# Same Browser Fingerprint Across Direct, HTTP, and SOCKS. Only the Document IPs Moved.

*Original FreePolicyTools research · 17 November 2026 · Amendment A1 (ProxySeller residential legs)*

A fresh Chromium context does not mint a new browser identity—and neither does routing document fetches through a residential HTTP or SOCKS proxy. In our cloud tests, **eighteen isolated Playwright Chromium visits** across three matched network conditions produced **one composite fingerprint hash**, while independently fetched document IP echoes shifted from cloud datacenter labels (`US` / `AS14618 Amazon.com, Inc.`) on direct egress to residential labels (`BR` / `AS28142 DIGITAL DESIGN SERVIÇOS DE TELECOMUNICAÇÕES EIRELI`) on both ProxySeller HTTP and SOCKS paths.

We measured researcher-operated fingerprint surfaces—canvas, WebGL, audio, navigator, screen, timezone, font metrics, storage availability, media-device availability, and WebRTC candidate counts—inside headless Chromium 131 on a Cursor-hosted USA VM. Network identity was recorded only as redacted echo labels and public country/org fields. We did not query commercial “how unique is my browser?” score pages as the primary sample.

Amendment **A1** activates the protocol’s ProxySeller residential comparison legs after Runtime Secrets injection. CometVPN remains VPN-only product context and affiliate CTA material; we did not claim a CometVPN tunnel measurement on this VM.

These are browser-surface stability findings on one automation stack under three researcher-operated egress conditions. They are not a population uniqueness rate, not a ProxySeller inventory or uptime ranking, not a VPN/proxy quality scoreboard, and not evidence of provider logging. For background, read [what browser fingerprinting is](/blog/what-is-browser-fingerprinting/), [why changing your IP does not change your fingerprint](/blog/why-changing-your-ip-doesnt-change-your-browser-fingerprint/), and [does a VPN change your browser fingerprint?](/blog/does-a-vpn-change-your-browser-fingerprint/).

## What we measured

We froze `research/browser-fingerprinting-study/protocol.md` before the first cloud-direct collection on 17 November 2026 (UTC), with a ≤7-day duration cap. Amendment A1 (same day) activated `proxy-us-http` and `proxy-us-socks` after env discovery confirmed component Runtime Secret **names** (values never recorded):

| Secret name | Presence at A1 |
|---|---|
| `PROXYSELLER_HOST` | PRESENT |
| `PROXYSELLER_HTTP_PORT` | PRESENT |
| `PROXYSELLER_SOCKS_PORT` | PRESENT |
| `PROXYSELLER_USER` | PRESENT |
| `PROXYSELLER_PASS` | PRESENT |

Collection used Playwright Chromium **131.0.6778.33**, headless, locale `en-US`, timezone `UTC`, viewport `1360×900`, one new browser context per visit, JavaScript on, ordinary TLS. Matched A1 collection windows (UTC):

| Condition | Run id | First visit | Last visit |
|---|---|---|---|
| `proxy-us-http` | `20261117T174906Z-proxy-us-http` | 17:49:16 | 17:50:13 |
| `proxy-us-socks` | `20261117T175129Z-proxy-us-socks` | 17:51:39 | 17:52:33 |
| `cloud-direct` (matched control) | `20261117T175322Z-cloud-direct` | 17:53:31 | 17:54:12 |

Each visit loaded a researcher fixture that:

- drew a fixed canvas 2D scene and hashed the data URL with SHA-256;
- probed WebGL vendor/renderer, a clear-color sample pixel, and a WebGL canvas hash;
- rendered an OfflineAudioContext snippet and recorded an absolute-sum metric (8s outer timeout);
- measured ten font strings via canvas text metrics;
- recorded a navigator and screen subset, timezone options, storage availability, and media-device availability (enumerateDevices raced at 4s);
- gathered WebRTC ICE candidates with `stun:stun.cloudflare.com:3478` only (no TURN), keeping type counts and address-class labels rather than raw IPs (10s outer timeout; 8s gather timeout inside the probe);
- fetched four public document echoes with a 12s `AbortSignal` each: `api.ipify.org`, `api64.ipify.org`, `ipinfo.io/json`, and Cloudflare trace.

