@tangle-network/agent-bench 0.4.0 → 0.4.3

package/scripts/verify-packed-consumer.mjs

@@ -52,11 +52,58 @@ async function resolveRuntimePackage(packDir) {
 
 try {
   const packDir = path.join(scratch, 'pack')
+  const unpackDir = path.join(scratch, 'unpack')
   const runtimePackDir = path.join(scratch, 'runtime-pack')
   const consumerDir = path.join(scratch, 'consumer')
+  const terminalBenchVenv = path.join(scratch, 'terminal-bench-venv')
+  const terminalBenchBinDir = path.join(terminalBenchVenv, 'bin')
   await mkdir(packDir)
+  await mkdir(unpackDir)
   await mkdir(runtimePackDir)
   await mkdir(consumerDir)
+  await mkdir(terminalBenchBinDir, { recursive: true })
+  await writeFile(path.join(terminalBenchVenv, 'package.json'), '{"type":"module"}\n')
+  await writeFile(
+    path.join(terminalBenchBinDir, 'python'),
+    String.raw`#!/usr/bin/env node
+const rows = [{
+  id: 'installed-absolute-venv',
+  instruction: 'prove the installed path',
+  task_dir: '/tmp/terminal-bench-task',
+  solution: 'echo ok\n',
+}]
+process.stdout.write(JSON.stringify(rows) + '\n')
+`,
+    { mode: 0o755 },
+  )
+  await writeFile(
+    path.join(terminalBenchBinDir, 'tb'),
+    String.raw`#!/usr/bin/env node
+import { mkdirSync, writeFileSync } from 'node:fs'
+import { join } from 'node:path'
+
+const args = process.argv.slice(2)
+function value(flag) {
+  const index = args.indexOf(flag)
+  if (index === -1 || index + 1 >= args.length) throw new Error('missing ' + flag)
+  return args[index + 1]
+}
+
+const outputPath = value('--output-path')
+const runId = value('--run-id')
+const taskId = value('-t')
+const reportDir = join(outputPath, runId)
+mkdirSync(reportDir, { recursive: true })
+writeFileSync(
+  join(reportDir, 'results.json'),
+  JSON.stringify({
+    resolved_ids: [taskId],
+    results: [{ task_id: taskId, is_resolved: true, parser_results: { installedVenv: true } }],
+  }),
+)
+`,
+    { mode: 0o755 },
+  )
 
