Mexela first-party proxy timing kit, 11 October 2026

Files: benchmark.php, observations.json. The original runner SHA-256 is a16ea4dd1007bdd1efcd4e009473af4ff3bc14040c1a8eef5dc773190ad97ce4.

Requires PHP CLI with cURL. Only the owned HTTPS checker is requested. Provide private JSON on stdin through your approved private-input mechanism, never argv or shell history. Fields: proxies (1 to 5 public IPv4 HTTP assignments, each host/port=80/user/password), rounds (2 to 20), interval_seconds (1 to 30). Supply your own authorized private configuration.

Run: php benchmark.php
Supply the JSON through stdin. The observed run used 20 rounds and 30-second pauses, 20 direct controls and 100 proxy attempts. Fresh connections, five-second connection and ten-second total budgets, normal TLS validation, no followed redirects or persistent cookies. All requests and failures are retained. p95 is nearest rank on qualified successful total times; failed durations stay in rows and denominators.

This is a dated bounded sample, not continuous uptime, an SLA or a causal product comparison. Physical geography, throughput, leaks, anonymity and third-party destinations were not measured. Never publish endpoints, credentials or customer data.

Runner/data/diagram copyright 2026 Mexela; attribution to Mexela and a link to https://mexela.com/blog/proxy-performance-case-study/ are requested when citing the case.
