axl-eval <path> --config <file> Use config file for runtime\n axl-eval <path> --conditions <list> Node.js import conditions (comma-separated)\n axl-eval compare <a> <b> Compare two eval result files\n axl-eval compare <a> <b> --fail-on-regression Exit 1 if regressions\n\nConfig auto-detection (when --config is not specified):\n ${CONFIG_CANDIDATES.join(' -> ')}\n\nWhen a config is found, the exported AxlRuntime is passed to executeWorkflow\nand cost is tracked automatically via runtime.trackCost().\nWhen no config is found, a bare AxlRuntime is created (providers from env vars).\n`);\n process.exit(0);\n }\n\n if (args[0] === 'compare') {\n await runCompare(args.slice(1));\n return;\n }\n\n await runEvalCommand(args);\n}\n\nasync function runCompare(args: string[]) {\n const failOnRegression = args.includes('--fail-on-regression');\n const files = args.filter((a) => !a.startsWith('--'));\n\n if (files.length !== 2) {\n console.error('Usage: axl-eval compare <baseline.json> <candidate.json>');\n process.exit(1);\n }\n\n const baseline: EvalResult = JSON.parse(await readFileAsync(files[0], 'utf-8'));\n const candidate: EvalResult = JSON.parse(await readFileAsync(files[1], 'utf-8'));\n const comparison = evalCompare(baseline, candidate);\n\n console.log(\n `\\nCompare: baseline (${comparison.baseline.id.slice(0, 8)}) -> candidate (${comparison.candidate.id.slice(0, 8)})\\n`,\n );\n\n const scorerNames = Object.keys(comparison.scorers);\n const maxNameLen = Math.max(...scorerNames.map((n) => n.length), 6);\n\n console.log(` ${'Scorer'.padEnd(maxNameLen)} Baseline Candidate Delta Change`);\n console.log(` ${''.padEnd(maxNameLen, '-')} -------- --------- -------- ------`);\n\n for (const name of scorerNames) {\n const s = comparison.scorers[name];\n const sign = s.delta > 0 ? '+' : '';\n console.log(\n ` ${name.padEnd(maxNameLen)} ${s.baselineMean.toFixed(3).padStart(8)} ${s.candidateMean.toFixed(3).padStart(9)} ${(sign + s.delta.toFixed(3)).padStart(8)} ${(sign + s.deltaPercent.toFixed(1) + '%').padStart(6)}`,\n );\n }\n\n const stable = Math.max(\n 0,\n baseline.items.length - comparison.regressions.length - comparison.improvements.length,\n );\n console.log(\n `\\n Regressions: ${comparison.regressions.length} | Improvements: ${comparison.improvements.length} | Stable: ${stable}\\n`,\n );\n\n if (failOnRegression && comparison.regressions.length > 0) {\n process.exit(1);\n }\n}\n\nfunction collectEvalFiles(p: string): string[] {\n const resolved = path.resolve(p);\n try {\n const stat = statSync(resolved);\n if (stat.isDirectory()) {\n const entries = readdirSync(resolved);\n return entries\n .filter((e) => /\\.eval\\.[mc]?[jt]sx?$/.test(e))\n .map((e) => path.join(resolved, e));\n }\n return [resolved];\n } catch {\n return [resolved];\n }\n}\n\nfunction formatTable(result: EvalResult): string {\n const lines: string[] = [];\n const scorerNames = Object.keys(result.summary.scorers);\n const maxNameLen = Math.max(...scorerNames.map((n) => n.length), 'Scorer'.length);\n const colWidth = 8;\n\n lines.push(`Eval: ${result.workflow} x ${result.dataset} (${result.summary.count} items)`);\n lines.push(\n ` ${'Scorer'.padEnd(maxNameLen)} ${'Mean'.padStart(colWidth)} ${'Min'.padStart(colWidth)} ${'Max'.padStart(colWidth)} ${'p50'.padStart(colWidth)} ${'p95'.padStart(colWidth)}`,\n );\n\n const ruleLen = maxNameLen + 2 + (colWidth + 2) * 5;\n lines.push(' ' + '\\u2500'.repeat(ruleLen));\n\n for (const name of scorerNames) {\n const s = result.summary.scorers[name];\n // Detect scorers with no valid scores (all items errored → computeStats([]) → all zeros)\n const validScoreCount = result.items.filter((i) => !i.error && i.scores[name] != null).length;\n if (validScoreCount === 0) {\n lines.push(\n ` ${name.padEnd(maxNameLen)} ${'--'.padStart(colWidth)} ${'--'.padStart(colWidth)} ${'--'.padStart(colWidth)} ${'--'.padStart(colWidth)} ${'--'.padStart(colWidth)}`,\n );\n } else {\n lines.push(\n ` ${name.padEnd(maxNameLen)} ${s.mean.toFixed(2).padStart(colWidth)} ${s.min.toFixed(2).padStart(colWidth)} ${s.max.toFixed(2).padStart(colWidth)} ${s.p50.toFixed(2).padStart(colWidth)} ${s.p95.toFixed(2).padStart(colWidth)}`,\n );\n }\n }\n\n const durationSec = (result.duration / 1000).toFixed(1);\n const costStr = result.totalCost > 0 ? `$${result.totalCost.toFixed(2)}` : '$0.00';\n lines.push('');\n lines.push(\n ` Failures: ${result.summary.failures}/${result.summary.count} | Cost: ${costStr} | Duration: ${durationSec}s`,\n );\n\n const itemsWithErrors = result.items.filter((i) => i.scorerErrors?.length);\n if (itemsWithErrors.length > 0) {\n const uniqueErrors = [...new Set(itemsWithErrors.flatMap((i) => i.scorerErrors!))];\n lines.push('');\n lines.push(\n ` Scorer errors (${itemsWithErrors.length}/${result.summary.count} items affected):`,\n );\n for (const err of uniqueErrors.slice(0, 5)) {\n lines.push(` - ${err}`);\n }\n if (uniqueErrors.length > 5) {\n lines.push(` ... and ${uniqueErrors.length - 5} more`);\n }\n }\n\n return lines.join('\\n');\n}\n\n// ── Runtime resolution ─────────────────────────────────────────────\n\nasync function resolveRuntimeFromConfig(configPath: string): Promise<AxlRuntime> {\n try {\n const mod = await importModule(configPath, import.meta.url);\n const runtime = resolveRuntime(mod) as AxlRuntime;\n\n if (!runtime || typeof runtime.execute !== 'function') {\n console.error(`Config must export a default AxlRuntime instance.`);\n if (runtime) {\n const keys = Object.keys(runtime as object)\n .slice(0, 5)\n .join(', ');\n console.error(` Got: ${typeof runtime}${keys ? ` with keys: { ${keys} }` : ''}`);\n }\n console.error(\n `Example:\\n import { AxlRuntime } from '@axlsdk/axl';\\n export default new AxlRuntime({ ... });`,\n );\n process.exit(1);\n }\n\n return runtime;\n } catch (err) {\n const msg = err instanceof Error ? err.message : String(err);\n if (\n /Cannot use import statement|Unexpected reserved word|top-level await|exports is not defined/.test(\n msg,\n )\n ) {\n const ext = path.extname(configPath);\n console.error(`[axl-eval] Config failed to load due to a CJS/ESM compatibility issue.`);\n if (ext === '.ts' || ext === '.tsx') {\n console.error(\n ` Tip: try renaming to .mts to force ESM format, or ensure tsx is installed and up to date.`,\n );\n } else {\n console.error(` Tip: add \"type\": \"module\" to your package.json.`);\n }\n console.error();\n }\n console.error(`Failed to load config:`, err);\n process.exit(1);\n }\n}\n\nasync function getRuntime(configArg?: string, conditions?: string[]): Promise<AxlRuntime> {\n // Register import conditions before any config loading\n if (conditions && conditions.length > 0) {\n await registerConditions(conditions);\n }\n\n // 1. Explicit --config\n if (configArg) {\n const configPath = path.resolve(process.cwd(), configArg);\n const stat = statSync(configPath, { throwIfNoEntry: false });\n if (!stat?.isFile()) {\n console.error(`Config file not found: ${configPath}`);\n process.exit(1);\n }\n console.error(`[axl-eval] Loading config from ${configPath}`);\n return resolveRuntimeFromConfig(configPath);\n }\n\n // 2. Auto-detect\n const found = findConfig(process.cwd());\n if (found) {\n console.error(`[axl-eval] Auto-detected config: ${found}`);\n return resolveRuntimeFromConfig(found);\n }\n\n // 3. Bare runtime (providers from env vars)\n const { AxlRuntime } = await import('@axlsdk/axl');\n return new AxlRuntime();\n}\n\n// ── Arg parsing ────────────────────────────────────────────────────\n\nfunction parseEvalArgs(args: string[]): {\n outputPath?: string;\n configArg?: string;\n conditions: string[];\n paths: string[];\n} {\n let outputPath: string | undefined;\n let configArg: string | undefined;\n let conditions: string[] = [];\n const paths: string[] = [];\n\n for (let i = 0; i < args.length; i++) {\n const arg = args[i];\n if (arg === '--output' || arg === '--config' || arg === '--conditions') {\n if (i + 1 >= args.length) {\n console.error(`Error: ${arg} requires a value`);\n process.exit(1);\n }\n const value = args[++i];\n if (arg === '--output') outputPath = value;\n else if (arg === '--config') configArg = value;\n else\n conditions = value\n .split(',')\n .map((c) => c.trim())\n .filter(Boolean);\n } else if (arg.startsWith('--')) {\n if (!KNOWN_FLAGS.has(arg)) {\n console.error(`Unknown flag: ${arg}`);\n process.exit(1);\n }\n } else {\n paths.push(arg);\n }\n }\n\n return { outputPath, configArg, conditions, paths };\n}\n\n// ── Main eval command ──────────────────────────────────────────────\n\nasync function runEvalCommand(args: string[]) {\n const { outputPath, configArg, conditions, paths } = parseEvalArgs(args);\n\n if (paths.length === 0) {\n console.error('Error: No eval file path provided');\n process.exit(1);\n }\n\n const evalFiles: string[] = [];\n for (const p of paths) {\n evalFiles.push(...collectEvalFiles(p));\n }\n\n if (evalFiles.length === 0) {\n console.error('Error: No eval files found');\n process.exit(1);\n }\n\n const runtime = await getRuntime(configArg, conditions);\n const results: EvalResult[] = [];\n\n try {\n for (const filePath of evalFiles) {\n try {\n const mod = await importModule(path.resolve(filePath), import.meta.url);\n const evalConfig: EvalConfig = mod.default?.default ?? mod.default ?? mod.config ?? mod;\n\n if (!evalConfig.workflow || !evalConfig.dataset || !evalConfig.scorers) {\n console.error(\n `Error: ${filePath} does not export a valid eval config (missing workflow, dataset, or scorers)`,\n );\n continue;\n }\n\n // Resolve executeWorkflow: custom export > registered workflow > passthrough\n let executeWorkflow: (\n input: unknown,\n rt?: unknown,\n ) => Promise<{ output: unknown; cost?: number }>;\n\n if (mod.executeWorkflow) {\n // Wrap custom executeWorkflow with trackCost for automatic cost attribution\n executeWorkflow = async (input, rt) => {\n const { result, cost: trackedCost } = await runtime.trackCost(async () => {\n return mod.executeWorkflow(input, rt);\n });\n return { output: result.output, cost: result.cost ?? trackedCost };\n };\n } else if (runtime.getWorkflow(evalConfig.workflow)) {\n // No executeWorkflow exported but workflow is registered — use runtime.execute()\n executeWorkflow = async (input) => {\n const { result, cost } = await runtime.trackCost(async () => {\n return runtime.execute(evalConfig.workflow, input);\n });\n return { output: result, cost };\n };\n } else {\n console.warn(\n `[axl-eval] Warning: ${filePath} does not export executeWorkflow — using input passthrough`,\n );\n executeWorkflow = async (input) => ({ output: input });\n }\n\n const result = await runEval(evalConfig, executeWorkflow, runtime);\n results.push(result);\n\n console.log('\\n' + formatTable(result) + '\\n');\n } catch (err) {\n console.error(\n `Error running eval ${filePath}: ${err instanceof Error ? err.message : String(err)}`,\n );\n }\n }\n\n if (outputPath && results.length > 0) {\n const output = results.length === 1 ? results[0] : results;\n const outputDir = path.dirname(path.resolve(outputPath));\n await mkdir(outputDir, { recursive: true });\n await writeFileAsync(outputPath, JSON.stringify(output, null, 2), 'utf-8');\n console.log(`Results saved to ${outputPath}`);\n }\n } finally {\n await runtime.shutdown().catch(() => {});\n }\n\n // Exit with non-zero code if no evals succeeded\n if (results.length === 0) {\n process.exit(1);\n }\n}\n\nmain().catch((err) => {\n console.error(err);\n process.exit(1);\n});\n","import type { EvalResult, EvalComparison, EvalRegression, EvalImprovement } from './types.js';\n\nfunction round(n: number): number {\n return Math.round(n * 100) / 100;\n}\n\nexport function evalCompare(baseline: EvalResult, candidate: EvalResult): EvalComparison {\n if (baseline.dataset !