Direct-download PDF tool benchmark from April 28, 2026

Study scope

On April 28, 2026, we ran the public direct-download tool set through the repo’s existing npm run benchmark:tools harness.

This study measured the built Astro app against the mock Stirling contract with:

• 20 live direct-download tools
• 3 runs per tool
• a fixed 75ms mock upstream delay
• in-memory Postgres, Redis, and S3-compatible storage

This is first-party runtime data from the current repository. It is useful for tracking app-layer overhead and regression drift, but it does not represent full Stirling PDF output-quality or real-world processing time.

Key findings

• 18 of the 20 direct-download tools posted median times between 92.22ms and 97.32ms.
• jpg-to-pdf (109.37ms p50) and merge-pdf (111.32ms p50) sat above the main cluster.
• merge-pdf showed the widest spread, with a 1,974.18ms max run that pulled its average to 729.41ms.
• pdf-to-excel also showed a meaningful average-versus-median gap: 109.71ms average versus 93.60ms p50.

Notable rows

Tool	Avg	P50	Min	Max
split-pdf	92.53ms	92.22ms	90.99ms	94.38ms
compress-pdf	103.04ms	94.21ms	89.48ms	125.43ms
pdf-to-word	99.43ms	97.32ms	90.27ms	110.71ms
jpg-to-pdf	107.18ms	109.37ms	95.61ms	116.56ms
merge-pdf	729.41ms	111.32ms	102.73ms	1,974.18ms

Download the raw data

• Direct-download benchmark CSV

Methodology notes

• The benchmark comes from the repo’s shared npm run benchmark:tools harness.
• Each tool is exercised through the public Astro route, not by calling the upstream mock directly.
• The script reports average, median, minimum, and maximum so latency outliers are visible instead of being smoothed away.

Press-ready angles

• Most direct-download PDF flows are currently clustering in the low-90ms median band at the app layer.
• merge-pdf and jpg-to-pdf are the first direct-download flows to watch as traffic and payload size increase.
• Median plus max is a more honest operational story than average alone for public PDF tools.