A composite hash covered the canonical JSON of the selected surfaces. Raw IP strings stayed in a gitignored private directory as salted comparison material only. After each successful visit we captured a full-page PNG of the fixture UI (hashes / renderer / timezone only—no raw IPs in the visible surface).

### Fixture delivery (important for interpretation)

Early proxy attempts that navigated to a loopback HTTP fixture failed: the residential proxy returned **`503 Target host denied`** for `127.0.0.1` even when Chromium bypass lists included loopback. Amendment A1 therefore delivers the fixture with Playwright **`page.setContent(fixtureHtml)`** for every condition—including the matched cloud-direct control—so fingerprint surfaces are compared under identical document origin constraints.

Consequences of `about:blank` / opaque-origin delivery (observed uniformly across conditions):

- `localStorage` / `sessionStorage` report unavailable;
- `navigator.deviceMemory` is absent;
- `mediaDevices.enumerateDevices` does not complete usefully within the timeout (`media.available: false`).

Those origin-bound fields are part of the composite. They are **not** treated as “proxy changed the fingerprint.” A historical localhost-served cloud-direct run from earlier the same day produced a different composite for exactly these origin-bound reasons and is retained as historical evidence, not pooled into A1 matched totals.

SHA-256 uses `crypto.subtle` when present and a pure-JS SHA-256 implementation otherwise (required because `setContent` is not a Secure Context). Canvas hashes matched the historical localhost run, confirming digest parity for ASCII data-URL inputs.

### Proxy assembly

- **HTTP (`proxy-us-http`):** Playwright context proxy `http://$PROXYSELLER_HOST:$PROXYSELLER_HTTP_PORT` with username/password fields from Runtime Secrets. Bypass list retained for residual local traffic.
- **SOCKS (`proxy-us-socks`):** Chromium does not accept SOCKS5 username/password. The collector uses [`proxy-chain`](https://github.com/apify/proxy-chain) **3.x** to expose a local unauthenticated HTTP forwarder that authenticates to upstream `socks5://$HOST:$SOCKS_PORT`. Design records `proxyViaLocalAnonymizer: true`. The local forwarder is a Chromium constraint workaround, not a second commercial hop.
- **CometVPN:** not used as HTTP/SOCKS; no tunnel claim.

### Main matrix (Amendment A1 matched totals)

| Condition | Run id | Visits | Composite hashes | Fixture delivery |
|---|---|---|---|---|
| `cloud-direct` (matched control) | `20261117T175322Z-cloud-direct` | 6 / 6 | 1 | `page.setContent` |
| `proxy-us-http` | `20261117T174906Z-proxy-us-http` | 6 / 6 | 1 | `page.setContent` |
| `proxy-us-socks` | `20261117T175129Z-proxy-us-socks` | 6 / 6 | 1 | `page.setContent` |

Shared composite across all eighteen matched visits:

`b661fd318fd710e98a86095230b3c5d423829410cb55806e1a38ecad1a1dd1f1`

### Protocol conditions not activated in A1

The frozen protocol lists a supplementary `proxy-dead` condition (invalid local proxy port; fixture with loopback bypass; record whether fingerprint collection still completes and how public fetches fail). **A1 did not activate `proxy-dead`.** It remains out of scope for matched totals and is not treated as a silent skip of a required main-matrix cell. Main-matrix cells `cloud-direct`, `proxy-us-http`, and `proxy-us-socks` are fully documented below.

CometVPN tunnel measurement was not activated (VPN-only product; hosted VM path not proven for this package).