== candidate.dataset) {\n throw new Error(\n `Cannot compare evals from different datasets: \"${baseline.dataset}\" vs \"${candidate.dataset}\"`,\n );\n }\n\n const baselineScorerNames = Object.keys(baseline.summary.scorers).sort();\n const candidateScorerNames = Object.keys(candidate.summary.scorers).sort();\n if (JSON.stringify(baselineScorerNames) !== JSON.stringify(candidateScorerNames)) {\n throw new Error(`Cannot compare evals with different scorers`);\n }\n\n const scorers: EvalComparison['scorers'] = {};\n for (const name of baselineScorerNames) {\n const bMean = baseline.summary.scorers[name].mean;\n const cMean = candidate.summary.scorers[name].mean;\n const delta = round(cMean - bMean);\n const deltaPercent = bMean > 0 ? round((delta / bMean) * 100) : 0;\n scorers[name] = { baselineMean: bMean, candidateMean: cMean, delta, deltaPercent };\n }\n\n const regressions: EvalRegression[] = [];\n const improvements: EvalImprovement[] = [];\n\n const minLength = Math.min(baseline.items.length, candidate.items.length);\n for (let i = 0; i < minLength; i++) {\n const bItem = baseline.items[i];\n const cItem = candidate.items[i];\n if (bItem.error || cItem.error) continue;\n for (const name of baselineScorerNames) {\n const bScore = bItem.scores[name];\n const cScore = cItem.scores[name];\n if (bScore == null || cScore == null) continue;\n const delta = round(cScore - bScore);\n if (delta < -0.1)\n regressions.push({\n input: bItem.input,\n scorer: name,\n baselineScore: bScore,\n candidateScore: cScore,\n delta,\n });\n else if (delta > 0.1)\n improvements.push({\n input: bItem.input,\n scorer: name,\n baselineScore: bScore,\n candidateScore: cScore,\n delta,\n });\n }\n }\n\n const parts: string[] = [];\n for (const name of baselineScorerNames) {\n const s = scorers[name];\n if (Math.abs(s.delta) > 0.001) {\n const direction = s.delta > 0 ? 'improves' : 'regresses';\n const sign = s.delta > 0 ? '+' : '';\n parts.push(\n `'${name}' ${direction} by ${sign}${s.deltaPercent.toFixed(1)}% (${s.baselineMean.toFixed(2)} -> ${s.candidateMean.toFixed(2)})`,\n );\n }\n }\n const summaryStr = `candidate ${parts.length > 0 ? parts.join(', ') : 'no significant changes'} with ${regressions.length} regressions and ${improvements.length} improvements`;\n\n return {\n baseline: { id: baseline.id, metadata: baseline.metadata },\n candidate: { id: candidate.id, metadata: candidate.metadata },\n scorers,\n regressions,\n improvements,\n summary: summaryStr,\n };\n}\n","import type { AxlRuntime } from '@axlsdk/axl';\nimport type { EvalConfig, EvalResult, EvalItem, EvalSummary } from './types.js';\nimport type { ScorerContext } from './scorer.js';\nimport { randomUUID } from 'node:crypto';\n\nfunction computeStats(scores: number[]): {\n mean: number;\n min: number;\n max: number;\n p50: number;\n p95: number;\n} {\n if (scores.length === 0) return { mean: 0, min: 0, max: 0, p50: 0, p95: 0 };\n const sorted = [...scores].sort((a, b) => a - b);\n const mean = sorted.reduce((a, b) => a + b, 0) / sorted.length;\n const min = sorted[0];\n const max = sorted[sorted.length - 1];\n const p50 = sorted[Math.floor(sorted.length * 0.5)];\n const p95 = sorted[Math.floor(sorted.length * 0.95)];\n return { mean: round(mean), min: round(min), max: round(max), p50: round(p50), p95: round(p95) };\n}\n\nfunction round(n: number): number {\n return Math.round(n * 1000) / 1000;\n}\n\nfunction parseCost(cost: string): number {\n const match = cost.match(/^\\$?([\\d.]+)$/);\n if (!match) throw new Error(`Invalid cost format: \"${cost}\"`);\n return parseFloat(match[1]);\n}\n\nexport async function runEval(\n config: EvalConfig,\n executeWorkflow: (\n input: unknown,\n runtime: AxlRuntime,\n ) => Promise<{ output: unknown; cost?: number }>,\n runtime: AxlRuntime,\n): Promise<EvalResult> {\n const startTime = Date.now();\n const id = randomUUID();\n const items = await config.dataset.getItems();\n const concurrency = config.concurrency ?? 5;\n const budgetLimit = config.budget ? parseCost(config.budget) : undefined;\n\n // Create a scorer context that LLM scorers use to resolve providers.\n const scorerContext: ScorerContext = {\n resolveProvider: (uri: string) => {\n if (typeof runtime.resolveProvider !== 'function') {\n throw new Error(\n `LLM scorers require a runtime with resolveProvider(). ` +\n `Ensure you are using a real AxlRuntime instance, not a mock.`,\n );\n }\n return runtime.resolveProvider(uri);\n },\n };\n\n const evalItems: EvalItem[] = [];\n let totalCost = 0;\n let budgetExceeded = false;\n\n async function processItem(item: (typeof items)[0]): Promise<void> {\n if (budgetExceeded) {\n evalItems.push({\n input: item.input,\n annotations: item.annotations,\n output: null,\n error: 'Budget exceeded',\n scores: {},\n });\n return;\n }\n\n const evalItem: EvalItem = {\n input: item.input,\n annotations: item.annotations,\n output: null,\n scores: {},\n };\n try {\n const result = await executeWorkflow(item.input, runtime);\n evalItem.output = result.output;\n if (result.cost != null) {\n totalCost += result.cost;\n }\n } catch (err) {\n evalItem.error = err instanceof Error ? err.message : String(err);\n evalItems.push(evalItem);\n return;\n }\n\n if (budgetLimit != null && totalCost > budgetLimit) {\n budgetExceeded = true;\n }\n\n for (const scorer of config.scorers) {\n try {\n const score = await scorer.score(\n evalItem.output,\n item.input,\n item.annotations,\n scorerContext,\n );\n // Accumulate LLM scorer cost if available\n if (scorer.isLlm) {\n const lastCost = (scorer as unknown as { _lastCost?: number })._lastCost;\n if (lastCost != null) {\n totalCost += lastCost;\n }\n }\n if (!Number.isFinite(score) || score < 0 || score > 1) {\n if (!evalItem.scorerErrors) evalItem.scorerErrors = [];\n evalItem.scorerErrors.push(\n `Scorer \"${scorer.name}\" returned out-of-range score ${score} for input ${JSON.stringify(item.input)}`,\n );\n evalItem.scores[scorer.name] = null;\n } else {\n evalItem.scores[scorer.name] = round(score);\n }\n } catch (err) {\n // Accumulate cost even on failure — the LLM call may have succeeded\n // before a downstream error (e.g., invalid JSON, schema validation)\n if (scorer.isLlm) {\n const lastCost = (scorer as unknown as { _lastCost?: number })._lastCost;\n if (lastCost != null) {\n totalCost += lastCost;\n }\n }\n if (!evalItem.scorerErrors) evalItem.scorerErrors = [];\n evalItem.scorerErrors.push(\n `Scorer \"${scorer.name}\" threw: ${err instanceof Error ? err.message : String(err)}`,\n );\n evalItem.scores[scorer.name] = null;\n }\n }\n evalItems.push(evalItem);\n }\n\n let index = 0;\n async function runNext(): Promise<void> {\n while (index < items.length) {\n const currentIndex = index++;\n await processItem(items[currentIndex]);\n }\n }\n\n const workers = Array.from({ length: Math.min(concurrency, items.length) }, () => runNext());\n await Promise.all(workers);\n\n const failures = evalItems.filter((i) => i.error).length;\n const scorerNames = config.scorers.map((s) => s.name);\n const scorerStats: EvalSummary['scorers'] = {};\n for (const name of scorerNames) {\n const scores = evalItems\n .filter((i) => !i.error && i.scores[name] != null)\n .map((i) => i.scores[name] as number);\n scorerStats[name] = computeStats(scores);\n }\n\n return {\n id,\n workflow: config.workflow,\n dataset: config.dataset.name,\n metadata: config.metadata ?? {},\n timestamp: new Date().toISOString(),\n totalCost,\n duration: Date.now() - startTime,\n items: evalItems,\n summary: { count: items.length, failures, scorers: scorerStats },\n };\n}\n","/**\n * CLI utilities for config detection, runtime resolution, and loader registration.\n *\n * These are duplicated from @axlsdk/studio (cli-utils.ts, resolve-runtime.ts,\n * eval-loader.ts, cli.ts) because @axlsdk/eval cannot depend on @axlsdk/studio.\n * Keep in sync with the studio versions if either changes.\n */\n\nimport { resolve } from 'node:path';\nimport { existsSync } from 'node:fs';\nimport { pathToFileURL } from 'node:url';\n\n// ── Config auto-detection ──────────────────────────────────────────\n\nexport const CONFIG_CANDIDATES = [\n 'axl.config.mts',\n 'axl.config.ts',\n 'axl.config.mjs',\n 'axl.config.js',\n];\n\nexport function findConfig(cwd: string): string | undefined {\n for (const name of CONFIG_CANDIDATES) {\n const p = resolve(cwd, name);\n if (existsSync(p)) return p;\n }\n return undefined;\n}\n\n// ── Extension helpers ──────────────────────────────────────────────\n\n/** Returns true if the file is TypeScript and needs tsx to load. */\nexport function needsTsxLoader(filePath: string): boolean {\n return /\\.[mc]?tsx?$/.test(filePath);\n}\n\n// ── Runtime resolution ─────────────────────────────────────────────\n\n/**\n * Resolve the AxlRuntime from a dynamically imported config module.\n *\n * Module shapes handled:\n * - ESM `export default runtime` → mod.default is the runtime\n * - CJS compiled from `export default runtime` → mod.default.default is the runtime\n * - CJS `module.exports = runtime` → mod.default is the runtime\n * - Named `export { runtime }` → mod.runtime is the runtime\n */\nexport function resolveRuntime(mod: Record<string, any>): unknown {\n const def = mod.default as Record<string, any> | undefined;\n return def?.default ?? def ?? mod.runtime;\n}\n\n// ── Module loading ────────────────────────────────────────────────\n\n// Lazily resolved tsImport function from tsx. `undefined` = not yet checked,\n// `null` = tsx not available.\nlet tsImportFn:\n | ((specifier: string, parentURL: string) => Promise<Record<string, any>>)\n | null\n | undefined;\n\n/**\n * Import a module, using tsx's `tsImport()` for TypeScript files.