# Protocol: VPN/censorship event study (cloud-safe)

**Frozen before measured collection · 17 November 2026 · UTC**  
**Slug:** `vpn-censorship-event-study`  
**Duration:** ≤7 days (single cloud measurement session planned; no multi-week crawl)  
**Vantage:** Cursor-hosted USA cloud VM. Freeze note recorded AWS/`IAD`; replacement-agent collection (17 Nov 2026) observed Cloudflare colo `EWR` with `loc=US`. Not Kenya PC.

## Research question

When independent organisations document VPN blocking or internet-censorship events, what can a USA cloud vantage still observe about the same named targets’ DNS, TCP/TLS, and HTTPS fetchability — and how should document-review claims be separated from live measurements?

This study does **not** claim to reproduce in-country blocking from a USA datacenter. USA success is expected for many targets that remain blocked only on specific national networks. The measurement layer is a control and documentation check, not a censorship map of Russia, Kazakhstan, Iran, Myanmar, or Uganda.

## Layer separation (mandatory labels)

Every claim in the article and datasets must be tagged:

| Label | Meaning |
|---|---|
| `DOCUMENT` | Sourced from named public reports (OONI, Freedom House, Access Now, Filterwatch, Human Rights Myanmar, etc.). Not re-measured inside the affected country in this study. |
| `MEASURED_CLOUD_US` | Collected from this USA cloud VM with the frozen cloud-direct collector. |
| `MEASURED_PROXYSELLER_RESIDENTIAL` | Collected through ProxySeller residential HTTP and/or SOCKS5h using injected Runtime Secrets. Values never printed or committed. |
| `BLOCKED_PENDING_SECRETS` | Historical label used before secrets injected (initial cloud-direct session). Superseded once `PROXYSELLER_*` secrets were present. |
| `NOT_RUN` | CometVPN tunnel conditions. CometVPN is VPN-only; no HTTP/SOCKS use. No tunnel claim without a proven working VPN path. |

## Document-review sample (frozen events)

Purposive sample of five well-documented events. Not a census of global censorship.

| ID | Event (short) | Primary sources | Documented mechanism (as reported) | Named target class for our probes |
|---|---|---|---|---|
| E1 | Russia VPN protocol disruption (2023–2024, fluctuating) | OONI *Censorship Chronicles* Russia report (9 Dec 2024) | TSPU / DPI; temporary and regional OpenVPN, IKEv2, WireGuard, IPsec disruption; media + circumvention pressure | Circumvention-tool *websites* (control reachability from US); protocol blocks themselves are DOCUMENT-only |
| E2 | Kazakhstan TLS interference + circumvention-site blocks (Jun 2023–Jun 2024) | OONI Kazakhstan report (19 Sep 2024) | TLS interference / MITM certificates; ≥73 circumvention-tool websites blocked in-country; Tor/Psiphon tools often still reachable | Domains named as blocked circumvention *websites* |
| E3 | Iran targeted VPN/protocol disruption (Dec 2023–Feb 2024; Aug 2024 follow-ups) | Filterwatch network monitors (Apr 2024; Sep 2024) | Protocol-level disruption (e.g. OpenVPN, Shadowsocks-family reports); policy crackdown on circumvention | Circumvention websites as US control; protocol disruption DOCUMENT-only |
| E4 | Myanmar nationwide VPN restrictions (from ~30 May 2024) | Access Now; Freedom House FoTN 2024/2025; Human Rights Myanmar *Great Firewall of Myanmar* | DPI / nationwide VPN app and site restrictions; social platforms also targeted | VPN provider websites + Psiphon |
| E5 | Uganda election-period social + VPN website restrictions (Jan 2021) | OONI Uganda election post (2021); Freedom House FoTN 2024 historical context | Ordered blocking of social platforms and ~100 VPN services; OONI noted website anomalies (`connection_reset`) for several VPN domains | Historical DOCUMENT event; same frozen website list measured from US as contemporary control |

Protocol blocks of WireGuard/OpenVPN *tunnels inside* those countries are **DOCUMENT-only**. This cloud VM cannot ethically or technically substitute for in-country OONI Probe vantage points.