### Excluded from matched totals (still in the package)

| Run | Why excluded |
|---|---|
| `20261117T150746Z-cloud-direct-pilot` | Same-day pilot before WebRTC address-class parser fix (composite matched historical localhost main run) |
| `20261117T150914Z-cloud-direct` | Historical localhost fixture delivery (secure-context storage/`deviceMemory`); composite `dd8a9bcc…` differs only on origin-bound fields |
| `20261117T173439Z-proxy-us-http` | Loopback-via-proxy timeouts (`503 Target host denied`) |
| `20261117T174350Z-proxy-us-http` | Smoke timeout before fixture hardening |
| `20261117T174720Z-proxy-us-http` | Pre SHA-256 fallback (`crypto.subtle` missing on `setContent`) |

## Finding 1: eighteen matched visits, one composite fingerprint

All eighteen A1 matched visits succeeded. They shared a single composite hash across cloud-direct, ProxySeller HTTP, and ProxySeller SOCKS.

Every individually tracked surface we scored for within-condition stability also returned one value across the six visits of each condition: canvas hash, WebGL data hash, WebGL renderer string, audio absolute sum, timezone, user agent, platform, hardware concurrency, screen JSON, font-metric JSON, and WebRTC candidate-count JSON. Cross-condition, those rendering and navigator fields matched as well; the shared composite is the formal statement of that agreement.

Closing a context and opening another did not rotate these surfaces. Routing document fetches through residential HTTP or SOCKS did not rotate them either. That matches the practical claim in our guide on [cookies versus local storage versus browser fingerprinting](/blog/cookies-vs-local-storage-vs-browser-fingerprinting/): clearing storage—or changing the network path—is not the same as changing rendering identity.

Canvas hash (all matched conditions):

`57e77ee7b491b06625fd9d58be3ddcdc3a61ac9df3fafadd4ef1ab9e3fcc59f5`

Audio absolute sum: `33.6482334`. WebGL renderer remained SwiftShader (`ANGLE (Google, Vulkan 1.3.0 (SwiftShader Device (Subzero) …))`).

### Collection screenshots (redacted fixture UI)

Public package PNGs show only the fixture’s hash / renderer / timezone summary. Raw IPs never appear on the captured surface.

![Cloud-direct matched control fixture after visit p1-v1](evidence/20261117T175322Z-cloud-direct/visit-p1-v1.png)

*Matched cloud-direct control (`20261117T175322Z-cloud-direct`), visit p1-v1. Composite hash `b661fd…`.*

![ProxySeller HTTP fixture after visit p1-v1](evidence/20261117T174906Z-proxy-us-http/visit-p1-v1.png)

*ProxySeller residential HTTP (`20261117T174906Z-proxy-us-http`), visit p1-v1. Same composite hash.*

![ProxySeller SOCKS fixture after visit p1-v1](evidence/20261117T175129Z-proxy-us-socks/visit-p1-v1.png)

*ProxySeller residential SOCKS via local anonymizer (`20261117T175129Z-proxy-us-socks`), visit p1-v1. Same composite hash.*

Each matched condition also stores the remaining five visit screenshots (`p1-v2` … `p2-v3`) in the public evidence directories.

## Finding 2: document IP labels moved with the proxy; the fingerprint did not

### Cloud-direct (matched control)

Across six visits, all four echo kinds succeeded (**24 / 24**). Endpoints returned **seven distinct salted IP labels** in total. Per-kind unique salted labels: ipify **4**, api64 **4**, ipinfo **5**, Cloudflare trace **4**. Country/loc labels stayed `US`; ipinfo org stayed `AS14618 Amazon.com, Inc.`. Within-visit agreement was incomplete: ipify and Cloudflare shared a label in **0 of 6** visits; ipify and ipinfo shared a label in **1 of 6** visits.

| Echo kind | Successes | Unique salted labels |
|---|---|---|
| ipify | 6 / 6 | 4 |
| api64 | 6 / 6 | 4 |
| ipinfo | 6 / 6 | 5 |
| cloudflare-trace | 6 / 6 | 4 |