   // Build explicitly so verification cannot inherit a machine-level
   // ignore-scripts setting and accidentally pack stale or missing output.
@@ -70,21 +117,14 @@ try {
     throw new Error(`expected one packed agent-bench tarball, found ${packedFiles.length}`)
   }
   const tarball = path.join(packDir, packedFiles[0])
+  await run('tar', ['-xzf', tarball, '-C', unpackDir], benchDir)
   const runtimePackage = await resolveRuntimePackage(runtimePackDir)
-  const manifest = JSON.parse(await readFile(path.join(benchDir, 'package.json'), 'utf8'))
-  const devDependencies = manifest.devDependencies
-  if (
-    typeof devDependencies?.['@types/node'] !== 'string' ||
-    typeof devDependencies.typescript !== 'string' ||
-    typeof devDependencies.tsx !== 'string' ||
-    !devDependencies['@types/node'] ||
-    !devDependencies.typescript ||
-    !devDependencies.tsx
-  ) {
-    throw new Error(
-      'package verification requires @types/node, typescript, and tsx devDependencies',
-    )
-  }
+  const packedManifest = JSON.parse(
+    await readFile(path.join(unpackDir, 'package', 'package.json'), 'utf8'),
+  )
+  const nodeTypes = requiredPackedDevelopmentDependency(packedManifest, '@types/node')
+  requiredPackedDevelopmentDependency(packedManifest, 'typescript')
+  const tsx = requiredPackedDevelopmentDependency(packedManifest, 'tsx')
   const publicTsconfig = JSON.parse(
     await readFile(path.join(benchDir, 'tsconfig.public.json'), 'utf8'),
   )
@@ -105,9 +145,9 @@ try {
             : {}),
         },
         devDependencies: {
-          '@types/node': devDependencies['@types/node'],
+          '@types/node': nodeTypes,
           typescript: TYPESCRIPT_5,
-          tsx: devDependencies.tsx,
+          tsx,
         },
       },
       null,
@@ -122,6 +162,21 @@ try {
     path.join(consumerDir, 'index.mjs'),
     "import { resolveAdapter, runBenchmarks } from '@tangle-network/agent-bench'\nimport { ADAPTERS } from '@tangle-network/agent-bench/adapters'\nimport { createCragAdapter } from '@tangle-network/agent-bench/benchmarks/crag'\n\nif (typeof resolveAdapter !== 'function' || typeof runBenchmarks !== 'function') throw new Error('root exports are not executable')\nif (typeof ADAPTERS !== 'object' || typeof createCragAdapter !== 'function') throw new Error('subpath exports are not executable')\nif (resolveAdapter('crag').name !== 'crag') throw new Error('compiled adapter registry returned the wrong adapter')\nprocess.env.TOOLLM_FIXTURES = '1'\nconst toolLlmTasks = await resolveAdapter('toollm').loadTasks({ limit: 1 })\nif (toolLlmTasks.length !== 1 || toolLlmTasks[0]?.id !== '1') throw new Error('packed ToolLLM fixture loading failed')\n",
   )
+  await writeFile(
+    path.join(consumerDir, 'terminal-bench-absolute-venv.mjs'),
+    `import { createTerminalBenchAdapter } from '@tangle-network/agent-bench/benchmarks/terminal-bench'
+
+const adapter = createTerminalBenchAdapter()
+const tasks = await adapter.loadTasks({ ids: ['installed-absolute-venv'] })
+if (tasks.length !== 1 || tasks[0]?.id !== 'installed-absolute-venv') {
+  throw new Error('packed Terminal-Bench adapter did not load through the absolute venv')
+}
+const score = await adapter.judge(tasks[0], 'echo installed-package-path')
+if (!score.resolved || score.score !== 1) {
+  throw new Error(\`packed Terminal-Bench adapter did not judge through the absolute venv: \${JSON.stringify(score)}\`)
+}
+`,
+  )
   await writeFile(
     path.join(consumerDir, 'tsconfig.json'),
     `${JSON.stringify(
@@ -165,6 +220,10 @@ for name in sorted(expected):
     consumerDir,
   )
   await run('node', ['index.mjs'], consumerDir)
+  await run('node', ['terminal-bench-absolute-venv.mjs'], consumerDir, {
+    ...process.env,
+    TERMINAL_BENCH_VENV: terminalBenchVenv,
+  })
   await run('npm', ['exec', '--', 'tsc', '-p', 'tsconfig.json'], consumerDir)
   const typescript5 = await run('npm', ['exec', '--', 'tsc', '--version'], consumerDir)
   if (typescript5.stdout.trim() !== `Version ${TYPESCRIPT_5}`) {
@@ -228,8 +287,16 @@ for name in sorted(expected):
     })
   }
   console.log(
-    `packed consumer verified: ${manifest.name}@${manifest.version} with @tangle-network/agent-runtime@${runtimeManifest.version}, TypeScript ${TYPESCRIPT_5} and ${TYPESCRIPT_6}; prepared ${prepareProof.executionPlanDigest}`,
+    `packed consumer verified: ${packedManifest.name}@${packedManifest.version} with @tangle-network/agent-runtime@${runtimeManifest.version}, TypeScript ${TYPESCRIPT_5} and ${TYPESCRIPT_6}, absolute Terminal-Bench venv; prepared ${prepareProof.executionPlanDigest}`,
   )
 } finally {
   await rm(scratch, { recursive: true, force: true })
 }
+
+function requiredPackedDevelopmentDependency(packageJson, name) {
+  const version = packageJson.devDependencies?.[name]
+  if (typeof version !== 'string' || version.length === 0 || version.startsWith('catalog:')) {
+    throw new Error(`packed consumer requires a resolved ${name} development dependency`)
+  }
+  return version
+}