\n *\n * `tsImport()` handles ESM/CJS format correctly without process-wide side effects —\n * no need for `register()` hooks or ESM-forcing workarounds. Falls back to regular\n * `import()` for non-TypeScript files or when tsx is not installed.\n */\nexport async function importModule(\n filePath: string,\n parentURL: string,\n): Promise<Record<string, any>> {\n if (needsTsxLoader(filePath)) {\n if (tsImportFn === undefined) {\n try {\n // @ts-expect-error — tsx is an optional runtime dependency\n const mod = await import('tsx/esm/api');\n tsImportFn = mod.tsImport ?? null;\n } catch {\n tsImportFn = null;\n console.warn(\n '[axl-eval] Warning: tsx is not installed. TypeScript files require tsx.\\n' +\n ' Install it with: npm install -D tsx',\n );\n }\n }\n if (tsImportFn) {\n return (await tsImportFn(pathToFileURL(filePath).href, parentURL)) as Record<string, any>;\n }\n }\n return await import(pathToFileURL(filePath).href);\n}\n\n// ── Conditions ────────────────────────────────────────────────────\n\nexport async function registerConditions(conditions: string[]): Promise<void> {\n try {\n const nodeModule = await import('node:module');\n const hookCode = [\n `const extra = ${JSON.stringify(conditions)};`,\n `export async function resolve(specifier, context, nextResolve) {`,\n ` return nextResolve(specifier, {`,\n ` ...context,`,\n ` conditions: [...new Set([...context.conditions, ...extra])],`,\n ` });`,\n `}`,\n ].join('\\n');\n nodeModule.register(`data:text/javascript,${encodeURIComponent(hookCode)}`);\n } catch {\n console.warn('[axl-eval] Warning: --conditions requires Node.js 20.6+');\n }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;AAEA,IAAAA,kBAAsC;AACtC,sBAA8E;AAC9E,WAAsB;;;ACFtB,SAAS,MAAM,GAAmB;AAChC,SAAO,KAAK,MAAM,IAAI,GAAG,IAAI;AAC/B;AAEO,SAAS,YAAY,UAAsB,WAAuC;AACvF,MAAI,SAAS,YAAY,UAAU,SAAS;AAC1C,UAAM,IAAI;AAAA,MACR,kDAAkD,SAAS,OAAO,SAAS,UAAU,OAAO;AAAA,IAC9F;AAAA,EACF;AAEA,QAAM,sBAAsB,OAAO,KAAK,SAAS,QAAQ,OAAO,EAAE,KAAK;AACvE,QAAM,uBAAuB,OAAO,KAAK,UAAU,QAAQ,OAAO,EAAE,KAAK;AACzE,MAAI,KAAK,UAAU,mBAAmB,MAAM,KAAK,UAAU,oBAAoB,GAAG;AAChF,UAAM,IAAI,MAAM,6CAA6C;AAAA,EAC/D;AAEA,QAAM,UAAqC,CAAC;AAC5C,aAAW,QAAQ,qBAAqB;AACtC,UAAM,QAAQ,SAAS,QAAQ,QAAQ,IAAI,EAAE;AAC7C,UAAM,QAAQ,UAAU,QAAQ,QAAQ,IAAI,EAAE;AAC9C,UAAM,QAAQ,MAAM,QAAQ,KAAK;AACjC,UAAM,eAAe,QAAQ,IAAI,MAAO,QAAQ,QAAS,GAAG,IAAI;AAChE,YAAQ,IAAI,IAAI,EAAE,cAAc,OAAO,eAAe,OAAO,OAAO,aAAa;AAAA,EACnF;AAEA,QAAM,cAAgC,CAAC;AACvC,QAAM,eAAkC,CAAC;AAEzC,QAAM,YAAY,KAAK,IAAI,SAAS,MAAM,QAAQ,UAAU,MAAM,MAAM;AACxE,WAAS,IAAI,GAAG,IAAI,WAAW,KAAK;AAClC,UAAM,QAAQ,SAAS,MAAM,CAAC;AAC9B,UAAM,QAAQ,UAAU,MAAM,CAAC;AAC/B,QAAI,MAAM,SAAS,MAAM,MAAO;AAChC,eAAW,QAAQ,qBAAqB;AACtC,YAAM,SAAS,MAAM,OAAO,IAAI;AAChC,YAAM,SAAS,MAAM,OAAO,IAAI;AAChC,UAAI,UAAU,QAAQ,UAAU,KAAM;AACtC,YAAM,QAAQ,MAAM,SAAS,MAAM;AACnC,UAAI,QAAQ;AACV,oBAAY,KAAK;AAAA,UACf,OAAO,MAAM;AAAA,UACb,QAAQ;AAAA,UACR,eAAe;AAAA,UACf,gBAAgB;AAAA,UAChB;AAAA,QACF,CAAC;AAAA,eACM,QAAQ;AACf,qBAAa,KAAK;AAAA,UAChB,OAAO,MAAM;AAAA,UACb,QAAQ;AAAA,UACR,eAAe;AAAA,UACf,gBAAgB;AAAA,UAChB;AAAA,QACF,CAAC;AAAA,IACL;AAAA,EACF;AAEA,QAAM,QAAkB,CAAC;AACzB,aAAW,QAAQ,qBAAqB;AACtC,UAAM,IAAI,QAAQ,IAAI;AACtB,QAAI,KAAK,IAAI,EAAE,KAAK,IAAI,MAAO;AAC7B,YAAM,YAAY,EAAE,QAAQ,IAAI,aAAa;AAC7C,YAAM,OAAO,EAAE,QAAQ,IAAI,MAAM;AACjC,YAAM;AAAA,QACJ,IAAI,IAAI,KAAK,SAAS,OAAO,IAAI,GAAG,EAAE,aAAa,QAAQ,CAAC,CAAC,MAAM,EAAE,aAAa,QAAQ,CAAC,CAAC,OAAO,EAAE,cAAc,QAAQ,CAAC,CAAC;AAAA,MAC/H;AAAA,IACF;AAAA,EACF;AACA,QAAM,aAAa,aAAa,MAAM,SAAS,IAAI,MAAM,KAAK,IAAI,IAAI,wBAAwB,SAAS,YAAY,MAAM,oBAAoB,aAAa,MAAM;AAEhK,SAAO;AAAA,IACL,UAAU,EAAE,IAAI,SAAS,IAAI,UAAU,SAAS,SAAS;AAAA,IACzD,WAAW,EAAE,IAAI,UAAU,IAAI,UAAU,UAAU,SAAS;AAAA,IAC5D;AAAA,IACA;AAAA,IACA;AAAA,IACA,SAAS;AAAA,EACX;AACF;;;AC9EA,yBAA2B;AAE3B,SAAS,aAAa,QAMpB;AACA,MAAI,OAAO,WAAW,EAAG,QAAO,EAAE,MAAM,GAAG,KAAK,GAAG,KAAK,GAAG,KAAK,GAAG,KAAK,EAAE;AAC1E,QAAM,SAAS,CAAC,GAAG,MAAM,EAAE,KAAK,CAAC,GAAG,MAAM,IAAI,CAAC;AAC/C,QAAM,OAAO,OAAO,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,OAAO;AACxD,QAAM,MAAM,OAAO,CAAC;AACpB,QAAM,MAAM,OAAO,OAAO,SAAS,CAAC;AACpC,QAAM,MAAM,OAAO,KAAK,MAAM,OAAO,SAAS,GAAG,CAAC;AAClD,QAAM,MAAM,OAAO,KAAK,MAAM,OAAO,SAAS,IAAI,CAAC;AACnD,SAAO,EAAE,MAAMC,OAAM,IAAI,GAAG,KAAKA,OAAM,GAAG,GAAG,KAAKA,OAAM,GAAG,GAAG,KAAKA,OAAM,GAAG,GAAG,KAAKA,OAAM,GAAG,EAAE;AACjG;AAEA,SAASA,OAAM,GAAmB;AAChC,SAAO,KAAK,MAAM,IAAI,GAAI,IAAI;AAChC;AAEA,SAAS,UAAU,MAAsB;AACvC,QAAM,QAAQ,KAAK,MAAM,eAAe;AACxC,MAAI,CAAC,MAAO,OAAM,IAAI,MAAM,yBAAyB,IAAI,GAAG;AAC5D,SAAO,WAAW,MAAM,CAAC,CAAC;AAC5B;AAEA,eAAsB,QACpB,QACA,iBAIA,SACqB;AACrB,QAAM,YAAY,KAAK,IAAI;AAC3B,QAAM,SAAK,+BAAW;AACtB,QAAM,QAAQ,MAAM,OAAO,QAAQ,SAAS;AAC5C,QAAM,cAAc,OAAO,eAAe;AAC1C,QAAM,cAAc,OAAO,SAAS,UAAU,OAAO,MAAM,IAAI;AAG/D,QAAM,gBAA+B;AAAA,IACnC,iBAAiB,CAAC,QAAgB;AAChC,UAAI,OAAO,QAAQ,oBAAoB,YAAY;AACjD,cAAM,IAAI;AAAA,UACR;AAAA,QAEF;AAAA,MACF;AACA,aAAO,QAAQ,gBAAgB,GAAG;AAAA,IACpC;AAAA,EACF;AAEA,QAAM,YAAwB,CAAC;AAC/B,MAAI,YAAY;AAChB,MAAI,iBAAiB;AAErB,iBAAe,YAAY,MAAwC;AACjE,QAAI,gBAAgB;AAClB,gBAAU,KAAK;AAAA,QACb,OAAO,KAAK;AAAA,QACZ,aAAa,KAAK;AAAA,QAClB,QAAQ;AAAA,QACR,OAAO;AAAA,QACP,QAAQ,CAAC;AAAA,MACX,CAAC;AACD;AAAA,IACF;AAEA,UAAM,WAAqB;AAAA,MACzB,OAAO,KAAK;AAAA,MACZ,aAAa,KAAK;AAAA,MAClB,QAAQ;AAAA,MACR,QAAQ,CAAC;AAAA,IACX;AACA,QAAI;AACF,YAAM,SAAS,MAAM,gBAAgB,KAAK,OAAO,OAAO;AACxD,eAAS,SAAS,OAAO;AACzB,UAAI,OAAO,QAAQ,MAAM;AACvB,qBAAa,OAAO;AAAA,MACtB;AAAA,IACF,SAAS,KAAK;AACZ,eAAS,QAAQ,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAChE,gBAAU,KAAK,QAAQ;AACvB;AAAA,IACF;AAEA,QAAI,eAAe,QAAQ,YAAY,aAAa;AAClD,uBAAiB;AAAA,IACnB;AAEA,eAAW,UAAU,OAAO,SAAS;AACnC,UAAI;AACF,cAAM,QAAQ,MAAM,OAAO;AAAA,UACzB,SAAS;AAAA,UACT,KAAK;AAAA,UACL,KAAK;AAAA,UACL;AAAA,QACF;AAEA,YAAI,OAAO,OAAO;AAChB,gBAAM,WAAY,OAA6C;AAC/D,cAAI,YAAY,MAAM;AACpB,yBAAa;AAAA,UACf;AAAA,QACF;AACA,YAAI,CAAC,OAAO,SAAS,KAAK,KAAK,QAAQ,KAAK,QAAQ,GAAG;AACrD,cAAI,CAAC,SAAS,aAAc,UAAS,eAAe,CAAC;AACrD,mBAAS,aAAa;AAAA,YACpB,WAAW,OAAO,IAAI,iCAAiC,KAAK,cAAc,KAAK,UAAU,KAAK,KAAK,CAAC;AAAA,UACtG;AACA,mBAAS,OAAO,OAAO,IAAI,IAAI;AAAA,QACjC,OAAO;AACL,mBAAS,OAAO,OAAO,IAAI,IAAIA,OAAM,KAAK;AAAA,QAC5C;AAAA,MACF,SAAS,KAAK;AAGZ,YAAI,OAAO,OAAO;AAChB,gBAAM,WAAY,OAA6C;AAC/D,cAAI,YAAY,MAAM;AACpB,yBAAa;AAAA,UACf;AAAA,QACF;AACA,YAAI,CAAC,SAAS,aAAc,UAAS,eAAe,CAAC;AACrD,iBAAS,aAAa;AAAA,UACpB,WAAW,OAAO,IAAI,YAAY,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,CAAC;AAAA,QACpF;AACA,iBAAS,OAAO,OAAO,IAAI,IAAI;AAAA,MACjC;AAAA,IACF;AACA,cAAU,KAAK,QAAQ;AAAA,EACzB;AAEA,MAAI,QAAQ;AACZ,iBAAe,UAAyB;AACtC,WAAO,QAAQ,MAAM,QAAQ;AAC3B,YAAM,eAAe;AACrB,YAAM,YAAY,MAAM,YAAY,CAAC;AAAA,IACvC;AAAA,EACF;AAEA,QAAM,UAAU,MAAM,KAAK,EAAE,QAAQ,KAAK,IAAI,aAAa,MAAM,MAAM,EAAE,GAAG,MAAM,QAAQ,CAAC;AAC3F,QAAM,QAAQ,IAAI,OAAO;AAEzB,QAAM,WAAW,UAAU,OAAO,CAAC,MAAM,EAAE,KAAK,EAAE;AAClD,QAAM,cAAc,OAAO,QAAQ,IAAI,CAAC,MAAM,EAAE,IAAI;AACpD,QAAM,cAAsC,CAAC;AAC7C,aAAW,QAAQ,aAAa;AAC9B,UAAM,SAAS,UACZ,OAAO,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,OAAO,IAAI,KAAK,IAAI,EAChD,IAAI,CAAC,MAAM,EAAE,OAAO,IAAI,CAAW;AACtC,gBAAY,IAAI,IAAI,aAAa,MAAM;AAAA,EACzC;AAEA,SAAO;AAAA,IACL;AAAA,IACA,UAAU,OAAO;AAAA,IACjB,SAAS,OAAO,QAAQ;AAAA,IACxB,UAAU,OAAO,YAAY,CAAC;AAAA,IAC9B,YAAW,oBAAI,KAAK,GAAE,YAAY;AAAA,IAClC;AAAA,IACA,UAAU,KAAK,IAAI,IAAI;AAAA,IACvB,OAAO;AAAA,IACP,SAAS,EAAE,OAAO,MAAM,QAAQ,UAAU,SAAS,YAAY;AAAA,EACjE;AACF;;;ACpKA,uBAAwB;AACxB,qBAA2B;AAC3B,sBAA8B;AAIvB,IAAM,oBAAoB;AAAA,EAC/B;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;AAEO,SAAS,WAAW,KAAiC;AAC1D,aAAW,QAAQ,mBAAmB;AACpC,UAAM,QAAI,0BAAQ,KAAK,IAAI;AAC3B,YAAI,2BAAW,CAAC,EAAG,QAAO;AAAA,EAC5B;AACA,SAAO;AACT;AAKO,SAAS,eAAe,UAA2B;AACxD,SAAO,eAAe,KAAK,QAAQ;AACrC;AAaO,SAAS,eAAe,KAAmC;AAChE,QAAM,MAAM,IAAI;AAChB,SAAO,KAAK,WAAW,OAAO,IAAI;AACpC;AAMA,IAAI;AAYJ,eAAsB,aACpB,UACA,WAC8B;AAC9B,MAAI,eAAe,QAAQ,GAAG;AAC5B,QAAI,eAAe,QAAW;AAC5B,UAAI;AAEF,cAAM,MAAM,MAAM,OAAO,aAAa;AACtC,qBAAa,IAAI,YAAY;AAAA,MAC/B,QAAQ;AACN,qBAAa;AACb,gBAAQ;AAAA,UACN;AAAA,QAEF;AAAA,MACF;AAAA,IACF;AACA,QAAI,YAAY;AACd,aAAQ,MAAM,eAAW,+BAAc,QAAQ,EAAE,MAAM,SAAS;AAAA,IAClE;AAAA,EACF;AACA,SAAO,MAAM,WAAO,+BAAc,QAAQ,EAAE;AAC9C;AAIA,eAAsB,mBAAmB,YAAqC;AAC5E,MAAI;AACF,UAAM,aAAa,MAAM,OAAO,QAAa;AAC7C,UAAM,WAAW;AAAA,MACf,iBAAiB,KAAK,UAAU,UAAU,CAAC;AAAA,MAC3C;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF,EAAE,KAAK,IAAI;AACX,eAAW,SAAS,wBAAwB,mBAAmB,QAAQ,CAAC,EAAE;AAAA,EAC5E,QAAQ;AACN,YAAQ,KAAK,yDAAyD;AAAA,EACxE;AACF;;;AH/GA;AAiBA,IAAM,cAAc,oBAAI,IAAI,CAAC,YAAY,YAAY,gBAAgB,sBAAsB,CAAC;AAE5F,eAAe,OAAO;AACpB,QAAM,OAAO,QAAQ,KAAK,MAAM,CAAC;AAEjC,MAAI,KAAK,WAAW,KAAK,KAAK,CAAC,MAAM,YAAY,KAAK,CAAC,MAAM,MAAM;AACjE,YAAQ,IAAI;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,IAUZ,kBAAkB,KAAK,MAAM,CAAC;AAAA;AAAA;AAAA;AAAA;AAAA,CAKjC;AACG,YAAQ,KAAK,CAAC;AAAA,EAChB;AAEA,MAAI,KAAK,CAAC,MAAM,WAAW;AACzB,UAAM,WAAW,KAAK,MAAM,CAAC,CAAC;AAC9B;AAAA,EACF;AAEA,QAAM,eAAe,IAAI;AAC3B;AAEA,eAAe,WAAW,MAAgB;AACxC,QAAM,mBAAmB,KAAK,SAAS,sBAAsB;AAC7D,QAAM,QAAQ,KAAK,OAAO,CAAC,MAAM,CAAC,EAAE,WAAW,IAAI,CAAC;AAEpD,MAAI,MAAM,WAAW,GAAG;AACtB,YAAQ,MAAM,0DAA0D;AACxE,YAAQ,KAAK,CAAC;AAAA,EAChB;AAEA,QAAM,WAAuB,KAAK,MAAM,UAAM,gBAAAC,UAAc,MAAM,CAAC,GAAG,OAAO,CAAC;AAC9E,QAAM,YAAwB,KAAK,MAAM,UAAM,gBAAAA,UAAc,MAAM,CAAC,GAAG,OAAO,CAAC;AAC/E,QAAM,aAAa,YAAY,UAAU,SAAS;AAElD,UAAQ;AAAA,IACN;AAAA,qBAAwB,WAAW,SAAS,GAAG,MAAM,GAAG,CAAC,CAAC,mBAAmB,WAAW,UAAU,GAAG,MAAM,GAAG,CAAC,CAAC;AAAA;AAAA,EAClH;AAEA,QAAM,cAAc,OAAO,KAAK,WAAW,OAAO;AAClD,QAAM,aAAa,KAAK,IAAI,GAAG,YAAY,IAAI,CAAC,MAAM,EAAE,MAAM,GAAG,CAAC;AAElE,UAAQ,IAAI,KAAK,SAAS,OAAO,UAAU,CAAC,yCAAyC;AACrF,UAAQ,IAAI,KAAK,GAAG,OAAO,YAAY,GAAG,CAAC,yCAAyC;AAEpF,aAAW,QAAQ,aAAa;AAC9B,UAAM,IAAI,WAAW,QAAQ,IAAI;AACjC,UAAM,OAAO,EAAE,QAAQ,IAAI,MAAM;AACjC,YAAQ;AAAA,MACN,KAAK,KAAK,OAAO,UAAU,CAAC,KAAK,EAAE,aAAa,QAAQ,CAAC,EAAE,SAAS,CAAC,CAAC,KAAK,EAAE,cAAc,QAAQ,CAAC,EAAE,SAAS,CAAC,CAAC,MAAM,OAAO,EAAE,MAAM,QAAQ,CAAC,GAAG,SAAS,CAAC,CAAC,MAAM,OAAO,EAAE,aAAa,QAAQ,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC;AAAA,IACxN;AAAA,EACF;AAEA,QAAM,SAAS,KAAK;AAAA,IAClB;AAAA,IACA,SAAS,MAAM,SAAS,WAAW,YAAY,SAAS,WAAW,aAAa;AAAA,EAClF;AACA,UAAQ;AAAA,IACN;AAAA,iBAAoB,WAAW,YAAY,MAAM,oBAAoB,WAAW,aAAa,MAAM,cAAc,MAAM;AAAA;AAAA,EACzH;AAEA,MAAI,oBAAoB,WAAW,YAAY,SAAS,GAAG;AACzD,YAAQ,KAAK,CAAC;AAAA,EAChB;AACF;AAEA,SAAS,iBAAiB,GAAqB;AAC7C,QAAM,WAAgB,aAAQ,CAAC;AAC/B,MAAI;AACF,UAAM,WAAO,0BAAS,QAAQ;AAC9B,QAAI,KAAK,YAAY,GAAG;AACtB,YAAM,cAAU,6BAAY,QAAQ;AACpC,aAAO,QACJ,OAAO,CAAC,MAAM,wBAAwB,KAAK,CAAC,CAAC,EAC7C,IAAI,CAAC,MAAW,UAAK,UAAU,CAAC,CAAC;AAAA,IACtC;AACA,WAAO,CAAC,QAAQ;AAAA,EAClB,QAAQ;AACN,WAAO,CAAC,QAAQ;AAAA,EAClB;AACF;AAEA,SAAS,YAAY,QAA4B;AAC/C,QAAM,QAAkB,CAAC;AACzB,QAAM,cAAc,OAAO,KAAK,OAAO,QAAQ,OAAO;AACtD,QAAM,aAAa,KAAK,IAAI,GAAG,YAAY,IAAI,CAAC,MAAM,EAAE,MAAM,GAAG,SAAS,MAAM;AAChF,QAAM,WAAW;AAEjB,QAAM,KAAK,SAAS,OAAO,QAAQ,MAAM,OAAO,OAAO,KAAK,OAAO,QAAQ,KAAK,SAAS;AACzF,QAAM;AAAA,IACJ,KAAK,SAAS,OAAO,UAAU,CAAC,KAAK,OAAO,SAAS,QAAQ,CAAC,KAAK,MAAM,SAAS,QAAQ,CAAC,KAAK,MAAM,SAAS,QAAQ,CAAC,KAAK,MAAM,SAAS,QAAQ,CAAC,KAAK,MAAM,SAAS,QAAQ,CAAC;AAAA,EACpL;AAEA,QAAM,UAAU,aAAa,KAAK,WAAW,KAAK;AAClD,QAAM,KAAK,OAAO,SAAS,OAAO,OAAO,CAAC;AAE1C,aAAW,QAAQ,aAAa;AAC9B,UAAM,IAAI,OAAO,QAAQ,QAAQ,IAAI;AAErC,UAAM,kBAAkB,OAAO,MAAM,OAAO,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,OAAO,IAAI,KAAK,IAAI,EAAE;AACvF,QAAI,oBAAoB,GAAG;AACzB,YAAM;AAAA,QACJ,KAAK,KAAK,OAAO,UAAU,CAAC,KAAK,KAAK,SAAS,QAAQ,CAAC,KAAK,KAAK,SAAS,QAAQ,CAAC,KAAK,KAAK,SAAS,QAAQ,CAAC,KAAK,KAAK,SAAS,QAAQ,CAAC,KAAK,KAAK,SAAS,QAAQ,CAAC;AAAA,MAC1K;AAAA,IACF,OAAO;AACL,YAAM;AAAA,QACJ,KAAK,KAAK,OAAO,UAAU,CAAC,KAAK,EAAE,KAAK,QAAQ,CAAC,EAAE,SAAS,QAAQ,CAAC,KAAK,EAAE,IAAI,QAAQ,CAAC,EAAE,SAAS,QAAQ,CAAC,KAAK,EAAE,IAAI,QAAQ,CAAC,EAAE,SAAS,QAAQ,CAAC,KAAK,EAAE,IAAI,QAAQ,CAAC,EAAE,SAAS,QAAQ,CAAC,KAAK,EAAE,IAAI,QAAQ,CAAC,EAAE,SAAS,QAAQ,CAAC;AAAA,MACvO;AAAA,IACF;AAAA,EACF;AAEA,QAAM,eAAe,OAAO,WAAW,KAAM,QAAQ,CAAC;AACtD,QAAM,UAAU,OAAO,YAAY,IAAI,IAAI,OAAO,UAAU,QAAQ,CAAC,CAAC,KAAK;AAC3E,QAAM,KAAK,EAAE;AACb,QAAM;AAAA,IACJ,eAAe,OAAO,QAAQ,QAAQ,IAAI,OAAO,QAAQ,KAAK,YAAY,OAAO,gBAAgB,WAAW;AAAA,EAC9G;AAEA,QAAM,kBAAkB,OAAO,MAAM,OAAO,CAAC,MAAM,EAAE,cAAc,MAAM;AACzE,MAAI,gBAAgB,SAAS,GAAG;AAC9B,UAAM,eAAe,CAAC,GAAG,IAAI,IAAI,gBAAgB,QAAQ,CAAC,MAAM,EAAE,YAAa,CAAC,CAAC;AACjF,UAAM,KAAK,EAAE;AACb,UAAM;AAAA,MACJ,oBAAoB,gBAAgB,MAAM,IAAI,OAAO,QAAQ,KAAK;AAAA,IACpE;AACA,eAAW,OAAO,aAAa,MAAM,GAAG,CAAC,GAAG;AAC1C,YAAM,KAAK,SAAS,GAAG,EAAE;AAAA,IAC3B;AACA,QAAI,aAAa,SAAS,GAAG;AAC3B,YAAM,KAAK,eAAe,aAAa,SAAS,CAAC,OAAO;AAAA,IAC1D;AAAA,EACF;AAEA,SAAO,MAAM,KAAK,IAAI;AACxB;AAIA,eAAe,yBAAyB,YAAyC;AAC/E,MAAI;AACF,UAAM,MAAM,MAAM,aAAa,YAAY,YAAY,GAAG;AAC1D,UAAM,UAAU,eAAe,GAAG;AAElC,QAAI,CAAC,WAAW,OAAO,QAAQ,YAAY,YAAY;AACrD,cAAQ,MAAM,mDAAmD;AACjE,UAAI,SAAS;AACX,cAAM,OAAO,OAAO,KAAK,OAAiB,EACvC,MAAM,GAAG,CAAC,EACV,KAAK,IAAI;AACZ,gBAAQ,MAAM,UAAU,OAAO,OAAO,GAAG,OAAO,iBAAiB,IAAI,OAAO,EAAE,EAAE;AAAA,MAClF;AACA,cAAQ;AAAA,QACN;AAAA;AAAA;AAAA,MACF;AACA,cAAQ,KAAK,CAAC;AAAA,IAChB;AAEA,WAAO;AAAA,EACT,SAAS,KAAK;AACZ,UAAM,MAAM,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAC3D,QACE,8FAA8F;AAAA,MAC5F;AAAA,IACF,GACA;AACA,YAAM,MAAW,aAAQ,UAAU;AACnC,cAAQ,MAAM,wEAAwE;AACtF,UAAI,QAAQ,SAAS,QAAQ,QAAQ;AACnC,gBAAQ;AAAA,UACN;AAAA,QACF;AAAA,MACF,OAAO;AACL,gBAAQ,MAAM,mDAAmD;AAAA,MACnE;AACA,cAAQ,MAAM;AAAA,IAChB;AACA,YAAQ,MAAM,0BAA0B,GAAG;AAC3C,YAAQ,KAAK,CAAC;AAAA,EAChB;AACF;AAEA,eAAe,WAAW,WAAoB,YAA4C;AAExF,MAAI,cAAc,WAAW,SAAS,GAAG;AACvC,UAAM,mBAAmB,UAAU;AAAA,EACrC;AAGA,MAAI,WAAW;AACb,UAAM,aAAkB,aAAQ,QAAQ,IAAI,GAAG,SAAS;AACxD,UAAM,WAAO,0BAAS,YAAY,EAAE,gBAAgB,MAAM,CAAC;AAC3D,QAAI,CAAC,MAAM,OAAO,GAAG;AACnB,cAAQ,MAAM,0BAA0B,UAAU,EAAE;AACpD,cAAQ,KAAK,CAAC;AAAA,IAChB;AACA,YAAQ,MAAM,kCAAkC,UAAU,EAAE;AAC5D,WAAO,yBAAyB,UAAU;AAAA,EAC5C;AAGA,QAAM,QAAQ,WAAW,QAAQ,IAAI,CAAC;AACtC,MAAI,OAAO;AACT,YAAQ,MAAM,oCAAoC,KAAK,EAAE;AACzD,WAAO,yBAAyB,KAAK;AAAA,EACvC;AAGA,QAAM,EAAE,WAAW,IAAI,MAAM,OAAO,aAAa;AACjD,SAAO,IAAI,WAAW;AACxB;AAIA,SAAS,cAAc,MAKrB;AACA,MAAI;AACJ,MAAI;AACJ,MAAI,aAAuB,CAAC;AAC5B,QAAM,QAAkB,CAAC;AAEzB,WAAS,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;AACpC,UAAM,MAAM,KAAK,CAAC;AAClB,QAAI,QAAQ,cAAc,QAAQ,cAAc,QAAQ,gBAAgB;AACtE,UAAI,IAAI,KAAK,KAAK,QAAQ;AACxB,gBAAQ,MAAM,UAAU,GAAG,mBAAmB;AAC9C,gBAAQ,KAAK,CAAC;AAAA,MAChB;AACA,YAAM,QAAQ,KAAK,EAAE,CAAC;AACtB,UAAI,QAAQ,WAAY,cAAa;AAAA,eAC5B,QAAQ,WAAY,aAAY;AAAA;AAEvC,qBAAa,MACV,MAAM,GAAG,EACT,IAAI,CAAC,MAAM,EAAE,KAAK,CAAC,EACnB,OAAO,OAAO;AAAA,IACrB,WAAW,IAAI,WAAW,IAAI,GAAG;AAC/B,UAAI,CAAC,YAAY,IAAI,GAAG,GAAG;AACzB,gBAAQ,MAAM,iBAAiB,GAAG,EAAE;AACpC,gBAAQ,KAAK,CAAC;AAAA,MAChB;AAAA,IACF,OAAO;AACL,YAAM,KAAK,GAAG;AAAA,IAChB;AAAA,EACF;AAEA,SAAO,EAAE,YAAY,WAAW,YAAY,MAAM;AACpD;AAIA,eAAe,eAAe,MAAgB;AAC5C,QAAM,EAAE,YAAY,WAAW,YAAY,MAAM,IAAI,cAAc,IAAI;AAEvE,MAAI,MAAM,WAAW,GAAG;AACtB,YAAQ,MAAM,mCAAmC;AACjD,YAAQ,KAAK,CAAC;AAAA,EAChB;AAEA,QAAM,YAAsB,CAAC;AAC7B,aAAW,KAAK,OAAO;AACrB,cAAU,KAAK,GAAG,iBAAiB,CAAC,CAAC;AAAA,EACvC;AAEA,MAAI,UAAU,WAAW,GAAG;AAC1B,YAAQ,MAAM,4BAA4B;AAC1C,YAAQ,KAAK,CAAC;AAAA,EAChB;AAEA,QAAM,UAAU,MAAM,WAAW,WAAW,UAAU;AACtD,QAAM,UAAwB,CAAC;AAE/B,MAAI;AACF,eAAW,YAAY,WAAW;AAChC,UAAI;AACF,cAAM,MAAM,MAAM,aAAkB,aAAQ,QAAQ,GAAG,YAAY,GAAG;AACtE,cAAM,aAAyB,IAAI,SAAS,WAAW,IAAI,WAAW,IAAI,UAAU;AAEpF,YAAI,CAAC,WAAW,YAAY,CAAC,WAAW,WAAW,CAAC,WAAW,SAAS;AACtE,kBAAQ;AAAA,YACN,UAAU,QAAQ;AAAA,UACpB;AACA;AAAA,QACF;AAGA,YAAI;AAKJ,YAAI,IAAI,iBAAiB;AAEvB,4BAAkB,OAAO,OAAO,OAAO;AACrC,kBAAM,EAAE,QAAAC,SAAQ,MAAM,YAAY,IAAI,MAAM,QAAQ,UAAU,YAAY;AACxE,qBAAO,IAAI,gBAAgB,OAAO,EAAE;AAAA,YACtC,CAAC;AACD,mBAAO,EAAE,QAAQA,QAAO,QAAQ,MAAMA,QAAO,QAAQ,YAAY;AAAA,UACnE;AAAA,QACF,WAAW,QAAQ,YAAY,WAAW,QAAQ,GAAG;AAEnD,4BAAkB,OAAO,UAAU;AACjC,kBAAM,EAAE,QAAAA,SAAQ,KAAK,IAAI,MAAM,QAAQ,UAAU,YAAY;AAC3D,qBAAO,QAAQ,QAAQ,WAAW,UAAU,KAAK;AAAA,YACnD,CAAC;AACD,mBAAO,EAAE,QAAQA,SAAQ,KAAK;AAAA,UAChC;AAAA,QACF,OAAO;AACL,kBAAQ;AAAA,YACN,uBAAuB,QAAQ;AAAA,UACjC;AACA,4BAAkB,OAAO,WAAW,EAAE,QAAQ,MAAM;AAAA,QACtD;AAEA,cAAM,SAAS,MAAM,QAAQ,YAAY,iBAAiB,OAAO;AACjE,gBAAQ,KAAK,MAAM;AAEnB,gBAAQ,IAAI,OAAO,YAAY,MAAM,IAAI,IAAI;AAAA,MAC/C,SAAS,KAAK;AACZ,gBAAQ;AAAA,UACN,sBAAsB,QAAQ,KAAK,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,CAAC;AAAA,QACrF;AAAA,MACF;AAAA,IACF;AAEA,QAAI,cAAc,QAAQ,SAAS,GAAG;AACpC,YAAM,SAAS,QAAQ,WAAW,IAAI,QAAQ,CAAC,IAAI;AACnD,YAAM,YAAiB,aAAa,aAAQ,UAAU,CAAC;AACvD,gBAAM,uBAAM,WAAW,EAAE,WAAW,KAAK,CAAC;AAC1C,gBAAM,gBAAAC,WAAe,YAAY,KAAK,UAAU,QAAQ,MAAM,CAAC,GAAG,OAAO;AACzE,cAAQ,IAAI,oBAAoB,UAAU,EAAE;AAAA,IAC9C;AAAA,EACF,UAAE;AACA,UAAM,QAAQ,SAAS,EAAE,MAAM,MAAM;AAAA,IAAC,CAAC;AAAA,EACzC;AAGA,MAAI,QAAQ,WAAW,GAAG;AACxB,YAAQ,KAAK,CAAC;AAAA,EAChB;AACF;AAEA,KAAK,EAAE,MAAM,CAAC,QAAQ;AACpB,UAAQ,MAAM,GAAG;AACjB,UAAQ,KAAK,CAAC;AAChB,CAAC;","names":["import_node_fs","round","readFileAsync","result","writeFileAsync"]}
1
+ {"version":3,"sources":["../src/cli.ts","../src/compare.ts","../src/scorer.ts","../src/runner.ts","../src/cli-utils.ts"],"sourcesContent":["#!/usr/bin/env node\n\nimport { readdirSync, statSync } from 'node:fs';\nimport { readFile as readFileAsync, writeFile as writeFileAsync, mkdir } from 'node:fs/promises';\nimport * as path from 'node:path';\nimport type { AxlRuntime } from '@axlsdk/axl';\nimport { evalCompare } from './compare.js';\nimport { runEval } from './runner.js';\nimport type { EvalConfig, EvalResult } from './types.js';\nimport {\n findConfig,\n resolveRuntime,\n importModule,\n registerConditions,\n CONFIG_CANDIDATES,\n} from './cli-utils.js';\n\nconst KNOWN_FLAGS = new Set(['--output', '--config', '--conditions', '--fail-on-regression']);\n\nasync function main() {\n const args = process.argv.slice(2);\n\n if (args.length === 0 || args[0] === '--help' || args[0] === '-h') {\n console.log(`\nUsage:\n axl-eval <path> Run eval file(s)\n axl-eval <path> --output <file> Save results to JSON\n axl-eval <path> --config <file> Use config file for runtime\n axl-eval <path> --conditions <list> Node.js import conditions (comma-separated)\n axl-eval compare <a> <b> Compare two eval result files\n axl-eval compare <a> <b> --fail-on-regression Exit 1 if regressions\n\nConfig auto-detection (when --config is not specified):\n ${CONFIG_CANDIDATES.join(' -> ')}\n\nWhen a config is found, the exported AxlRuntime is passed to executeWorkflow\nand cost is tracked automatically via runtime.trackCost().\nWhen no config is found, a bare AxlRuntime is created (providers from env vars).\n`);\n process.exit(0);\n }\n\n if (args[0] === 'compare') {\n await runCompare(args.slice(1));\n return;\n }\n\n await runEvalCommand(args);\n}\n\nasync function runCompare(args: string[]) {\n const failOnRegression = args.includes('--fail-on-regression');\n const files = args.filter((a) => !a.startsWith('--'));\n\n if (files.length !== 2) {\n console.error('Usage: axl-eval compare <baseline.json> <candidate.json>');\n process.exit(1);\n }\n\n const baseline: EvalResult = JSON.parse(await readFileAsync(files[0], 'utf-8'));\n const candidate: EvalResult = JSON.parse(await readFileAsync(files[1], 'utf-8'));\n const comparison = evalCompare(baseline, candidate);\n\n console.log(\n `\\nCompare: baseline (${comparison.baseline.id.slice(0, 8)}) -> candidate (${comparison.candidate.id.slice(0, 8)})\\n`,\n );\n\n const scorerNames = Object.keys(comparison.scorers);\n const maxNameLen = Math.max(...scorerNames.map((n) => n.length), 6);\n\n console.log(` ${'Scorer'.padEnd(maxNameLen)} Baseline Candidate Delta Change`);\n console.log(` ${''.padEnd(maxNameLen, '-')} -------- --------- -------- ------`);\n\n for (const name of scorerNames) {\n const s = comparison.scorers[name];\n const sign = s.delta > 0 ? '+' : '';\n console.log(\n ` ${name.padEnd(maxNameLen)} ${s.baselineMean.toFixed(3).padStart(8)} ${s.candidateMean.toFixed(3).padStart(9)} ${(sign + s.delta.toFixed(3)).padStart(8)} ${(sign + s.deltaPercent.toFixed(1) + '%').padStart(6)}`,\n );\n }\n\n if (comparison.timing) {\n const t = comparison.timing;\n const sign = t.delta > 0 ? '+' : '';\n console.log(\n `\\n Timing: baseline ${(t.baselineMean / 1000).toFixed(2)}s -> candidate ${(t.candidateMean / 1000).toFixed(2)}s (${sign}${t.deltaPercent.toFixed(1)}%)`,\n );\n }\n if (comparison.cost) {\n const c = comparison.cost;\n const sign = c.delta > 0 ? '+' : '';\n console.log(\n ` Cost: baseline $${c.baselineTotal.toFixed(2)} -> candidate $${c.candidateTotal.toFixed(2)} (${sign}${c.deltaPercent.toFixed(1)}%)`,\n );\n }\n\n const stable = Math.max(\n 0,\n baseline.items.length - comparison.regressions.length - comparison.improvements.length,\n );\n console.log(\n `\\n Regressions: ${comparison.regressions.length} | Improvements: ${comparison.improvements.length} | Stable: ${stable}\\n`,\n );\n\n if (failOnRegression && comparison.regressions.length > 0) {\n process.exit(1);\n }\n}\n\nfunction collectEvalFiles(p: string): string[] {\n const resolved = path.resolve(p);\n try {\n const stat = statSync(resolved);\n if (stat.isDirectory()) {\n const entries = readdirSync(resolved);\n return entries\n .filter((e) => /\\.eval\\.[mc]?[jt]sx?