### ProxySeller HTTP

All four echo kinds succeeded on all six visits (**24 / 24**). Each kind returned **one** salted label across the six visits (no within-condition label churn in this session; **1** unique label across all kinds). Country labels were **`BR`**; ipinfo org was **`AS28142 DIGITAL DESIGN SERVIÇOS DE TELECOMUNICAÇÕES EIRELI`**. Within-visit agreement: ipify↔Cloudflare **6 / 6**; ipify↔ipinfo **6 / 6**.

| Echo kind | Successes | Unique salted labels |
|---|---|---|
| ipify | 6 / 6 | 1 |
| api64 | 6 / 6 | 1 |
| ipinfo | 6 / 6 | 1 |
| cloudflare-trace | 6 / 6 | 1 |

### ProxySeller SOCKS

Same pattern as HTTP in this session: **24 / 24** echo successes, one salted label per kind across visits (and across kinds), country **`BR`**, org **`AS28142…`**, within-visit ipify agreement **6 / 6** with both Cloudflare and ipinfo. The composite fingerprint hash was identical to HTTP and to the matched cloud-direct control.

### What this does and does not show

The browser composite hash remained identical while egress geography and ASN labels changed from Amazon US datacenter echoes to a Brazilian residential ASN on both proxy protocols. On this VM, network echo identity and browser rendering identity were separable observations—now shown with an actual residential proxy rotation, not only with within-datacenter label churn.

We do **not** claim that ProxySeller “always exits in Brazil,” that the product is marketed as Brazil-only, or that any particular city inventory was requested. The protocol describes residential comparison legs assembled from injected secrets; public-offer geo targeting was not the measured claim. Echo country/org are endpoint observations for this package and session. We also do not treat echo-provider agreement rates as a ProxySeller reliability metric. Within-datacenter label churn on cloud-direct (seven salted labels; incomplete within-visit agreement) is itself evidence that echo endpoints are not a single ground-truth address oracle.

Related reading: [why changing your IP does not change your browser fingerprint](/blog/why-changing-your-ip-doesnt-change-your-browser-fingerprint/), [HTTP vs SOCKS proxies](/blog/http-vs-socks-proxies/), and [ProxySeller review](/blog/proxyseller-review/).

## Finding 3: automation and software-GPU signals were part of the stable identity

The stable navigator subset included `webdriver: true`, `pluginsLength: 0`, and a HeadlessChrome user agent. WebGL reported a SwiftShader software renderer. Hardware concurrency was 4. Under `setContent` delivery, `deviceMemory` was absent and storage APIs reported unavailable—identical on direct and proxied visits.

Those fields are legitimate fingerprint inputs. They also mark this run as **automation on a cloud GPU-less stack**, not a consumer laptop browser. Readers should not treat the composite hash as representative of desktop Chrome entropy. They should treat it as evidence that the measured surfaces were repeatable under our fixture—and that the proxy path did not rewrite them.

| Marker | Observed value (all matched conditions) |
|---|---|
| `webdriver` | `true` |
| `pluginsLength` | `0` |
| WebGL renderer | SwiftShader / ANGLE Subzero |
| `audio.absSum` | `33.6482334` |
| `hardwareConcurrency` | `4` |
| `deviceMemory` | absent (`null`) |
| `localStorage` / `sessionStorage` | unavailable |
| `media.available` | `false` |
| canvas hash | `57e77ee7…` (unchanged from historical localhost run) |

For hands-on canvas checks in a normal browser, use our [free canvas fingerprint tester](/tools/free-canvas-fingerprint-tester/). For broader surface types, see [types of browser fingerprinting](/blog/types-of-browser-fingerprinting/). Accuracy context: [how accurate is browser fingerprinting](/blog/how-accurate-is-browser-fingerprinting/).

## Finding 4: WebRTC candidate structure repeated across conditions

Each matched visit recorded WebRTC as available, finished on the 8-second gather timeout, and returned **two candidates**: one `host` and one `srflx` (no relay). Address-class labels were stable as one `none` (no extractable IP in the candidate line—consistent with mDNS-style host candidates) and one `public-or-other-v4` for the srflx candidate.