package/scripts/verify-pier-agent.mts

@@ -435,7 +435,7 @@ ${proofArm === 'success' ? "(task / 'src/status.txt').write_text('ready\\nowner=
     workspace: taskWorkspace,
     evaluatorTaskContainer: container,
     limits: {
-      timeoutMs: 60_000,
+      timeoutMs: 180_000,
       maxSteps: 8,
       maxModelCalls: 0,
       maxInputTokens: 0,
@@ -671,12 +671,14 @@ ${proofArm === 'success' ? "(task / 'src/status.txt').write_text('ready\\nowner=
     )
   }
   assertTreeOmits(scratch, 'zero-model-proof')
-  const usage = finalized.receipt.value.usage
+  const usage = finalized.receipt.value.modelSettlement.material.usage
   if (
     usage.modelCalls !== 0 ||
     usage.inputTokens !== 0 ||
     usage.outputTokens !== 0 ||
-    usage.costUsd !== 0 ||
+    usage.cachedInputTokens !== 0 ||
+    usage.reasoningTokens !== 0 ||
+    usage.costUsdNanos !== 0 ||
     finalized.receipt.value.trace.modelCallCount !== 0
   ) {
     throw new Error(`runtime minted nonzero usage for a zero-model run: ${JSON.stringify(usage)}`)
@@ -691,7 +693,7 @@ ${proofArm === 'success' ? "(task / 'src/status.txt').write_text('ready\\nowner=
         modelCalls: usage.modelCalls,
         inputTokens: usage.inputTokens,
         outputTokens: usage.outputTokens,
-        costUsd: usage.costUsd,
+        costUsd: usage.costUsdNanos / 1_000_000_000,
         pierContextUsage: null,
         profileExcludedByVerifier: true,
         container: {

package/src/benchmarks/_harness.test.mts

@@ -4,7 +4,13 @@ import { mkdtemp, readFile, rm, writeFile } from 'node:fs/promises'
 import test from 'node:test'
 import { tmpdir } from 'node:os'
 import { join } from 'node:path'
-import { resolveBenchPython, runStagedJudge } from './_harness'
+import {
+  benchRoot,
+  resolveBenchPython,
+  runStagedJudge,
+  venvBin,
+  venvBinAt,
+} from './_harness'
 
 const digest = (bytes: Uint8Array): `sha256:${string}` =>
   `sha256:${createHash('sha256').update(bytes).digest('hex')}`
@@ -31,6 +37,15 @@ test('resolveBenchPython rejects relative and empty interpreter paths', () => {
   )
 })
 
+test('venv executable paths support package-owned and external environments', () => {
+  assert.equal(venvBinAt('/srv/terminal-bench', 'tb'), join('/srv/terminal-bench', 'bin', 'tb'))
+  assert.equal(
+    venvBinAt('.venv-commit0', 'python'),
+    join(benchRoot, '.venv-commit0', 'bin', 'python'),
+  )
+  assert.equal(venvBin('python'), venvBinAt('.venv', 'python'))
+})
+
 test('runStagedJudge terminates a hung evaluator at its explicit timeout', async () => {
   const started = Date.now()
   await assert.rejects(

package/src/benchmarks/_harness.ts

@@ -82,13 +82,20 @@ export const venvPython = resolveBenchPython()
 /** Interpreter for a NAMED isolated venv (e.g. `.venv-commit0`). Benches whose pip
  *  deps conflict with the shared `.venv` (commit0 downgrades pydantic/sqlalchemy)
  *  get their own venv and pass its python explicitly — keeping the shared one clean. */
-export const venvPythonAt = (venvDir: string): string => join(benchRoot, venvDir, 'bin', 'python')
+export const venvPythonAt = (venvDir: string): string => venvBinAt(venvDir, 'python')
+
+/** Resolve an executable in an isolated venv. Relative venv paths are package-owned;
+ * absolute paths allow installed consumers to keep large environments elsewhere. */
+export function venvBinAt(venvDir: string, name: string): string {
+  return join(resolve(benchRoot, venvDir), 'bin', name)
+}
+
 /** Report/transcript reads are large; 256 MiB matches the SWE harness budget. */
 export const bigBuffer = 1024 * 1024 * 256
 
 /** Path to a named executable inside the bench venv (e.g. `venvBin('tb')`). */
 export function venvBin(name: string): string {
-  return join(benchRoot, '.venv', 'bin', name)
+  return venvBinAt('.venv', name)
 }
 
 /**

package/src/benchmarks/terminal-bench.ts

@@ -25,6 +25,7 @@ import {
   runVenvPython,
   safeRunId,
   stageFile,
+  venvBinAt,
   venvPythonAt,
 } from './_harness'
 import type { BenchmarkAdapter, BenchScore, BenchTask, LoadOptions } from './types'
@@ -34,7 +35,7 @@ import type { BenchmarkAdapter, BenchScore, BenchTask, LoadOptions } from './typ
 // override the env without reloading this module.
 const terminalBenchVenvDir = (): string => process.env.TERMINAL_BENCH_VENV ?? '.venv-terminal-bench'
 const terminalBenchPython = (): string => venvPythonAt(terminalBenchVenvDir())
-const terminalBenchBin = (): string => join(benchRoot, terminalBenchVenvDir(), 'bin', 'tb')
+const terminalBenchBin = (): string => venvBinAt(terminalBenchVenvDir(), 'tb')
 
