Controlled static HTTPS GET check, 13 October 2026

Actual study: 6 direct GETs followed by 60 GETs through three previously reserved HTTP proxies.
Target: one owned, versioned 10,227-byte static JSON file. All responses were checked against the pinned SHA-256.
See the JSON for exact UTC dates, versions, per-attempt measurements, nearest-rank p95 summaries and explicit limitations.
Timing includes the complete cURL operation, not only proxy latency. No throughput or long-term uptime conclusion is supported.

To inspect the public reproducer without traffic:
  python -B reproduce-controlled-static-get.py
To run it from a fresh folder on Windows with Python 3 and curl 8.4 or newer:
  python -B reproduce-controlled-static-get.py --execute
Enter only your own authorized reserved HTTP proxy access in hidden prompts. The study's endpoint identities and credentials are not included.
The reproducer issues no more than 66 GETs, never more than four active at once, and refuses to replay an existing attempt.
It uses six-second connection/total timeouts, verified TLS, no retries or redirects, a 64 KiB body bound and fresh cURL processes.
One proxy is tested at a time. Each proxy has two rounds per level: three requests at level 1, three at level 2, four at level 4.
One second separates rounds; five seconds separates cells; the six direct controls have half-second gaps.
The Windows IPv4 TCP observer matches endpoints and process IDs in memory only, sampling about every 10 ms.

Credentials, endpoint addresses, raw bodies and raw error text are not written to result files. Results are local; no collector receives submissions.
Metrics: https://curl.se/docs/manpage.html