$/.test(e))\n .map((e) => path.join(resolved, e));\n }\n return [resolved];\n } catch {\n return [resolved];\n }\n}\n\nfunction formatTable(result: EvalResult): string {\n const lines: string[] = [];\n const scorerNames = Object.keys(result.summary.scorers);\n const maxNameLen = Math.max(...scorerNames.map((n) => n.length), 'Scorer'.length);\n const colWidth = 8;\n\n lines.push(`Eval: ${result.workflow} x ${result.dataset} (${result.summary.count} items)`);\n lines.push(\n ` ${'Scorer'.padEnd(maxNameLen)} ${'Mean'.padStart(colWidth)} ${'Min'.padStart(colWidth)} ${'Max'.padStart(colWidth)} ${'p50'.padStart(colWidth)} ${'p95'.padStart(colWidth)}`,\n );\n\n const ruleLen = maxNameLen + 2 + (colWidth + 2) * 5;\n lines.push(' ' + '\\u2500'.repeat(ruleLen));\n\n for (const name of scorerNames) {\n const s = result.summary.scorers[name];\n // Detect scorers with no valid scores (all items errored → computeStats([]) → all zeros)\n const validScoreCount = result.items.filter((i) => !i.error && i.scores[name] != null).length;\n if (validScoreCount === 0) {\n lines.push(\n ` ${name.padEnd(maxNameLen)} ${'--'.padStart(colWidth)} ${'--'.padStart(colWidth)} ${'--'.padStart(colWidth)} ${'--'.padStart(colWidth)} ${'--'.padStart(colWidth)}`,\n );\n } else {\n lines.push(\n ` ${name.padEnd(maxNameLen)} ${s.mean.toFixed(2).padStart(colWidth)} ${s.min.toFixed(2).padStart(colWidth)} ${s.max.toFixed(2).padStart(colWidth)} ${s.p50.toFixed(2).padStart(colWidth)} ${s.p95.toFixed(2).padStart(colWidth)}`,\n );\n }\n }\n\n if (result.summary.timing) {\n const t = result.summary.timing;\n lines.push(\n ` ${'Timing'.padEnd(maxNameLen)} ${(t.mean / 1000).toFixed(2).padStart(colWidth)}s ${(t.p50 / 1000).toFixed(2).padStart(colWidth)}s ${(t.p95 / 1000).toFixed(2).padStart(colWidth)}s`,\n );\n }\n\n const durationSec = (result.duration / 1000).toFixed(1);\n const costStr = result.totalCost > 0 ? `$${result.totalCost.toFixed(2)}` : '$0.00';\n lines.push('');\n lines.push(\n ` Failures: ${result.summary.failures}/${result.summary.count} | Cost: ${costStr} | Duration: ${durationSec}s`,\n );\n\n const itemsWithErrors = result.items.filter((i) => i.scorerErrors?.length);\n if (itemsWithErrors.length > 0) {\n const uniqueErrors = [...new Set(itemsWithErrors.flatMap((i) => i.scorerErrors!))];\n lines.push('');\n lines.push(\n ` Scorer errors (${itemsWithErrors.length}/${result.summary.count} items affected):`,\n );\n for (const err of uniqueErrors.slice(0, 5)) {\n lines.push(` - ${err}`);\n }\n if (uniqueErrors.length > 5) {\n lines.push(` ... and ${uniqueErrors.length - 5} more`);\n }\n }\n\n return lines.join('\\n');\n}\n\n// ── Runtime resolution ─────────────────────────────────────────────\n\nasync function resolveRuntimeFromConfig(configPath: string): Promise<AxlRuntime> {\n try {\n const mod = await importModule(configPath, import.meta.url);\n const runtime = resolveRuntime(mod) as AxlRuntime;\n\n if (!runtime || typeof runtime.execute !== 'function') {\n console.error(`Config must export a default AxlRuntime instance.`);\n if (runtime) {\n const keys = Object.keys(runtime as object)\n .slice(0, 5)\n .join(', ');\n console.error(` Got: ${typeof runtime}${keys ? ` with keys: { ${keys} }` : ''}`);\n }\n console.error(\n `Example:\\n import { AxlRuntime } from '@axlsdk/axl';\\n export default new AxlRuntime({ ... });`,\n );\n process.exit(1);\n }\n\n return runtime;\n } catch (err) {\n const msg = err instanceof Error ? err.message : String(err);\n if (\n /Cannot use import statement|Unexpected reserved word|top-level await|exports is not defined/.test(\n msg,\n )\n ) {\n const ext = path.extname(configPath);\n console.error(`[axl-eval] Config failed to load due to a CJS/ESM compatibility issue.`);\n if (ext === '.ts' || ext === '.tsx') {\n console.error(\n ` Tip: try renaming to .mts to force ESM format, or ensure tsx is installed and up to date.`,\n );\n } else {\n console.error(` Tip: add \"type\": \"module\" to your package.json.`);\n }\n console.error();\n }\n console.error(`Failed to load config:`, err);\n process.exit(1);\n }\n}\n\nasync function getRuntime(configArg?: string, conditions?: string[]): Promise<AxlRuntime> {\n // Register import conditions before any config loading\n if (conditions && conditions.length > 0) {\n await registerConditions(conditions);\n }\n\n // 1. Explicit --config\n if (configArg) {\n const configPath = path.resolve(process.cwd(), configArg);\n const stat = statSync(configPath, { throwIfNoEntry: false });\n if (!stat?.isFile()) {\n console.error(`Config file not found: ${configPath}`);\n process.exit(1);\n }\n console.error(`[axl-eval] Loading config from ${configPath}`);\n return resolveRuntimeFromConfig(configPath);\n }\n\n // 2. Auto-detect\n const found = findConfig(process.cwd());\n if (found) {\n console.error(`[axl-eval] Auto-detected config: ${found}`);\n return resolveRuntimeFromConfig(found);\n }\n\n // 3. Bare runtime (providers from env vars)\n const { AxlRuntime } = await import('@axlsdk/axl');\n return new AxlRuntime();\n}\n\n// ── Arg parsing ────────────────────────────────────────────────────\n\nfunction parseEvalArgs(args: string[]): {\n outputPath?: string;\n configArg?: string;\n conditions: string[];\n paths: string[];\n} {\n let outputPath: string | undefined;\n let configArg: string | undefined;\n let conditions: string[] = [];\n const paths: string[] = [];\n\n for (let i = 0; i < args.length; i++) {\n const arg = args[i];\n if (arg === '--output' || arg === '--config' || arg === '--conditions') {\n if (i + 1 >= args.length) {\n console.error(`Error: ${arg} requires a value`);\n process.exit(1);\n }\n const value = args[++i];\n if (arg === '--output') outputPath = value;\n else if (arg === '--config') configArg = value;\n else\n conditions = value\n .split(',')\n .map((c) => c.trim())\n .filter(Boolean);\n } else if (arg.startsWith('--')) {\n if (!KNOWN_FLAGS.has(arg)) {\n console.error(`Unknown flag: ${arg}`);\n process.exit(1);\n }\n } else {\n paths.push(arg);\n }\n }\n\n return { outputPath, configArg, conditions, paths };\n}\n\n// ── Main eval command ──────────────────────────────────────────────\n\nasync function runEvalCommand(args: string[]) {\n const { outputPath, configArg, conditions, paths } = parseEvalArgs(args);\n\n if (paths.length === 0) {\n console.error('Error: No eval file path provided');\n process.exit(1);\n }\n\n const evalFiles: string[] = [];\n for (const p of paths) {\n evalFiles.push(...collectEvalFiles(p));\n }\n\n if (evalFiles.length === 0) {\n console.error('Error: No eval files found');\n process.exit(1);\n }\n\n const runtime = await getRuntime(configArg, conditions);\n const results: EvalResult[] = [];\n\n try {\n for (const filePath of evalFiles) {\n try {\n const mod = await importModule(path.resolve(filePath), import.meta.url);\n const evalConfig: EvalConfig = mod.default?.default ?? mod.default ?? mod.config ?? mod;\n\n if (!evalConfig.workflow || !evalConfig.dataset || !evalConfig.scorers) {\n console.error(\n `Error: ${filePath} does not export a valid eval config (missing workflow, dataset, or scorers)`,\n );\n continue;\n }\n\n // Resolve executeWorkflow: custom export > registered workflow > passthrough\n let executeWorkflow: (\n input: unknown,\n rt?: unknown,\n ) => Promise<{ output: unknown; cost?: number }>;\n\n if (mod.executeWorkflow) {\n // Wrap custom executeWorkflow with trackCost for automatic cost attribution\n executeWorkflow = async (input, rt) => {\n const { result, cost: trackedCost } = await runtime.trackCost(async () => {\n return mod.executeWorkflow(input, rt);\n });\n return { output: result.output, cost: result.cost ?? trackedCost };\n };\n } else if (runtime.getWorkflow(evalConfig.workflow)) {\n // No executeWorkflow exported but workflow is registered — use runtime.execute()\n executeWorkflow = async (input) => {\n const { result, cost } = await runtime.trackCost(async () => {\n return runtime.execute(evalConfig.workflow, input);\n });\n return { output: result, cost };\n };\n } else {\n console.warn(\n `[axl-eval] Warning: ${filePath} does not export executeWorkflow — using input passthrough`,\n );\n executeWorkflow = async (input) => ({ output: input });\n }\n\n const result = await runEval(evalConfig, executeWorkflow, runtime);\n results.push(result);\n\n console.log('\\n' + formatTable(result) + '\\n');\n } catch (err) {\n console.error(\n `Error running eval ${filePath}: ${err instanceof Error ? err.message : String(err)}`,\n );\n }\n }\n\n if (outputPath && results.length > 0) {\n const output = results.length === 1 ? results[0] : results;\n const outputDir = path.dirname(path.resolve(outputPath));\n await mkdir(outputDir, { recursive: true });\n await writeFileAsync(outputPath, JSON.stringify(output, null, 2), 'utf-8');\n console.log(`Results saved to ${outputPath}`);\n }\n } finally {\n await runtime.shutdown().catch(() => {});\n }\n\n // Exit with non-zero code if no evals succeeded\n if (results.length === 0) {\n process.exit(1);\n }\n}\n\nmain().catch((err) => {\n console.error(err);\n process.exit(1);\n});\n","import type { EvalResult, EvalComparison, EvalRegression, EvalImprovement } from './types.js';\n\nfunction round(n: number): number {\n return Math.round(n * 100) / 100;\n}\n\nexport function evalCompare(baseline: EvalResult, candidate: EvalResult): EvalComparison {\n if (baseline.dataset !== candidate.dataset) {\n throw new Error(\n `Cannot compare evals from different datasets: \"${baseline.dataset}\" vs \"${candidate.dataset}\"`,\n );\n }\n\n const baselineScorerNames = Object.keys(baseline.summary.scorers).sort();\n const candidateScorerNames = Object.keys(candidate.summary.scorers).sort();\n if (JSON.stringify(baselineScorerNames) !== JSON.stringify(candidateScorerNames)) {\n throw new Error(`Cannot compare evals with different scorers`);\n }\n\n const scorers: EvalComparison['scorers'] = {};\n for (const name of baselineScorerNames) {\n const bMean = baseline.summary.scorers[name].mean;\n const cMean = candidate.summary.scorers[name].mean;\n const delta = round(cMean - bMean);\n const deltaPercent = bMean > 0 ? round((delta / bMean) * 100) : 0;\n scorers[name] = { baselineMean: bMean, candidateMean: cMean, delta, deltaPercent };\n }\n\n const regressions: EvalRegression[] = [];\n const improvements: EvalImprovement[] = [];\n\n const minLength = Math.min(baseline.items.length, candidate.items.length);\n for (let i = 0; i < minLength; i++) {\n const bItem = baseline.items[i];\n const cItem = candidate.items[i];\n if (bItem.error || cItem.error) continue;\n for (const name of baselineScorerNames) {\n const bScore = bItem.scores[name];\n const cScore = cItem.scores[name];\n if (bScore == null || cScore == null) continue;\n const delta = round(cScore - bScore);\n if (delta < -0.1)\n regressions.push({\n itemIndex: i,\n input: bItem.input,\n scorer: name,\n baselineScore: bScore,\n candidateScore: cScore,\n delta,\n });\n else if (delta > 0.1)\n improvements.push({\n itemIndex: i,\n input: bItem.input,\n scorer: name,\n baselineScore: bScore,\n candidateScore: cScore,\n delta,\n });\n }\n }\n\n // Timing comparison (use pre-computed summary.timing when available)\n let timing: EvalComparison['timing'];\n if (baseline.summary.timing && candidate.summary.timing) {\n const bMean = baseline.summary.timing.mean;\n const cMean = candidate.summary.timing.mean;\n const delta = round(cMean - bMean);\n const deltaPercent = bMean > 0 ? round((delta / bMean) * 100) : 0;\n timing = { baselineMean: bMean, candidateMean: cMean, delta, deltaPercent };\n }\n\n // Cost comparison\n let cost: EvalComparison['cost'];\n if (baseline.totalCost > 0 || candidate.totalCost > 0) {\n const delta = round(candidate.totalCost - baseline.totalCost);\n const deltaPercent = baseline.totalCost > 0 ? round((delta / baseline.totalCost) * 100) : 0;\n cost = {\n baselineTotal: baseline.totalCost,\n candidateTotal: candidate.totalCost,\n delta,\n deltaPercent,\n };\n }\n\n const parts: string[] = [];\n for (const name of baselineScorerNames) {\n const s = scorers[name];\n if (Math.abs(s.delta) > 0.001) {\n const direction = s.delta > 0 ? 