 // Pinned dataset: the 0.1.1 core set is patched for terminal-bench >=0.2.4 (the
 // installed CLI) and is the published launch task set. name==version is what `tb

package/src/corpus.test.mts

@@ -109,6 +109,8 @@ const baseRec = (attempts: AttemptRecord[], over: Partial<RunRecord> = {}): RunR
   assert.equal(r0?.commitSha, 'abc123')
   assert.equal(r0?.tokenUsage.input, 100, 'real tokens carried (not zeroed)')
   assert.equal(r0?.costUsd, 0.01, 'real cost carried')
+  assert.deepEqual(r0?.costProvenance, { kind: 'observed', usd: 0.01 }, 'reported cost remains observed')
+  assert.equal(r0?.terminalOutcome, 'succeeded', 'failed task quality is not an execution failure')
   assert.equal(r0?.prompt, 'q0', 'verbatim prompt survives the validator')
   assert.equal(r0?.completion, 'alpha', 'verbatim completion survives the validator')
   assert.equal(r0?.outcome.searchScore, 0, 'search split → searchScore from attempt score')
@@ -117,6 +119,17 @@ const baseRec = (attempts: AttemptRecord[], over: Partial<RunRecord> = {}): RunR
   assert.notEqual(r0?.runId, r1?.runId, 'per-attempt runIds are distinct')
 }
 
+// --- attempt error is execution state, not a fabricated zero-quality label ---
+{
+  const { records, unmappable } = await benchRecordToCorpusRecords(
+    baseRec([{ ...measuredAttempt(0, 'partial output', false), error: 'provider disconnected' }]),
+    { commitSha: 'abc123', model: 'gpt-5-2025-08-07' },
+  )
+  assert.equal(unmappable.length, 0)
+  assert.equal(records[0]?.terminalOutcome, 'failed')
+  assert.equal(records[0]?.outcome.searchScore, 0, 'task quality remains independently recorded')
+}
+
 // --- unmeasured economics → unmappable, never forged with phantom zeros ---
 {
   const bare: AttemptRecord = { round: 0, prompt: 'q', output: 'x', valid: true, eventCount: 0, eventTypes: {} }

package/src/corpus.ts

@@ -288,7 +288,11 @@ export async function benchRecordToCorpusRecords(
       commitSha: opts.commitSha,
       wallMs: a.wallMs as number,
       costUsd: a.costUsd as number,
+      // Attempt costs are present only when the worker reported them. Missing
+      // costs exited above rather than becoming a made-up zero.
+      costProvenance: { kind: 'observed', usd: a.costUsd as number },
       tokenUsage: { input: a.tokensIn as number, output: a.tokensOut as number },
+      terminalOutcome: a.error === undefined ? 'succeeded' : 'failed',
       outcome: {
         ...(splitTag === 'holdout' ? { holdoutScore: score } : { searchScore: score }),
         raw: { valid: a.valid === true ? 1 : 0, score },

package/src/profile-coordinates.ts

@@ -14,8 +14,8 @@
  */
 import { readFileSync, readdirSync } from 'node:fs'
 import { join } from 'node:path'
-import type { AgentProfile } from '@tangle-network/sandbox'
-import { defineInlineResource } from '@tangle-network/sandbox'
+import type { AgentProfile } from '@tangle-network/agent-interface'
+import { defineInlineResource } from '@tangle-network/agent-interface'
 
 export interface ProfileCoordinate {
   /** Coordinate id (the COORDINATE= knob value). */