We do not publish raw candidate IPs. Candidate **structure** repeated with the fingerprint surfaces across cloud-direct, HTTP, and SOCKS. STUN still contacted Cloudflare’s STUN host; under a browser HTTP/SOCKS proxy, UDP STUN behavior is implementation-dependent and was not separately instrumented beyond candidate counts and classes. We therefore treat WebRTC here as a fingerprint-adjacent structural signal, not as a full leak verdict for proxy egress.

WebRTC remains a separate leak and fingerprint topic—see [what is a WebRTC leak](/blog/what-is-a-webrtc-leak/), [IP leaks, WebRTC and IPv6](/blog/ip-leaks-webrtc-and-ipv6/), and [WebRTC and browser privacy](/blog/what-is-webrtc-and-how-does-it-affect-browser-privacy/).

## Finding 5: HTTP and SOCKS agreed with each other under Chromium constraints

ProxySeller HTTP (native Playwright auth) and ProxySeller SOCKS (local `proxy-chain` anonymizer → upstream SOCKS5) produced:

- the same composite fingerprint hash;
- the same canvas / WebGL / audio / navigator / screen / font / WebRTC-count surfaces;
- the same echo country/org labels in this session (`BR` / `AS28142…`);
- the same within-visit echo agreement (6 / 6);
- six fixture screenshots each (public package).

That agreement is useful for practitioners who wonder whether SOCKS “looks different” to fingerprint scripts than HTTP when both only change the document fetch path. In this fixture, they did not. The SOCKS path’s local anonymizer is disclosed in design JSON (`proxyViaLocalAnonymizer: true`) so readers can weight Chromium’s lack of SOCKS5 username/password support appropriately.

## What ProxySeller and CometVPN mean in this package

**ProxySeller.** Amendment A1 executed residential HTTP and SOCKS Chromium proxy legs using injected Runtime Secret component names. Measurements are researcher-operated fixtures, IP echoes, and document fetches—not competitive benchmarking of ProxySeller uptime or catalog breadth. When you need an HTTP/SOCKS egress path for research or tooling, evaluate [ProxySeller](https://freepolicytools.com/proxyseller/) as a product separately from this fingerprint matrix. See also our [ProxySeller review](/blog/proxyseller-review/) and [types of proxy servers](/blog/types-of-proxy-servers/).

**CometVPN.** CometVPN is VPN-only. This cloud study does not use it as HTTP/SOCKS and does not claim tunnel measurements. A VPN can change the addresses on the wire without rewriting canvas, WebGL, or audio surfaces—the separation Findings 1–2 illustrate for document echoes. For VPN product context, see [CometVPN](https://freepolicytools.com/cometvpn/), our [CometVPN review](/blog/cometvpn-review/), and [what VPNs do not protect](/blog/vpn-limitations-what-vpns-dont-protect/).

Neither affiliate relationship changed the fixture, the hashes, the exclusion rules, or the echo redaction policy.

## A better way to check fingerprint assumptions

1. Fix the browser, viewport, locale, timezone, **and fixture delivery origin**.
2. Collect the same surface set in fresh contexts.
3. Hash surfaces you can explain, not a third-party mystery score alone.
4. Record network identity separately (echoes / route checks), with redaction.
5. Change one network condition (proxy or VPN) and compare under the same fixture delivery.
6. If loopback fixtures fail through a residential proxy, do not silently skip—change delivery (`setContent` / non-loopback host) and re-baseline.

If the composite hash stays put while egress labels move, you have measured separation—not “anonymity.” If you need intentional profile isolation, that is a different product class than a tunnel; start from [what an anti-detect browser is](/blog/what-is-an-anti-detect-browser/), [how to check your browser fingerprint](/blog/how-to-check-your-browser-fingerprint/), and [can browser fingerprinting be prevented](/blog/can-browser-fingerprinting-be-prevented/).