'improves' : 'regresses';\n const sign = s.delta > 0 ? '+' : '';\n parts.push(\n `'${name}' ${direction} by ${sign}${s.deltaPercent.toFixed(1)}% (${s.baselineMean.toFixed(2)} -> ${s.candidateMean.toFixed(2)})`,\n );\n }\n }\n if (timing && Math.abs(timing.deltaPercent) > 1) {\n const dir = timing.delta > 0 ? 'slower' : 'faster';\n parts.push(`${Math.abs(timing.deltaPercent).toFixed(0)}% ${dir}`);\n }\n if (cost && Math.abs(cost.deltaPercent) > 1) {\n const dir = cost.delta > 0 ? 'more expensive' : 'cheaper';\n parts.push(`${Math.abs(cost.deltaPercent).toFixed(0)}% ${dir}`);\n }\n const summaryStr = `candidate ${parts.length > 0 ? parts.join(', ') : 'no significant changes'} with ${regressions.length} regressions and ${improvements.length} improvements`;\n\n return {\n baseline: { id: baseline.id, metadata: baseline.metadata },\n candidate: { id: candidate.id, metadata: candidate.metadata },\n scorers,\n timing,\n cost,\n regressions,\n improvements,\n summary: summaryStr,\n };\n}\n","/** Result from a scorer that includes metadata beyond the numeric score. */\nexport type ScorerResult = {\n score: number;\n metadata?: Record<string, unknown>;\n cost?: number;\n};\n\n/** Normalize a scorer return value to a ScorerResult. */\nexport function normalizeScorerResult(result: number | ScorerResult): ScorerResult {\n return typeof result === 'number' ? { score: result } : result;\n}\n\n/** Context passed to scorers by the eval runner. */\nexport type ScorerContext = {\n /** Resolve a provider:model URI to a provider instance and model name. */\n resolveProvider: (modelUri: string) => {\n provider: {\n chat(\n messages: { role: string; content: string }[],\n options: {\n model: string;\n temperature?: number;\n responseFormat?: { type: string; json_schema?: unknown };\n },\n ): Promise<{ content: string; cost?: number }>;\n };\n model: string;\n };\n};\n\nexport type ScorerFn<TOutput, TInput, TAnnotations> = (\n output: TOutput,\n input: TInput,\n annotations?: TAnnotations,\n) => number | ScorerResult;\n\nexport type ScorerConfig<TOutput = unknown, TInput = unknown, TAnnotations = unknown> = {\n name: string;\n description: string;\n score: ScorerFn<TOutput, TInput, TAnnotations>;\n};\n\nexport type Scorer<TOutput = unknown, TInput = unknown, TAnnotations = unknown> = {\n readonly name: string;\n readonly description: string;\n readonly isLlm: boolean;\n score(\n output: TOutput,\n input: TInput,\n annotations?: TAnnotations,\n context?: ScorerContext,\n ): number | ScorerResult | Promise<number | ScorerResult>;\n};\n\nexport function scorer<TOutput = unknown, TInput = unknown, TAnnotations = unknown>(\n config: ScorerConfig<TOutput, TInput, TAnnotations>,\n): Scorer<TOutput, TInput, TAnnotations> {\n return {\n name: config.name,\n description: config.description,\n isLlm: false,\n score: config.score,\n };\n}\n","import type { AxlRuntime } from '@axlsdk/axl';\nimport type { EvalConfig, EvalResult, EvalItem, EvalSummary } from './types.js';\nimport type { ScorerContext } from './scorer.js';\nimport { normalizeScorerResult } from './scorer.js';\nimport { randomUUID } from 'node:crypto';\n\nfunction computeStats(scores: number[]): {\n mean: number;\n min: number;\n max: number;\n p50: number;\n p95: number;\n} {\n if (scores.length === 0) return { mean: 0, min: 0, max: 0, p50: 0, p95: 0 };\n const sorted = [...scores].sort((a, b) => a - b);\n const mean = sorted.reduce((a, b) => a + b, 0) / sorted.length;\n const min = sorted[0];\n const max = sorted[sorted.length - 1];\n const p50 = sorted[Math.floor(sorted.length * 0.5)];\n const p95 = sorted[Math.floor(sorted.length * 0.95)];\n return { mean: round(mean), min: round(min), max: round(max), p50: round(p50), p95: round(p95) };\n}\n\nfunction round(n: number): number {\n return Math.round(n * 1000) / 1000;\n}\n\nfunction parseCost(cost: string): number {\n const match = cost.match(/^\\$?([\\d.]+)$/);\n if (!match) throw new Error(`Invalid cost format: \"${cost}\"`);\n return parseFloat(match[1]);\n}\n\nexport async function runEval(\n config: EvalConfig,\n executeWorkflow: (\n input: unknown,\n runtime: AxlRuntime,\n ) => Promise<{ output: unknown; cost?: number }>,\n runtime: AxlRuntime,\n): Promise<EvalResult> {\n const startTime = Date.now();\n const id = randomUUID();\n const items = await config.dataset.getItems();\n const concurrency = config.concurrency ?? 5;\n const budgetLimit = config.budget ? parseCost(config.budget) : undefined;\n\n // Create a scorer context that LLM scorers use to resolve providers.\n const scorerContext: ScorerContext = {\n resolveProvider: (uri: string) => {\n if (typeof runtime.resolveProvider !== 'function') {\n throw new Error(\n `LLM scorers require a runtime with resolveProvider(). ` +\n `Ensure you are using a real AxlRuntime instance, not a mock.`,\n );\n }\n return runtime.resolveProvider(uri);\n },\n };\n\n const evalItems: EvalItem[] = new Array(items.length);\n let totalCost = 0;\n let budgetExceeded = false;\n\n async function processItem(item: (typeof items)[0], itemIndex: number): Promise<void> {\n if (budgetExceeded) {\n evalItems[itemIndex] = {\n input: item.input,\n annotations: item.annotations,\n output: null,\n error: 'Budget exceeded',\n scores: {},\n };\n return;\n }\n\n const evalItem: EvalItem = {\n input: item.input,\n annotations: item.annotations,\n output: null,\n scores: {},\n };\n const itemStart = Date.now();\n try {\n const result = await executeWorkflow(item.input, runtime);\n evalItem.duration = Date.now() - itemStart;\n evalItem.output = result.output;\n evalItem.cost = result.cost;\n if (result.cost != null) {\n totalCost += result.cost;\n }\n } catch (err) {\n evalItem.duration = Date.now() - itemStart;\n evalItem.error = err instanceof Error ? err.message : String(err);\n evalItems[itemIndex] = evalItem;\n return;\n }\n\n if (budgetLimit != null && totalCost > budgetLimit) {\n budgetExceeded = true;\n }\n\n evalItem.scoreDetails = {};\n let itemScorerCost = 0;\n\n for (const scorer of config.scorers) {\n const scorerStart = Date.now();\n try {\n const raw = await scorer.score(\n evalItem.output,\n item.input,\n item.annotations,\n scorerContext,\n );\n const scorerResult = normalizeScorerResult(raw);\n\n if (scorerResult.cost != null) {\n itemScorerCost += scorerResult.cost;\n totalCost += scorerResult.cost;\n }\n\n const scorerDuration = Date.now() - scorerStart;\n\n if (\n !Number.isFinite(scorerResult.score) ||\n scorerResult.score < 0 ||\n scorerResult.score > 1\n ) {\n if (!evalItem.scorerErrors) evalItem.scorerErrors = [];\n evalItem.scorerErrors.push(\n `Scorer \"${scorer.name}\" returned out-of-range score ${scorerResult.score} for input ${JSON.stringify(item.input)}`,\n );\n evalItem.scores[scorer.name] = null;\n evalItem.scoreDetails[scorer.name] = {\n score: null,\n metadata: scorerResult.metadata,\n duration: scorerDuration,\n cost: scorerResult.cost,\n };\n } else {\n evalItem.scores[scorer.name] = round(scorerResult.score);\n evalItem.scoreDetails[scorer.name] = {\n score: round(scorerResult.score),\n metadata: scorerResult.metadata,\n duration: scorerDuration,\n cost: scorerResult.cost,\n };\n }\n } catch (err) {\n // Capture cost from error (LLM scorer attaches it)\n const errCost = typeof (err as any)?.cost === 'number' ? (err as any).cost : undefined;\n if (errCost != null) {\n itemScorerCost += errCost;\n totalCost += errCost;\n }\n if (!evalItem.scorerErrors) evalItem.scorerErrors = [];\n evalItem.scorerErrors.push(\n `Scorer \"${scorer.name}\" threw: ${err instanceof Error ? err.message : String(err)}`,\n );\n evalItem.scores[scorer.name] = null;\n evalItem.scoreDetails[scorer.name] = {\n score: null,\n duration: Date.now() - scorerStart,\n cost: errCost,\n };\n }\n }\n evalItem.scorerCost = itemScorerCost > 0 ? itemScorerCost : undefined;\n evalItems[itemIndex] = evalItem;\n }\n\n let index = 0;\n async function runNext(): Promise<void> {\n while (index < items.length) {\n const currentIndex = index++;\n await processItem(items[currentIndex], currentIndex);\n }\n }\n\n const workers = Array.from({ length: Math.min(concurrency, items.length) }, () => runNext());\n await Promise.all(workers);\n\n const failures = evalItems.filter((i) => i.error).length;\n const scorerNames = config.scorers.map((s) => s.name);\n const scorerStats: EvalSummary['scorers'] = {};\n for (const name of scorerNames) {\n const scores = evalItems\n .filter((i) => !i.error && i.scores[name] != null)\n .map((i) => i.scores[name] as number);\n scorerStats[name] = computeStats(scores);\n }\n\n const durations = evalItems.filter((i) => !i.error && i.duration != null).map((i) => i.duration!);\n const timing = durations.length > 0 ? computeStats(durations) : undefined;\n\n return {\n id,\n workflow: config.workflow,\n dataset: config.dataset.name,\n metadata: config.metadata ?? {},\n timestamp: new Date().toISOString(),\n totalCost,\n duration: Date.now() - startTime,\n items: evalItems,\n summary: { count: items.length, failures, scorers: scorerStats, timing },\n };\n}\n","/**\n * CLI utilities for config detection, runtime resolution, and loader registration.\n *\n * These are duplicated from @axlsdk/studio (cli-utils.ts, resolve-runtime.ts,\n * eval-loader.ts, cli.ts) because @axlsdk/eval cannot depend on @axlsdk/studio.\n * Keep in sync with the studio versions if either changes.\n */\n\nimport { resolve } from 'node:path';\nimport { existsSync } from 'node:fs';\nimport { pathToFileURL } from 'node:url';\n\n// ── Config auto-detection ──────────────────────────────────────────\n\nexport const CONFIG_CANDIDATES = [\n 'axl.config.mts',\n 'axl.config.ts',\n 'axl.config.mjs',\n 'axl.config.js',\n];\n\nexport function findConfig(cwd: string): string | undefined {\n for (const name of CONFIG_CANDIDATES) {\n const p = resolve(cwd, name);\n if (existsSync(p)) return p;\n }\n return undefined;\n}\n\n// ── Extension helpers ──────────────────────────────────────────────\n\n/** Returns true if the file is TypeScript and needs tsx to load. */\nexport function needsTsxLoader(filePath: string): boolean {\n return /\\.[mc]?tsx?$/.test(filePath);\n}\n\n// ── Runtime resolution ─────────────────────────────────────────────\n\n/**\n * Resolve the AxlRuntime from a dynamically imported config module.