package/src/rollout-ledger/backfill-swe-arena.mts

@@ -141,10 +141,11 @@ interface Ctx {
   stats: BackfillStats
 }
 
-function baseLine(ctx: Ctx): Pick<RolloutLine, 'schema' | 'run_id' | 'provenance'> {
+function baseLine(ctx: Ctx): Pick<RolloutLine, 'schema' | 'run_id' | 'experiment_id' | 'provenance'> {
   return {
     schema: ROLLOUT_SCHEMA,
     run_id: ctx.runId,
+    experiment_id: null,
     provenance: { captured_at: ctx.capturedAt, capture: 'backfill' },
   }
 }
@@ -216,6 +217,7 @@ async function emitCellLines(
       reward,
       reward_source: reward === null ? null : OFFICIAL_JUDGE,
       verdict: judgeVerdict,
+      realness_gated: false,
       metrics: {
         resolved: artifact?.resolved ?? null,
         verify_pass: artifact?.verifyPass ?? null,
@@ -335,6 +337,7 @@ function workerLine(
       reward,
       reward_source: reward === null ? null : `${OFFICIAL_JUDGE}/inherited`,
       verdict: null,
+      realness_gated: false,
       metrics: {
         worker_cwd: cwd,
         session_agent: session?.agent ?? null,
@@ -504,6 +507,7 @@ async function emitProposerLines(ctx: Ctx, lines: RolloutLine[], entries: Rollou
         reward,
         reward_source: reward === null ? null : `${OFFICIAL_JUDGE}/candidate-resolved-fraction`,
         verdict: null,
+        realness_gated: false,
         metrics: {
           resolved_count: entry?.outcome.resolvedCount ?? null,
           instance_count: ctx.instanceCount,

package/src/rollout-ledger/settle-capture.mts

@@ -194,9 +194,12 @@ export function createSettleCapture(opts: SettleCaptureOptions): SettleCapture {
   const log = opts.log ?? (() => {})
   const dbPath = opts.opencodeDb ?? DEFAULT_OPENCODE_DB
 
-  const base = (capturedAt: string): Pick<RolloutLine, 'schema' | 'run_id'> & { provenance: RolloutLine['provenance'] } => ({
+  const base = (
+    capturedAt: string,
+  ): Pick<RolloutLine, 'schema' | 'run_id' | 'experiment_id'> & { provenance: RolloutLine['provenance'] } => ({
     schema: ROLLOUT_SCHEMA,
     run_id: opts.runId,
+    experiment_id: null,
     provenance: { captured_at: capturedAt, capture: 'settle-time' },
   })
 
@@ -232,6 +235,7 @@ export function createSettleCapture(opts: SettleCaptureOptions): SettleCapture {
           reward,
           reward_source: reward === null ? null : OFFICIAL_JUDGE,
           verdict: args.judgeVerdict,
+          realness_gated: false,
           metrics: {
             ...args.metrics,
             ...(args.splitVisibility !== null ? { split_visibility: args.splitVisibility } : {}),
@@ -311,6 +315,7 @@ export function createSettleCapture(opts: SettleCaptureOptions): SettleCapture {
                 reward: v2.reward,
                 reward_source: v2.reward === null && v2.deliveredMatch !== 'unknown' ? null : WORKER_REWARD_SOURCE_V2,
                 verdict: null,
+                realness_gated: false,
                 metrics: {
                   worker_label: worker.label,
                   worker_cwd: worker.cwd,
@@ -400,6 +405,7 @@ export function createSettleCapture(opts: SettleCaptureOptions): SettleCapture {
           reward,
           reward_source: PROPOSER_REWARD_SOURCE_V2,
           verdict: null,
+          realness_gated: false,
           metrics: {
             resolved_count: args.candResolved,
             baseline_resolved_count: args.baselineResolved,

package/src/search-bench/profiles.ts

@@ -17,7 +17,7 @@
  * shape — a `transport:'http'` server at the router's `/v1/search/mcp` endpoint,
  * provider pinned via the `?provider=` query param.
  */
-import type { AgentProfile } from '@tangle-network/sandbox'
+import type { AgentProfile } from '@tangle-network/agent-interface'
 
 export type SearchArm = 'native' | 'off' | { provider: string }
 

package/src/skill-sandbox-smoke.mts

@@ -14,7 +14,8 @@
  *
  * Run: dotenvx run -f ~/company/devops/secrets/.env.keys -- pnpm exec tsx bench/src/skill-sandbox-smoke.mts
  */
-import { Sandbox, defineInlineResource } from '@tangle-network/sandbox'
+import { defineInlineResource } from '@tangle-network/agent-interface'
+import { Sandbox } from '@tangle-network/sandbox'
 
 const must = (k: string): string => {
   const v = process.env[k]

package/src/swe-arena/gepa-seat.mts

@@ -233,7 +233,7 @@ export function innerSmokeJudge(): JudgeConfig<SmokeVerdict, GepaSeatScenario> {
 // ---------------------------------------------------------------------------
 