## Methods appendix: collector guards and failure modes

- Env discovery writes **names only** to `evidence/env-discovery.json` (presence map + matching keys). Cursor meta keys matching `CLOUD_AGENT_*` are ignored even if their names contain unrelated tokens. A1 recorded all five component names as PRESENT; `/dev/net/tun` existed as a device node and was **not** treated as a CometVPN tunnel proof.
- Credentials are never written to public visit JSON, design JSON, screenshots, or findings.
- Document echo fetches use 12s abort timeouts; media/audio/WebRTC stages have outer timeouts so a hung API cannot stall a visit past the collector wait.
- Failed guards exclude that condition until reviewed; they are not successes. The three pre-fix HTTP attempts are retained as excluded evidence of the loopback and Secure Context issues.
- Public screenshots show only the fixture’s hash summary UI. Private directories hold salts and raw echo bodies.
- Reproducibility: install Playwright Chromium from the study `package.json`, freeze `protocol.md`, run `collect.mjs` per condition with injected `PROXYSELLER_*` secrets for proxy legs, then `analyze.py`. Fixture delivery must stay `page.setContent` for matched comparisons. Do not invent credentials when secrets are absent.

## What this study cannot tell you

Concrete limits, not generic caveats:

- **One automation stack.** Headless Chromium 131 with SwiftShader on one USA Cursor VM is not desktop Chrome entropy and not a multi-browser panel.
- **One session window.** Matched A1 legs ran within about five UTC minutes on 17 November 2026. We did not measure day-to-day drift, Chromium upgrades, or ProxySeller session rotation policies over time.
- **`proxy-dead` not run.** Failure-mode proxy behavior remains unmeasured in this package.
- **No CometVPN tunnel.** Affiliate/product context only; no WireGuard/TUN claim.
- **No population uniqueness.** A shared composite hash is repeatability, not a “how unique am I?” percentile.
- **No ProxySeller ranking.** Echo geo/org labels and agreement rates are session observations, not uptime, city-inventory, or competitive benchmarks.
- **WebRTC is structural only.** Candidate counts/classes repeated; UDP STUN through browser proxies was not separately instrumented into a leak verdict.
- **Opaque-origin fixture.** `setContent` disables some storage/`deviceMemory`/media fields by design for matched delivery; those absences are not proxy effects.
- **No provider logging claim.** We did not inspect provider logs, retention, or law-enforcement processes.
- **No headed consumer browser, no anti-detect profile, no commercial uniqueness score page** as the primary sample.

## Frequently asked questions

### Does this prove fingerprinting uniquely identifies every person?

No. We measured stability of researcher-selected surfaces across eighteen matched Playwright Chromium visits on one cloud VM under three egress conditions. A single shared composite hash shows repeatability here. It is not a population uniqueness estimate.

### Did you test ProxySeller residential proxies?

Yes, in Amendment A1. After Runtime Secrets injected the five component names, we ran six visits each on residential HTTP and SOCKS. Both shared the matched cloud-direct composite hash while document echoes moved to `BR` / `AS28142` labels in this session. That is not a geo-inventory or uptime claim.

### Did you measure CometVPN?

No tunnel measurement. CometVPN is VPN-only product context and affiliate CTA material. Hosted VMs are not assumed to provide a proven CometVPN path. A VPN can change wire addresses without rewriting canvas or WebGL surfaces.

### Why did the composite hash differ from the earlier localhost cloud-direct run?

Fixture delivery. Residential proxies denied loopback HTTP (`503 Target host denied`), so Amendment A1 delivers the fixture with `page.setContent` for every condition. Opaque-origin delivery disables storage and `deviceMemory` and times out media enumeration. Those origin-bound fields change the composite. Under matched `setContent` delivery, direct, HTTP, and SOCKS share one hash. The historical localhost run remains excluded from matched totals.

### Why did document IP labels disagree on cloud-direct but agree on proxy legs?

On matched cloud-direct, public echo endpoints returned seven salted labels total and incomplete within-visit agreement while country stayed `US` and org stayed `AS14618`. On both ProxySeller legs in this session, each echo kind returned one salted label and ipify agreed with Cloudflare and ipinfo on all six visits, with country `BR` and org `AS28142`. Network echo identity and browser rendering identity remained separate observations.