## Measurement sample (frozen targets)

### Primary circumvention-website list (12)

All are publicly marketed landing pages repeatedly named in circumvention-blocking literature (OONI Explorer circumvention charts / Kazakhstan report examples / OONI Run VPN list). Order is fixed for both passes:

1. `https://nordvpn.com/`
2. `https://www.expressvpn.com/`
3. `https://protonvpn.com/`
4. `https://surfshark.com/`
5. `https://mullvad.net/`
6. `https://psiphon.ca/`
7. `https://www.torproject.org/`
8. `https://www.tunnelbear.com/`
9. `https://www.getoutline.org/`
10. `https://www.privateinternetaccess.com/`
11. `https://www.hotspotshield.com/`
12. `https://www.cyberghostvpn.com/`

### Documentation URLs (5) — fetchability of the event record itself

Frozen HTML list used by both collectors (order fixed):

1. `https://ooni.org/post/2024-russia-report/`
2. `https://ooni.org/reports/circumvention/` — **actual frozen URL.** Kazakhstan’s primary DOCUMENT citation remains the English PDF in `events-catalogue.json` (`https://ooni.org/documents/OONI-Report-Kazakhstan-2024-English.pdf`). An earlier draft HTML slug (`/post/2024-kazakhstan-report/`) was not used in measured collection.
3. `https://filter.watch/english/2024/04/02/network-monitor-feb-2024-targeted-vpn-protocol/`
4. `https://www.accessnow.org/myanmar-vpn-ban/`
5. `https://ooni.org/post/2021-uganda-general-election-blocks-and-outage/`

### Controls (3)

1. `https://example.com/` — should succeed
2. `https://www.wikipedia.org/` — should succeed
3. `https://nxdomain-fpt-vpn-censorship-study.invalid/` — expected DNS failure

## Conditions

| Condition ID | Path | Status |
|---|---|---|
| `cloud-direct-us` | No HTTP/SOCKS proxy; default VM egress | **RUN** (`2026-11-17T15:07:32Z`, colo `EWR`) |
| `proxyseller-residential` | HTTP CONNECT + SOCKS5h via ProxySeller residential Runtime Secrets (`PROXYSELLER_HOST`, `PROXYSELLER_HTTP_PORT`, `PROXYSELLER_SOCKS_PORT`, `PROXYSELLER_USER`, `PROXYSELLER_PASS`; never printed) | **RUN** (`2026-11-17T17:41:34Z`; Cloudflare via proxy `loc=BR`, colo `GRU`) |
| `cometvpn-tunnel` | CometVPN WireGuard/app tunnel | **NOT_RUN** — product/affiliate context only; no fake tunnel claims |

**Amendment (17 Nov 2026, same frozen URL list):** Initial freeze marked residential `BLOCKED_PENDING_SECRETS`. After Runtime Secrets injected on a subsequent agent, residential HTTP/SOCKS5h legs ran without changing DOCUMENT claims or inventing credentials.

## Collector rules

- Client (cloud-direct): Python 3 stdlib + OpenSSL/`dig` subprocesses as recorded in `scripts/collect_access.py`.
- Client (residential): `curl` via private `--config` proxy-user file + `--location`, as recorded in `scripts/collect_residential.py` (credentials never on argv; never printed).
- Per URL (cloud-direct), record: DNS A/AAAA outcome classes; TCP:443 connect ms; TLS handshake outcome; leaf cert subject CN/O (org string only, truncate); HTTPS GET status, final URL host (not full query), body SHA-256 of first 64 KiB, HTML `<title>` text truncated to 120 chars, elapsed ms.
- Per URL (residential), record: HTTPS GET through proxy (status, final host, body SHA-256 of first 64 KiB, title, timings, ssl_verify_result). Local dig/TCP/TLS are not the residential path.
- Timeouts: DNS 5s; TCP 8s; TLS 8s; HTTP 15s (cloud); curl connect 12s / max-time 25s (residential).
- Two passes (`pass-1`, `pass-2`) per condition; same URL order. Residential runs both HTTP and SOCKS5h.
- Do not solve CAPTCHAs, create accounts, download VPN installers, or attempt tunnel establishment to third-party VPN networks.
- Do not print or commit ProxySeller/CometVPN secrets, raw residential endpoints, or unredacted cloud public IPs in public files.
- Public IP: keep only salted opaque labels + ASN/colo notes (`EWR`/US for the published collection) in public JSON.

## Public vs private evidence

| Public (`evidence/`, `public/research/...`) | Private (`private/`, gitignored) |
|---|---|
| Outcome classes, statuses, timings, title snippets, body hashes, cert org strings | Raw IPs, full cert PEMs, proxy endpoints, any secret material |
| Event catalogue with DOCUMENT citations | Unredacted curl verbose logs if retained |

## Interpretation limits

- USA reachability ≠ absence of blocking elsewhere.
- Documented in-country TLS resets cannot be “confirmed” from USA success.
- Website reachability ≠ VPN tunnel usability inside a censored network.
- Affiliate relationships (ProxySeller, CometVPN) do not alter measurement outcomes.
- Residential website reachability is not VPN tunnel usability and is not in-country censorship measurement.

## Funding / affiliates

FreePolicyTools is a ProxySeller and CometVPN affiliate. Links: `/proxyseller/`, `/cometvpn/`. No vendor sponsored this measurement session. CometVPN was not used as HTTP/SOCKS.
