chenglou/pretext · error · Error

${batchReport.message ?? 'keep-all batch failed'}

Error message

${batchReport.message ?? 'keep-all batch failed'}

What it means

After the in-page keep-all probe batch reports back via the POST side channel, runBrowser() inspects batchReport.status. If the page set status to 'error', this wraps batchReport.message (falling back to a generic 'keep-all batch failed'). The root cause lives in the browser page that ran the probe, not in this CLI process.

Solutions

1. Read the wrapped batchReport.message printed in the error for the browser-side root cause.
2. Confirm the dev/probe page server is healthy and re-run; suspect stale tabs/servers before assuming an engine bug (per repo guidance).
3. Re-run with `--timeout=` raised if the message indicates a measurement timeout.
4. If the message points at a specific probe case, debug that case in isolation via the diagnostic /probe page.

const isBatchError = (r: { status?: string }): boolean => r.status === 'error'

try {
  await runBrowser(browser, port)
} catch (e) {
  const msg = e instanceof Error ? e.message : String(e)
  if (msg.includes('keep-all batch failed') || msg.startsWith('keep-all')) {
    console.error(`Probe batch error for ${browser}. Suspect stale tabs/servers first; see message: ${msg}`)
    process.exit(1)
  }
  throw e
}

When it happens

Trigger: The /probe?batch=keep-all page threw during measurement, failed to acquire a measurement, hit a DOM/layout error, or returned an explicit error status in the posted ProbeBatchReport. Possible when the test server didn't serve the probe page, the automation tab was backgrounded/throttled, or a probe case references text the page cannot render.

Common situations: A transient browser-automation hiccup, the dev server not running on the expected port, stale tab/server, or a regression in the probe page itself.

Related errors

• ${batchReport.message ?? 'letter-spacing batch failed'}
• ${batchReport.message ?? 'pre-wrap batch failed'}
• Missing keep-all result for
• Missing letter-spacing result for
• corpus-check exited with status

AI-assisted analysis of chenglou/pretext@ac49b09b7d (2026-08-12). Data as JSON: /api/errors/7af615902f326fa1. Report an issue: GitHub.

Appendix: source

Thrown at scripts/keep-all-check.ts:151

    serverProcess = pageServer.process
    const requestId = `${browser}-${Date.now()}-${Math.random().toString(36).slice(2)}`
    const reportServer = await startPostedReportServer<ProbeBatchReport>(requestId)

    try {
      const url =
        `${pageServer.baseUrl}/probe?batch=keep-all` +
        `&requestId=${encodeURIComponent(requestId)}` +
        `&reportEndpoint=${encodeURIComponent(reportServer.endpoint)}`
      const batchReport = await loadPostedReport(
        session,
        url,
        () => reportServer.waitForReport(null),
        requestId,
        reportBrowser,
        timeoutMs,
      )
      if (batchReport.status === 'error') {
        throw new Error(batchReport.message ?? 'keep-all batch failed')
      }

      const batchResults = batchReport.results ?? []
      const reportsByLabel = new Map(batchResults.map(result => [result.label, result.report]))
      for (const testCase of KEEP_ALL_ORACLE_CASES) {
        if (!caseRunsInBrowser(testCase, browser)) continue
        const report = reportsByLabel.get(testCase.label)
        if (report === undefined) {
          throw new Error(`Missing keep-all result for ${testCase.label}`)
        }
        printCaseResult(browser, testCase, report)
        if (!reportIsExact(report)) ok = false
      }
    } finally {
      reportServer.close()
    }
  } finally {
    session?.close()

View on GitHub (pinned to ac49b09b7d)