 export const GEPA_PYTHON_INSTALL_HINT =
-  'install `agent-eval-rpc==0.126.6`, then install ' +
+  'install `agent-eval-rpc==0.131.0`, then install ' +
   '`gepa[full] @ git+https://github.com/gepa-ai/gepa.git@f919db0a622e2e9f9204779b81fe00cc1b2d808f`'
 
 export type GepaMethodFactory = (

package/src/swe-arena/ledger-orphans.test.mts

@@ -7,7 +7,8 @@
  * path improve() rides.
  */
 
-import { mkdtempSync, readFileSync, writeFileSync } from 'node:fs'
+import { execFileSync } from 'node:child_process'
+import { mkdtempSync, readFileSync } from 'node:fs'
 import { tmpdir } from 'node:os'
 import { join } from 'node:path'
 import { describe, expect, it } from 'vitest'
@@ -22,40 +23,41 @@ import {
 
 const ATTRIBUTION = { channel: 'agent', phase: 'search.baseline', model: 'm' }
 
-/** A durable event line in the exact shape cost-ledger.jsonl persists. */
-const line = (record: Record<string, unknown>): string =>
-  `${JSON.stringify({ version: 1, record })}\n`
-
-/** Write a synthetic ledger: one settled pair + one crash-orphaned pending. */
+/**
+ * Write one settled receipt and then terminate a process with a real pending
+ * call. The child uses the public ledger API so this fixture cannot shadow its
+ * private JSONL event format.
+ */
 function writeCrashedLedger(dir: string): void {
-  const settledPending = {
-    status: 'pending',
-    callId: 'settled-1',
-    ...ATTRIBUTION,
-    actor: 'worker:astropy#r0',
-    timestamp: 1_000,
-  }
-  const settledReceipt = {
-    ...settledPending,
-    status: 'settled',
-    inputTokens: 10,
-    outputTokens: 5,
-    costUsd: 0.01,
-    costUnknown: false,
-    actualCostUsd: 0.01,
-  }
-  const orphanPending = {
-    status: 'pending',
-    callId: 'orphan-1',
-    ...ATTRIBUTION,
-    actor: 'worker:xarray#r0',
-    timestamp: 2_000,
-    tags: { cellId: 'pydata__xarray-4687:0' },
-  }
-  writeFileSync(
-    join(dir, 'cost-ledger.jsonl'),
-    line(settledPending) + line(settledReceipt) + line(orphanPending),
-  )
+  const child = `
+    import { createRunCostLedger, fsCampaignStorage } from '@tangle-network/agent-eval/campaign'
+    const runDir = process.env.COST_LEDGER_RUN_DIR
+    if (typeof runDir !== 'string' || runDir.length === 0) throw new Error('missing COST_LEDGER_RUN_DIR')
+    const attribution = { channel: 'agent', phase: 'search.baseline', model: 'm' }
+    const ledger = createRunCostLedger({ storage: fsCampaignStorage(), runDir })
+    await ledger.runPaidCall({
+      ...attribution,
+      callId: 'settled-1',
+      actor: 'worker:astropy#r0',
+      execute: async () => 'settled',
+      receipt: () => ({ model: 'm', inputTokens: 10, outputTokens: 5, actualCostUsd: 0.01 }),
+    })
+    let started
+    const pending = new Promise((resolve) => { started = resolve })
+    void ledger.runPaidCall({
+      ...attribution,
+      callId: 'orphan-1',
+      actor: 'worker:xarray#r0',
+      tags: { cellId: 'pydata__xarray-4687:0' },
+      execute: async () => { started(); return await new Promise(() => {}) },
+      receipt: () => ({ model: 'm', inputTokens: 0, outputTokens: 0, actualCostUsd: 0 }),
+    })
+    await pending
+  `
+  execFileSync(process.execPath, ['--input-type=module', '--eval', child], {
+    cwd: process.cwd(),
+    env: { ...process.env, COST_LEDGER_RUN_DIR: dir },
+  })
 }
 
 const openLedger = (dir: string) =>

package/dist/benchmarks/cadbench.js.map

@@ -1 +0,0 @@
-{"version":3,"sources":[],"sourcesContent":[],"mappings":"","names":[]}

package/dist/benchmarks/cadgenbench.js.map

@@ -1 +0,0 @@
-{"version":3,"sources":[],"sourcesContent":[],"mappings":"","names":[]}

package/dist/benchmarks/types.js.map

@@ -1 +0,0 @@
-{"version":3,"sources":[],"sourcesContent":[],"mappings":"","names":[]}