\n *\n * Module shapes handled:\n * - ESM `export default runtime` → mod.default is the runtime\n * - CJS compiled from `export default runtime` → mod.default.default is the runtime\n * - CJS `module.exports = runtime` → mod.default is the runtime\n * - Named `export { runtime }` → mod.runtime is the runtime\n */\nexport function resolveRuntime(mod: Record<string, any>): unknown {\n const def = mod.default as Record<string, any> | undefined;\n return def?.default ?? def ?? mod.runtime;\n}\n\n// ── Module loading ────────────────────────────────────────────────\n\n// Lazily resolved tsImport function from tsx. `undefined` = not yet checked,\n// `null` = tsx not available.\nlet tsImportFn:\n | ((specifier: string, parentURL: string) => Promise<Record<string, any>>)\n | null\n | undefined;\n\n/**\n * Import a module, using tsx's `tsImport()` for TypeScript files.\n *\n * `tsImport()` handles ESM/CJS format correctly without process-wide side effects —\n * no need for `register()` hooks or ESM-forcing workarounds. Falls back to regular\n * `import()` for non-TypeScript files or when tsx is not installed.\n */\nexport async function importModule(\n filePath: string,\n parentURL: string,\n): Promise<Record<string, any>> {\n if (needsTsxLoader(filePath)) {\n if (tsImportFn === undefined) {\n try {\n // @ts-expect-error — tsx is an optional runtime dependency\n const mod = await import('tsx/esm/api');\n tsImportFn = mod.tsImport ?? null;\n } catch {\n tsImportFn = null;\n console.warn(\n '[axl-eval] Warning: tsx is not installed. TypeScript files require tsx.\\n' +\n ' Install it with: npm install -D tsx',\n );\n }\n }\n if (tsImportFn) {\n return (await tsImportFn(pathToFileURL(filePath).href, parentURL)) as Record<string, any>;\n }\n }\n return await import(pathToFileURL(filePath).href);\n}\n\n// ── Conditions ────────────────────────────────────────────────────\n\nexport async function registerConditions(conditions: string[]): Promise<void> {\n try {\n const nodeModule = await import('node:module');\n const hookCode = [\n `const extra = ${JSON.stringify(conditions)};`,\n `export async function resolve(specifier, context, nextResolve) {`,\n ` return nextResolve(specifier, {`,\n ` ...context,`,\n ` conditions: [...new Set([...context.conditions, ...extra])],`,\n ` });`,\n `}`,\n ].join('\\n');\n nodeModule.register(`data:text/javascript,${encodeURIComponent(hookCode)}`);\n } catch {\n console.warn('[axl-eval] Warning: --conditions requires Node.js 20.6+');\n }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;AAEA,IAAAA,kBAAsC;AACtC,sBAA8E;AAC9E,WAAsB;;;ACFtB,SAAS,MAAM,GAAmB;AAChC,SAAO,KAAK,MAAM,IAAI,GAAG,IAAI;AAC/B;AAEO,SAAS,YAAY,UAAsB,WAAuC;AACvF,MAAI,SAAS,YAAY,UAAU,SAAS;AAC1C,UAAM,IAAI;AAAA,MACR,kDAAkD,SAAS,OAAO,SAAS,UAAU,OAAO;AAAA,IAC9F;AAAA,EACF;AAEA,QAAM,sBAAsB,OAAO,KAAK,SAAS,QAAQ,OAAO,EAAE,KAAK;AACvE,QAAM,uBAAuB,OAAO,KAAK,UAAU,QAAQ,OAAO,EAAE,KAAK;AACzE,MAAI,KAAK,UAAU,mBAAmB,MAAM,KAAK,UAAU,oBAAoB,GAAG;AAChF,UAAM,IAAI,MAAM,6CAA6C;AAAA,EAC/D;AAEA,QAAM,UAAqC,CAAC;AAC5C,aAAW,QAAQ,qBAAqB;AACtC,UAAM,QAAQ,SAAS,QAAQ,QAAQ,IAAI,EAAE;AAC7C,UAAM,QAAQ,UAAU,QAAQ,QAAQ,IAAI,EAAE;AAC9C,UAAM,QAAQ,MAAM,QAAQ,KAAK;AACjC,UAAM,eAAe,QAAQ,IAAI,MAAO,QAAQ,QAAS,GAAG,IAAI;AAChE,YAAQ,IAAI,IAAI,EAAE,cAAc,OAAO,eAAe,OAAO,OAAO,aAAa;AAAA,EACnF;AAEA,QAAM,cAAgC,CAAC;AACvC,QAAM,eAAkC,CAAC;AAEzC,QAAM,YAAY,KAAK,IAAI,SAAS,MAAM,QAAQ,UAAU,MAAM,MAAM;AACxE,WAAS,IAAI,GAAG,IAAI,WAAW,KAAK;AAClC,UAAM,QAAQ,SAAS,MAAM,CAAC;AAC9B,UAAM,QAAQ,UAAU,MAAM,CAAC;AAC/B,QAAI,MAAM,SAAS,MAAM,MAAO;AAChC,eAAW,QAAQ,qBAAqB;AACtC,YAAM,SAAS,MAAM,OAAO,IAAI;AAChC,YAAM,SAAS,MAAM,OAAO,IAAI;AAChC,UAAI,UAAU,QAAQ,UAAU,KAAM;AACtC,YAAM,QAAQ,MAAM,SAAS,MAAM;AACnC,UAAI,QAAQ;AACV,oBAAY,KAAK;AAAA,UACf,WAAW;AAAA,UACX,OAAO,MAAM;AAAA,UACb,QAAQ;AAAA,UACR,eAAe;AAAA,UACf,gBAAgB;AAAA,UAChB;AAAA,QACF,CAAC;AAAA,eACM,QAAQ;AACf,qBAAa,KAAK;AAAA,UAChB,WAAW;AAAA,UACX,OAAO,MAAM;AAAA,UACb,QAAQ;AAAA,UACR,eAAe;AAAA,UACf,gBAAgB;AAAA,UAChB;AAAA,QACF,CAAC;AAAA,IACL;AAAA,EACF;AAGA,MAAI;AACJ,MAAI,SAAS,QAAQ,UAAU,UAAU,QAAQ,QAAQ;AACvD,UAAM,QAAQ,SAAS,QAAQ,OAAO;AACtC,UAAM,QAAQ,UAAU,QAAQ,OAAO;AACvC,UAAM,QAAQ,MAAM,QAAQ,KAAK;AACjC,UAAM,eAAe,QAAQ,IAAI,MAAO,QAAQ,QAAS,GAAG,IAAI;AAChE,aAAS,EAAE,cAAc,OAAO,eAAe,OAAO,OAAO,aAAa;AAAA,EAC5E;AAGA,MAAI;AACJ,MAAI,SAAS,YAAY,KAAK,UAAU,YAAY,GAAG;AACrD,UAAM,QAAQ,MAAM,UAAU,YAAY,SAAS,SAAS;AAC5D,UAAM,eAAe,SAAS,YAAY,IAAI,MAAO,QAAQ,SAAS,YAAa,GAAG,IAAI;AAC1F,WAAO;AAAA,MACL,eAAe,SAAS;AAAA,MACxB,gBAAgB,UAAU;AAAA,MAC1B;AAAA,MACA;AAAA,IACF;AAAA,EACF;AAEA,QAAM,QAAkB,CAAC;AACzB,aAAW,QAAQ,qBAAqB;AACtC,UAAM,IAAI,QAAQ,IAAI;AACtB,QAAI,KAAK,IAAI,EAAE,KAAK,IAAI,MAAO;AAC7B,YAAM,YAAY,EAAE,QAAQ,IAAI,aAAa;AAC7C,YAAM,OAAO,EAAE,QAAQ,IAAI,MAAM;AACjC,YAAM;AAAA,QACJ,IAAI,IAAI,KAAK,SAAS,OAAO,IAAI,GAAG,EAAE,aAAa,QAAQ,CAAC,CAAC,MAAM,EAAE,aAAa,QAAQ,CAAC,CAAC,OAAO,EAAE,cAAc,QAAQ,CAAC,CAAC;AAAA,MAC/H;AAAA,IACF;AAAA,EACF;AACA,MAAI,UAAU,KAAK,IAAI,OAAO,YAAY,IAAI,GAAG;AAC/C,UAAM,MAAM,OAAO,QAAQ,IAAI,WAAW;AAC1C,UAAM,KAAK,GAAG,KAAK,IAAI,OAAO,YAAY,EAAE,QAAQ,CAAC,CAAC,KAAK,GAAG,EAAE;AAAA,EAClE;AACA,MAAI,QAAQ,KAAK,IAAI,KAAK,YAAY,IAAI,GAAG;AAC3C,UAAM,MAAM,KAAK,QAAQ,IAAI,mBAAmB;AAChD,UAAM,KAAK,GAAG,KAAK,IAAI,KAAK,YAAY,EAAE,QAAQ,CAAC,CAAC,KAAK,GAAG,EAAE;AAAA,EAChE;AACA,QAAM,aAAa,aAAa,MAAM,SAAS,IAAI,MAAM,KAAK,IAAI,IAAI,wBAAwB,SAAS,YAAY,MAAM,oBAAoB,aAAa,MAAM;AAEhK,SAAO;AAAA,IACL,UAAU,EAAE,IAAI,SAAS,IAAI,UAAU,SAAS,SAAS;AAAA,IACzD,WAAW,EAAE,IAAI,UAAU,IAAI,UAAU,UAAU,SAAS;AAAA,IAC5D;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA,SAAS;AAAA,EACX;AACF;;;AC5GO,SAAS,sBAAsB,QAA6C;AACjF,SAAO,OAAO,WAAW,WAAW,EAAE,OAAO,OAAO,IAAI;AAC1D;;;ACNA,yBAA2B;AAE3B,SAAS,aAAa,QAMpB;AACA,MAAI,OAAO,WAAW,EAAG,QAAO,EAAE,MAAM,GAAG,KAAK,GAAG,KAAK,GAAG,KAAK,GAAG,KAAK,EAAE;AAC1E,QAAM,SAAS,CAAC,GAAG,MAAM,EAAE,KAAK,CAAC,GAAG,MAAM,IAAI,CAAC;AAC/C,QAAM,OAAO,OAAO,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,OAAO;AACxD,QAAM,MAAM,OAAO,CAAC;AACpB,QAAM,MAAM,OAAO,OAAO,SAAS,CAAC;AACpC,QAAM,MAAM,OAAO,KAAK,MAAM,OAAO,SAAS,GAAG,CAAC;AAClD,QAAM,MAAM,OAAO,KAAK,MAAM,OAAO,SAAS,IAAI,CAAC;AACnD,SAAO,EAAE,MAAMC,OAAM,IAAI,GAAG,KAAKA,OAAM,GAAG,GAAG,KAAKA,OAAM,GAAG,GAAG,KAAKA,OAAM,GAAG,GAAG,KAAKA,OAAM,GAAG,EAAE;AACjG;AAEA,SAASA,OAAM,GAAmB;AAChC,SAAO,KAAK,MAAM,IAAI,GAAI,IAAI;AAChC;AAEA,SAAS,UAAU,MAAsB;AACvC,QAAM,QAAQ,KAAK,MAAM,eAAe;AACxC,MAAI,CAAC,MAAO,OAAM,IAAI,MAAM,yBAAyB,IAAI,GAAG;AAC5D,SAAO,WAAW,MAAM,CAAC,CAAC;AAC5B;AAEA,eAAsB,QACpB,QACA,iBAIA,SACqB;AACrB,QAAM,YAAY,KAAK,IAAI;AAC3B,QAAM,SAAK,+BAAW;AACtB,QAAM,QAAQ,MAAM,OAAO,QAAQ,SAAS;AAC5C,QAAM,cAAc,OAAO,eAAe;AAC1C,QAAM,cAAc,OAAO,SAAS,UAAU,OAAO,MAAM,IAAI;AAG/D,QAAM,gBAA+B;AAAA,IACnC,iBAAiB,CAAC,QAAgB;AAChC,UAAI,OAAO,QAAQ,oBAAoB,YAAY;AACjD,cAAM,IAAI;AAAA,UACR;AAAA,QAEF;AAAA,MACF;AACA,aAAO,QAAQ,gBAAgB,GAAG;AAAA,IACpC;AAAA,EACF;AAEA,QAAM,YAAwB,IAAI,MAAM,MAAM,MAAM;AACpD,MAAI,YAAY;AAChB,MAAI,iBAAiB;AAErB,iBAAe,YAAY,MAAyB,WAAkC;AACpF,QAAI,gBAAgB;AAClB,gBAAU,SAAS,IAAI;AAAA,QACrB,OAAO,KAAK;AAAA,QACZ,aAAa,KAAK;AAAA,QAClB,QAAQ;AAAA,QACR,OAAO;AAAA,QACP,QAAQ,CAAC;AAAA,MACX;AACA;AAAA,IACF;AAEA,UAAM,WAAqB;AAAA,MACzB,OAAO,KAAK;AAAA,MACZ,aAAa,KAAK;AAAA,MAClB,QAAQ;AAAA,MACR,QAAQ,CAAC;AAAA,IACX;AACA,UAAM,YAAY,KAAK,IAAI;AAC3B,QAAI;AACF,YAAM,SAAS,MAAM,gBAAgB,KAAK,OAAO,OAAO;AACxD,eAAS,WAAW,KAAK,IAAI,IAAI;AACjC,eAAS,SAAS,OAAO;AACzB,eAAS,OAAO,OAAO;AACvB,UAAI,OAAO,QAAQ,MAAM;AACvB,qBAAa,OAAO;AAAA,MACtB;AAAA,IACF,SAAS,KAAK;AACZ,eAAS,WAAW,KAAK,IAAI,IAAI;AACjC,eAAS,QAAQ,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAChE,gBAAU,SAAS,IAAI;AACvB;AAAA,IACF;AAEA,QAAI,eAAe,QAAQ,YAAY,aAAa;AAClD,uBAAiB;AAAA,IACnB;AAEA,aAAS,eAAe,CAAC;AACzB,QAAI,iBAAiB;AAErB,eAAW,UAAU,OAAO,SAAS;AACnC,YAAM,cAAc,KAAK,IAAI;AAC7B,UAAI;AACF,cAAM,MAAM,MAAM,OAAO;AAAA,UACvB,SAAS;AAAA,UACT,KAAK;AAAA,UACL,KAAK;AAAA,UACL;AAAA,QACF;AACA,cAAM,eAAe,sBAAsB,GAAG;AAE9C,YAAI,aAAa,QAAQ,MAAM;AAC7B,4BAAkB,aAAa;AAC/B,uBAAa,aAAa;AAAA,QAC5B;AAEA,cAAM,iBAAiB,KAAK,IAAI,IAAI;AAEpC,YACE,CAAC,OAAO,SAAS,aAAa,KAAK,KACnC,aAAa,QAAQ,KACrB,aAAa,QAAQ,GACrB;AACA,cAAI,CAAC,SAAS,aAAc,UAAS,eAAe,CAAC;AACrD,mBAAS,aAAa;AAAA,YACpB,WAAW,OAAO,IAAI,iCAAiC,aAAa,KAAK,cAAc,KAAK,UAAU,KAAK,KAAK,CAAC;AAAA,UACnH;AACA,mBAAS,OAAO,OAAO,IAAI,IAAI;AAC/B,mBAAS,aAAa,OAAO,IAAI,IAAI;AAAA,YACnC,OAAO;AAAA,YACP,UAAU,aAAa;AAAA,YACvB,UAAU;AAAA,YACV,MAAM,aAAa;AAAA,UACrB;AAAA,QACF,OAAO;AACL,mBAAS,OAAO,OAAO,IAAI,IAAIA,OAAM,aAAa,KAAK;AACvD,mBAAS,aAAa,OAAO,IAAI,IAAI;AAAA,YACnC,OAAOA,OAAM,aAAa,KAAK;AAAA,YAC/B,UAAU,aAAa;AAAA,YACvB,UAAU;AAAA,YACV,MAAM,aAAa;AAAA,UACrB;AAAA,QACF;AAAA,MACF,SAAS,KAAK;AAEZ,cAAM,UAAU,OAAQ,KAAa,SAAS,WAAY,IAAY,OAAO;AAC7E,YAAI,WAAW,MAAM;AACnB,4BAAkB;AAClB,uBAAa;AAAA,QACf;AACA,YAAI,CAAC,SAAS,aAAc,UAAS,eAAe,CAAC;AACrD,iBAAS,aAAa;AAAA,UACpB,WAAW,OAAO,IAAI,YAAY,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,CAAC;AAAA,QACpF;AACA,iBAAS,OAAO,OAAO,IAAI,IAAI;AAC/B,iBAAS,aAAa,OAAO,IAAI,IAAI;AAAA,UACnC,OAAO;AAAA,UACP,UAAU,KAAK,IAAI,IAAI;AAAA,UACvB,MAAM;AAAA,QACR;AAAA,MACF;AAAA,IACF;AACA,aAAS,aAAa,iBAAiB,IAAI,iBAAiB;AAC5D,cAAU,SAAS,IAAI;AAAA,EACzB;AAEA,MAAI,QAAQ;AACZ,iBAAe,UAAyB;AACtC,WAAO,QAAQ,MAAM,QAAQ;AAC3B,YAAM,eAAe;AACrB,YAAM,YAAY,MAAM,YAAY,GAAG,YAAY;AAAA,IACrD;AAAA,EACF;AAEA,QAAM,UAAU,MAAM,KAAK,EAAE,QAAQ,KAAK,IAAI,aAAa,MAAM,MAAM,EAAE,GAAG,MAAM,QAAQ,CAAC;AAC3F,QAAM,QAAQ,IAAI,OAAO;AAEzB,QAAM,WAAW,UAAU,OAAO,CAAC,MAAM,EAAE,KAAK,EAAE;AAClD,QAAM,cAAc,OAAO,QAAQ,IAAI,CAAC,MAAM,EAAE,IAAI;AACpD,QAAM,cAAsC,CAAC;AAC7C,aAAW,QAAQ,aAAa;AAC9B,UAAM,SAAS,UACZ,OAAO,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,OAAO,IAAI,KAAK,IAAI,EAChD,IAAI,CAAC,MAAM,EAAE,OAAO,IAAI,CAAW;AACtC,gBAAY,IAAI,IAAI,aAAa,MAAM;AAAA,EACzC;AAEA,QAAM,YAAY,UAAU,OAAO,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,YAAY,IAAI,EAAE,IAAI,CAAC,MAAM,EAAE,QAAS;AAChG,QAAM,SAAS,UAAU,SAAS,IAAI,aAAa,SAAS,IAAI;AAEhE,SAAO;AAAA,IACL;AAAA,IACA,UAAU,OAAO;AAAA,IACjB,SAAS,OAAO,QAAQ;AAAA,IACxB,UAAU,OAAO,YAAY,CAAC;AAAA,IAC9B,YAAW,oBAAI,KAAK,GAAE,YAAY;AAAA,IAClC;AAAA,IACA,UAAU,KAAK,IAAI,IAAI;AAAA,IACvB,OAAO;AAAA,IACP,SAAS,EAAE,OAAO,MAAM,QAAQ,UAAU,SAAS,aAAa,OAAO;AAAA,EACzE;AACF;;;ACtMA,uBAAwB;AACxB,qBAA2B;AAC3B,sBAA8B;AAIvB,IAAM,oBAAoB;AAAA,EAC/B;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;AAEO,SAAS,WAAW,KAAiC;AAC1D,aAAW,QAAQ,mBAAmB;AACpC,UAAM,QAAI,0BAAQ,KAAK,IAAI;AAC3B,YAAI,2BAAW,CAAC,EAAG,QAAO;AAAA,EAC5B;AACA,SAAO;AACT;AAKO,SAAS,eAAe,UAA2B;AACxD,SAAO,eAAe,KAAK,QAAQ;AACrC;AAaO,SAAS,eAAe,KAAmC;AAChE,QAAM,MAAM,IAAI;AAChB,SAAO,KAAK,WAAW,OAAO,IAAI;AACpC;AAMA,IAAI;AAYJ,eAAsB,aACpB,UACA,WAC8B;AAC9B,MAAI,eAAe,QAAQ,GAAG;AAC5B,QAAI,eAAe,QAAW;AAC5B,UAAI;AAEF,cAAM,MAAM,MAAM,OAAO,aAAa;AACtC,qBAAa,IAAI,YAAY;AAAA,MAC/B,QAAQ;AACN,qBAAa;AACb,gBAAQ;AAAA,UACN;AAAA,QAEF;AAAA,MACF;AAAA,IACF;AACA,QAAI,YAAY;AACd,aAAQ,MAAM,eAAW,+BAAc,QAAQ,EAAE,MAAM,SAAS;AAAA,IAClE;AAAA,EACF;AACA,SAAO,MAAM,WAAO,+BAAc,QAAQ,EAAE;AAC9C;AAIA,eAAsB,mBAAmB,YAAqC;AAC5E,MAAI;AACF,UAAM,aAAa,MAAM,OAAO,QAAa;AAC7C,UAAM,WAAW;AAAA,MACf,iBAAiB,KAAK,UAAU,UAAU,CAAC;AAAA,MAC3C;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF,EAAE,KAAK,IAAI;AACX,eAAW,SAAS,wBAAwB,mBAAmB,QAAQ,CAAC,EAAE;AAAA,EAC5E,QAAQ;AACN,YAAQ,KAAK,yDAAyD;AAAA,EACxE;AACF;;;AJ/GA;AAiBA,IAAM,cAAc,oBAAI,IAAI,CAAC,YAAY,YAAY,gBAAgB,sBAAsB,CAAC;AAE5F,eAAe,OAAO;AACpB,QAAM,OAAO,QAAQ,KAAK,MAAM,CAAC;AAEjC,MAAI,KAAK,WAAW,KAAK,KAAK,CAAC,MAAM,YAAY,KAAK,CAAC,MAAM,MAAM;AACjE,YAAQ,IAAI;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,IAUZ,kBAAkB,KAAK,MAAM,CAAC;AAAA;AAAA;AAAA;AAAA;AAAA,CAKjC;AACG,YAAQ,KAAK,CAAC;AAAA,EAChB;AAEA,MAAI,KAAK,CAAC,MAAM,WAAW;AACzB,UAAM,WAAW,KAAK,MAAM,CAAC,CAAC;AAC9B;AAAA,EACF;AAEA,QAAM,eAAe,IAAI;AAC3B;AAEA,eAAe,WAAW,MAAgB;AACxC,QAAM,mBAAmB,KAAK,SAAS,sBAAsB;AAC7D,QAAM,QAAQ,KAAK,OAAO,CAAC,MAAM,CAAC,EAAE,WAAW,IAAI,CAAC;AAEpD,MAAI,MAAM,WAAW,GAAG;AACtB,YAAQ,MAAM,0DAA0D;AACxE,YAAQ,KAAK,CAAC;AAAA,EAChB;AAEA,QAAM,WAAuB,KAAK,MAAM,UAAM,gBAAAC,UAAc,MAAM,CAAC,GAAG,OAAO,CAAC;AAC9E,QAAM,YAAwB,KAAK,MAAM,UAAM,gBAAAA,UAAc,MAAM,CAAC,GAAG,OAAO,CAAC;AAC/E,QAAM,aAAa,YAAY,UAAU,SAAS;AAElD,UAAQ;AAAA,IACN;AAAA,qBAAwB,WAAW,SAAS,GAAG,MAAM,GAAG,CAAC,CAAC,mBAAmB,WAAW,UAAU,GAAG,MAAM,GAAG,CAAC,CAAC;AAAA;AAAA,EAClH;AAEA,QAAM,cAAc,OAAO,KAAK,WAAW,OAAO;AAClD,QAAM,aAAa,KAAK,IAAI,GAAG,YAAY,IAAI,CAAC,MAAM,EAAE,MAAM,GAAG,CAAC;AAElE,UAAQ,IAAI,KAAK,SAAS,OAAO,UAAU,CAAC,yCAAyC;AACrF,UAAQ,IAAI,KAAK,GAAG,OAAO,YAAY,GAAG,CAAC,yCAAyC;AAEpF,aAAW,QAAQ,aAAa;AAC9B,UAAM,IAAI,WAAW,QAAQ,IAAI;AACjC,UAAM,OAAO,EAAE,QAAQ,IAAI,MAAM;AACjC,YAAQ;AAAA,MACN,KAAK,KAAK,OAAO,UAAU,CAAC,KAAK,EAAE,aAAa,QAAQ,CAAC,EAAE,SAAS,CAAC,CAAC,KAAK,EAAE,cAAc,QAAQ,CAAC,EAAE,SAAS,CAAC,CAAC,MAAM,OAAO,EAAE,MAAM,QAAQ,CAAC,GAAG,SAAS,CAAC,CAAC,MAAM,OAAO,EAAE,aAAa,QAAQ,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC;AAAA,IACxN;AAAA,EACF;AAEA,MAAI,WAAW,QAAQ;AACrB,UAAM,IAAI,WAAW;AACrB,UAAM,OAAO,EAAE,QAAQ,IAAI,MAAM;AACjC,YAAQ;AAAA,MACN;AAAA,sBAAyB,EAAE,eAAe,KAAM,QAAQ,CAAC,CAAC,mBAAmB,EAAE,gBAAgB,KAAM,QAAQ,CAAC,CAAC,MAAM,IAAI,GAAG,EAAE,aAAa,QAAQ,CAAC,CAAC;AAAA,IACvJ;AAAA,EACF;AACA,MAAI,WAAW,MAAM;AACnB,UAAM,IAAI,WAAW;AACrB,UAAM,OAAO,EAAE,QAAQ,IAAI,MAAM;AACjC,YAAQ;AAAA,MACN,qBAAqB,EAAE,cAAc,QAAQ,CAAC,CAAC,kBAAkB,EAAE,eAAe,QAAQ,CAAC,CAAC,KAAK,IAAI,GAAG,EAAE,aAAa,QAAQ,CAAC,CAAC;AAAA,IACnI;AAAA,EACF;AAEA,QAAM,SAAS,KAAK;AAAA,IAClB;AAAA,IACA,SAAS,MAAM,SAAS,WAAW,YAAY,SAAS,WAAW,aAAa;AAAA,EAClF;AACA,UAAQ;AAAA,IACN;AAAA,iBAAoB,WAAW,YAAY,MAAM,oBAAoB,WAAW,aAAa,MAAM,cAAc,MAAM;AAAA;AAAA,EACzH;AAEA,MAAI,oBAAoB,WAAW,YAAY,SAAS,GAAG;AACzD,YAAQ,KAAK,CAAC;AAAA,EAChB;AACF;AAEA,SAAS,iBAAiB,GAAqB;AAC7C,QAAM,WAAgB,aAAQ,CAAC;AAC/B,MAAI;AACF,UAAM,WAAO,0BAAS,QAAQ;AAC9B,QAAI,KAAK,YAAY,GAAG;AACtB,YAAM,cAAU,6BAAY,QAAQ;AACpC,aAAO,QACJ,OAAO,CAAC,MAAM,wBAAwB,KAAK,CAAC,CAAC,EAC7C,IAAI,CAAC,MAAW,UAAK,UAAU,CAAC,CAAC;AAAA,IACtC;AACA,WAAO,CAAC,QAAQ;AAAA,EAClB,QAAQ;AACN,WAAO,CAAC,QAAQ;AAAA,EAClB;AACF;AAEA,SAAS,YAAY,QAA4B;AAC/C,QAAM,QAAkB,CAAC;AACzB,QAAM,cAAc,OAAO,KAAK,OAAO,QAAQ,OAAO;AACtD,QAAM,aAAa,KAAK,IAAI,GAAG,YAAY,IAAI,CAAC,MAAM,EAAE,MAAM,GAAG,SAAS,MAAM;AAChF,QAAM,WAAW;AAEjB,QAAM,KAAK,SAAS,OAAO,QAAQ,MAAM,OAAO,OAAO,KAAK,OAAO,QAAQ,KAAK,SAAS;AACzF,QAAM;AAAA,IACJ,KAAK,SAAS,OAAO,UAAU,CAAC,KAAK,OAAO,SAAS,QAAQ,CAAC,KAAK,MAAM,SAAS,QAAQ,CAAC,KAAK,MAAM,SAAS,QAAQ,CAAC,KAAK,MAAM,SAAS,QAAQ,CAAC,KAAK,MAAM,SAAS,QAAQ,CAAC;AAAA,EACpL;AAEA,QAAM,UAAU,aAAa,KAAK,WAAW,KAAK;AAClD,QAAM,KAAK,OAAO,SAAS,OAAO,OAAO,CAAC;AAE1C,aAAW,QAAQ,aAAa;AAC9B,UAAM,IAAI,OAAO,QAAQ,QAAQ,IAAI;AAErC,UAAM,kBAAkB,OAAO,MAAM,OAAO,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,OAAO,IAAI,KAAK,IAAI,EAAE;AACvF,QAAI,oBAAoB,GAAG;AACzB,YAAM;AAAA,QACJ,KAAK,KAAK,OAAO,UAAU,CAAC,KAAK,KAAK,SAAS,QAAQ,CAAC,KAAK,KAAK,SAAS,QAAQ,CAAC,KAAK,KAAK,SAAS,QAAQ,CAAC,KAAK,KAAK,SAAS,QAAQ,CAAC,KAAK,KAAK,SAAS,QAAQ,CAAC;AAAA,MAC1K;AAAA,IACF,OAAO;AACL,YAAM;AAAA,QACJ,KAAK,KAAK,OAAO,UAAU,CAAC,KAAK,EAAE,KAAK,QAAQ,CAAC,EAAE,SAAS,QAAQ,CAAC,KAAK,EAAE,IAAI,QAAQ,CAAC,EAAE,SAAS,QAAQ,CAAC,KAAK,EAAE,IAAI,QAAQ,CAAC,EAAE,SAAS,QAAQ,CAAC,KAAK,EAAE,IAAI,QAAQ,CAAC,EAAE,SAAS,QAAQ,CAAC,KAAK,EAAE,IAAI,QAAQ,CAAC,EAAE,SAAS,QAAQ,CAAC;AAAA,MACvO;AAAA,IACF;AAAA,EACF;AAEA,MAAI,OAAO,QAAQ,QAAQ;AACzB,UAAM,IAAI,OAAO,QAAQ;AACzB,UAAM;AAAA,MACJ,KAAK,SAAS,OAAO,UAAU,CAAC,MAAM,EAAE,OAAO,KAAM,QAAQ,CAAC,EAAE,SAAS,QAAQ,CAAC,MAAM,EAAE,MAAM,KAAM,QAAQ,CAAC,EAAE,SAAS,QAAQ,CAAC,MAAM,EAAE,MAAM,KAAM,QAAQ,CAAC,EAAE,SAAS,QAAQ,CAAC;AAAA,IACtL;AAAA,EACF;AAEA,QAAM,eAAe,OAAO,WAAW,KAAM,QAAQ,CAAC;AACtD,QAAM,UAAU,OAAO,YAAY,IAAI,IAAI,OAAO,UAAU,QAAQ,CAAC,CAAC,KAAK;AAC3E,QAAM,KAAK,EAAE;AACb,QAAM;AAAA,IACJ,eAAe,OAAO,QAAQ,QAAQ,IAAI,OAAO,QAAQ,KAAK,YAAY,OAAO,gBAAgB,WAAW;AAAA,EAC9G;AAEA,QAAM,kBAAkB,OAAO,MAAM,OAAO,CAAC,MAAM,EAAE,cAAc,MAAM;AACzE,MAAI,gBAAgB,SAAS,GAAG;AAC9B,UAAM,eAAe,CAAC,GAAG,IAAI,IAAI,gBAAgB,QAAQ,CAAC,MAAM,EAAE,YAAa,CAAC,CAAC;AACjF,UAAM,KAAK,EAAE;AACb,UAAM;AAAA,MACJ,oBAAoB,gBAAgB,MAAM,IAAI,OAAO,QAAQ,KAAK;AAAA,IACpE;AACA,eAAW,OAAO,aAAa,MAAM,GAAG,CAAC,GAAG;AAC1C,YAAM,KAAK,SAAS,GAAG,EAAE;AAAA,IAC3B;AACA,QAAI,aAAa,SAAS,GAAG;AAC3B,YAAM,KAAK,eAAe,aAAa,SAAS,CAAC,OAAO;AAAA,IAC1D;AAAA,EACF;AAEA,SAAO,MAAM,KAAK,IAAI;AACxB;AAIA,eAAe,yBAAyB,YAAyC;AAC/E,MAAI;AACF,UAAM,MAAM,MAAM,aAAa,YAAY,YAAY,GAAG;AAC1D,UAAM,UAAU,eAAe,GAAG;AAElC,QAAI,CAAC,WAAW,OAAO,QAAQ,YAAY,YAAY;AACrD,cAAQ,MAAM,mDAAmD;AACjE,UAAI,SAAS;AACX,cAAM,OAAO,OAAO,KAAK,OAAiB,EACvC,MAAM,GAAG,CAAC,EACV,KAAK,IAAI;AACZ,gBAAQ,MAAM,UAAU,OAAO,OAAO,GAAG,OAAO,iBAAiB,IAAI,OAAO,EAAE,EAAE;AAAA,MAClF;AACA,cAAQ;AAAA,QACN;AAAA;AAAA;AAAA,MACF;AACA,cAAQ,KAAK,CAAC;AAAA,IAChB;AAEA,WAAO;AAAA,EACT,SAAS,KAAK;AACZ,UAAM,MAAM,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAC3D,QACE,8FAA8F;AAAA,MAC5F;AAAA,IACF,GACA;AACA,YAAM,MAAW,aAAQ,UAAU;AACnC,cAAQ,MAAM,wEAAwE;AACtF,UAAI,QAAQ,SAAS,QAAQ,QAAQ;AACnC,gBAAQ;AAAA,UACN;AAAA,QACF;AAAA,MACF,OAAO;AACL,gBAAQ,MAAM,mDAAmD;AAAA,MACnE;AACA,cAAQ,MAAM;AAAA,IAChB;AACA,YAAQ,MAAM,0BAA0B,GAAG;AAC3C,YAAQ,KAAK,CAAC;AAAA,EAChB;AACF;AAEA,eAAe,WAAW,WAAoB,YAA4C;AAExF,MAAI,cAAc,WAAW,SAAS,GAAG;AACvC,UAAM,mBAAmB,UAAU;AAAA,EACrC;AAGA,MAAI,WAAW;AACb,UAAM,aAAkB,aAAQ,QAAQ,IAAI,GAAG,SAAS;AACxD,UAAM,WAAO,0BAAS,YAAY,EAAE,gBAAgB,MAAM,CAAC;AAC3D,QAAI,CAAC,MAAM,OAAO,GAAG;AACnB,cAAQ,MAAM,0BAA0B,UAAU,EAAE;AACpD,cAAQ,KAAK,CAAC;AAAA,IAChB;AACA,YAAQ,MAAM,kCAAkC,UAAU,EAAE;AAC5D,WAAO,yBAAyB,UAAU;AAAA,EAC5C;AAGA,QAAM,QAAQ,WAAW,QAAQ,IAAI,CAAC;AACtC,MAAI,OAAO;AACT,YAAQ,MAAM,oCAAoC,KAAK,EAAE;AACzD,WAAO,yBAAyB,KAAK;AAAA,EACvC;AAGA,QAAM,EAAE,WAAW,IAAI,MAAM,OAAO,aAAa;AACjD,SAAO,IAAI,WAAW;AACxB;AAIA,SAAS,cAAc,MAKrB;AACA,MAAI;AACJ,MAAI;AACJ,MAAI,aAAuB,CAAC;AAC5B,QAAM,QAAkB,CAAC;AAEzB,WAAS,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;AACpC,UAAM,MAAM,KAAK,CAAC;AAClB,QAAI,QAAQ,cAAc,QAAQ,cAAc,QAAQ,gBAAgB;AACtE,UAAI,IAAI,KAAK,KAAK,QAAQ;AACxB,gBAAQ,MAAM,UAAU,GAAG,mBAAmB;AAC9C,gBAAQ,KAAK,CAAC;AAAA,MAChB;AACA,YAAM,QAAQ,KAAK,EAAE,CAAC;AACtB,UAAI,QAAQ,WAAY,cAAa;AAAA,eAC5B,QAAQ,WAAY,aAAY;AAAA;AAEvC,qBAAa,MACV,MAAM,GAAG,EACT,IAAI,CAAC,MAAM,EAAE,KAAK,CAAC,EACnB,OAAO,OAAO;AAAA,IACrB,WAAW,IAAI,WAAW,IAAI,GAAG;AAC/B,UAAI,CAAC,YAAY,IAAI,GAAG,GAAG;AACzB,gBAAQ,MAAM,iBAAiB,GAAG,EAAE;AACpC,gBAAQ,KAAK,CAAC;AAAA,MAChB;AAAA,IACF,OAAO;AACL,YAAM,KAAK,GAAG;AAAA,IAChB;AAAA,EACF;AAEA,SAAO,EAAE,YAAY,WAAW,YAAY,MAAM;AACpD;AAIA,eAAe,eAAe,MAAgB;AAC5C,QAAM,EAAE,YAAY,WAAW,YAAY,MAAM,IAAI,cAAc,IAAI;AAEvE,MAAI,MAAM,WAAW,GAAG;AACtB,YAAQ,MAAM,mCAAmC;AACjD,YAAQ,KAAK,CAAC;AAAA,EAChB;AAEA,QAAM,YAAsB,CAAC;AAC7B,aAAW,KAAK,OAAO;AACrB,cAAU,KAAK,GAAG,iBAAiB,CAAC,CAAC;AAAA,EACvC;AAEA,MAAI,UAAU,WAAW,GAAG;AAC1B,YAAQ,MAAM,4BAA4B;AAC1C,YAAQ,KAAK,CAAC;AAAA,EAChB;AAEA,QAAM,UAAU,MAAM,WAAW,WAAW,UAAU;AACtD,QAAM,UAAwB,CAAC;AAE/B,MAAI;AACF,eAAW,YAAY,WAAW;AAChC,UAAI;AACF,cAAM,MAAM,MAAM,aAAkB,aAAQ,QAAQ,GAAG,YAAY,GAAG;AACtE,cAAM,aAAyB,IAAI,SAAS,WAAW,IAAI,WAAW,IAAI,UAAU;AAEpF,YAAI,CAAC,WAAW,YAAY,CAAC,WAAW,WAAW,CAAC,WAAW,SAAS;AACtE,kBAAQ;AAAA,YACN,UAAU,QAAQ;AAAA,UACpB;AACA;AAAA,QACF;AAGA,YAAI;AAKJ,YAAI,IAAI,iBAAiB;AAEvB,4BAAkB,OAAO,OAAO,OAAO;AACrC,kBAAM,EAAE,QAAAC,SAAQ,MAAM,YAAY,IAAI,MAAM,QAAQ,UAAU,YAAY;AACxE,qBAAO,IAAI,gBAAgB,OAAO,EAAE;AAAA,YACtC,CAAC;AACD,mBAAO,EAAE,QAAQA,QAAO,QAAQ,MAAMA,QAAO,QAAQ,YAAY;AAAA,UACnE;AAAA,QACF,WAAW,QAAQ,YAAY,WAAW,QAAQ,GAAG;AAEnD,4BAAkB,OAAO,UAAU;AACjC,kBAAM,EAAE,QAAAA,SAAQ,KAAK,IAAI,MAAM,QAAQ,UAAU,YAAY;AAC3D,qBAAO,QAAQ,QAAQ,WAAW,UAAU,KAAK;AAAA,YACnD,CAAC;AACD,mBAAO,EAAE,QAAQA,SAAQ,KAAK;AAAA,UAChC;AAAA,QACF,OAAO;AACL,kBAAQ;AAAA,YACN,uBAAuB,QAAQ;AAAA,UACjC;AACA,4BAAkB,OAAO,WAAW,EAAE,QAAQ,MAAM;AAAA,QACtD;AAEA,cAAM,SAAS,MAAM,QAAQ,YAAY,iBAAiB,OAAO;AACjE,gBAAQ,KAAK,MAAM;AAEnB,gBAAQ,IAAI,OAAO,YAAY,MAAM,IAAI,IAAI;AAAA,MAC/C,SAAS,KAAK;AACZ,gBAAQ;AAAA,UACN,sBAAsB,QAAQ,KAAK,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,CAAC;AAAA,QACrF;AAAA,MACF;AAAA,IACF;AAEA,QAAI,cAAc,QAAQ,SAAS,GAAG;AACpC,YAAM,SAAS,QAAQ,WAAW,IAAI,QAAQ,CAAC,IAAI;AACnD,YAAM,YAAiB,aAAa,aAAQ,UAAU,CAAC;AACvD,gBAAM,uBAAM,WAAW,EAAE,WAAW,KAAK,CAAC;AAC1C,gBAAM,gBAAAC,WAAe,YAAY,KAAK,UAAU,QAAQ,MAAM,CAAC,GAAG,OAAO;AACzE,cAAQ,IAAI,oBAAoB,UAAU,EAAE;AAAA,IAC9C;AAAA,EACF,UAAE;AACA,UAAM,QAAQ,SAAS,EAAE,MAAM,MAAM;AAAA,IAAC,CAAC;AAAA,EACzC;AAGA,MAAI,QAAQ,WAAW,GAAG;AACxB,YAAQ,KAAK,CAAC;AAAA,EAChB;AACF;AAEA,KAAK,EAAE,MAAM,CAAC,QAAQ;AACpB,UAAQ,MAAM,GAAG;AACjB,UAAQ,KAAK,CAAC;AAChB,CAAC;","names":["import_node_fs","round","readFileAsync","result","writeFileAsync"]}
package/dist/cli.js CHANGED
@@ -2,7 +2,7 @@
2
2
  import {
3
3
  evalCompare,
4
4
  runEval
5
- } from "./chunk-OHZX3VFE.js";
5
+ } from "./chunk-F2AALK5U.js";
6
6
 
7
7
  // src/cli.ts
8
8
  import { readdirSync, statSync } from "fs";
@@ -126,6 +126,21 @@ Compare: baseline (${comparison.baseline.id.slice(0, 8)}) -> candidate (${compar
126
126
  ` ${name.padEnd(maxNameLen)} ${s.baselineMean.toFixed(3).padStart(8)} ${s.candidateMean.toFixed(3).padStart(9)} ${(sign + s.delta.toFixed(3)).padStart(8)} ${(sign + s.deltaPercent.toFixed(1) + "%").padStart(6)}`
127
127
  );
128
128
  }
129
+ if (comparison.timing) {
130
+ const t = comparison.timing;
131
+ const sign = t.delta > 0 ? "+" : "";
132
+ console.log(
133
+ `
134
+ Timing: baseline ${(t.baselineMean / 1e3).toFixed(2)}s -> candidate ${(t.candidateMean / 1e3).toFixed(2)}s (${sign}${t.deltaPercent.toFixed(1)}%)`
135
+ );
136
+ }
137
+ if (comparison.cost) {
138
+ const c = comparison.cost;
139
+ const sign = c.delta > 0 ? "+" : "";
140
+ console.log(
141
+ ` Cost: baseline $${c.baselineTotal.toFixed(2)} -> candidate $${c.candidateTotal.toFixed(2)} (${sign}${c.deltaPercent.toFixed(1)}%)`
142
+ );
143
+ }
129
144
  const stable = Math.max(
130
145
  0,
131
146
  baseline.items.length - comparison.regressions.length - comparison.improvements.length
@@ -176,6 +191,12 @@ function formatTable(result) {
176
191
  );
177
192
  }
178
193
  }
194
+ if (result.summary.timing) {
195
+ const t = result.summary.timing;
196
+ lines.push(
197
+ ` ${"Timing".padEnd(maxNameLen)} ${(t.mean / 1e3).toFixed(2).padStart(colWidth)}s ${(t.p50 / 1e3).toFixed(2).padStart(colWidth)}s ${(t.p95 / 1e3).toFixed(2).padStart(colWidth)}s`
198
+ );
199
+ }
179
200
  const durationSec = (result.duration / 1e3).toFixed(1);
180
201
  const costStr = result.totalCost > 0 ? `$${result.totalCost.toFixed(2)}` : "$0.00";
181
202
  lines.push("");