### Is headless Chromium a realistic consumer fingerprint?

No. The stable identity included `webdriver: true`, zero plugins, a HeadlessChrome user agent, and a SwiftShader WebGL renderer. Treat the result as automation-stack repeatability, not desktop Chrome entropy.

### How did SOCKS authentication work if Chromium rejects SOCKS5 passwords?

The collector uses `proxy-chain` to expose a local unauthenticated HTTP forwarder that authenticates to upstream ProxySeller SOCKS5. Design JSON records `proxyViaLocalAnonymizer: true`. The local forwarder is a Chromium constraint workaround, not a second commercial hop.

### Can I reproduce the study?

Yes. The downloadable ZIP includes the amended protocol, Playwright collector, fixture, analyzer, redacted visits for all three matched conditions, design files, source snapshots, fixture screenshots, and SHA-256 manifests. Raw IP bodies remain private. Proxy legs require injected `PROXYSELLER_*` secrets.

## Evidence, exclusions, and disclosure

The public ZIP includes the amended protocol, collector, fixture, analyzer, env-discovery record (names only), matched-control and proxy-leg visit JSON, design files, source snapshots, SHA-256 manifests, analysis, findings, fixture screenshots for matched runs, and this article. Raw IP bodies and salts remain private.

**Excluded from matched totals:** historical localhost cloud-direct; same-day pilot; three pre-fix HTTP attempts (loopback denial / smoke / missing SHA-256 fallback).

**Excluded from claims:** CometVPN tunnel measurements; commercial uniqueness scores; headed consumer browsers; ProxySeller uptime or geo-inventory rankings; any assertion about provider logging; any claim that SOCKS and HTTP are interchangeable for every application (this study measures Chromium document-fetch fingerprint surfaces only); `proxy-dead` failure-mode behavior (not activated).

No café customer traffic was captured. No competitive ProxySeller uptime benchmark was run.

**Affiliate disclosure:** This article contains ProxySeller and CometVPN affiliate links (`/proxyseller/`, `/cometvpn/`). FreePolicyTools may earn a commission if you purchase through them. These offers are separate from the research findings and do not constitute a comparative product test or change the study’s methods, results, or limitations. CometVPN is VPN-only and was not used as HTTP/SOCKS credentials.

### External references

- [W3C / MDN: Navigator](https://developer.mozilla.org/en-US/docs/Web/API/Navigator)
- [MDN: CanvasRenderingContext2D](https://developer.mozilla.org/en-US/docs/Web/API/CanvasRenderingContext2D)
- [MDN: WebGLRenderingContext](https://developer.mozilla.org/en-US/docs/Web/API/WebGLRenderingContext)
- [MDN: RTCPeerConnection](https://developer.mozilla.org/en-US/docs/Web/API/RTCPeerConnection)
- [MDN: Secure contexts](https://developer.mozilla.org/en-US/docs/Web/Security/Secure_Contexts)
- [Playwright: BrowserContext isolation](https://playwright.dev/docs/browser-contexts)
- [Playwright: network proxy](https://playwright.dev/docs/network#http-proxy)
- [proxy-chain (Apify)](https://github.com/apify/proxy-chain)
- Related FPT research: [proxy routing](/blog/proxy-routing-research/), [proxy encryption](/blog/proxy-encryption-research/), [VPN website protection warnings](/blog/vpn-website-protection-warnings/).

Cite as: FreePolicyTools, “Same Browser Fingerprint Across Direct, HTTP, and SOCKS. Only the Document IPs Moved,” 17 November 2026 (Amendment A1). Report the matched setContent matrix on Playwright Chromium 131 with ProxySeller residential HTTP/SOCKS comparison legs (`20261117T175322Z-cloud-direct`, `20261117T174906Z-proxy-us-http`, `20261117T175129Z-proxy-us-socks`). There is no representative user sample, provider ranking, or discovered vulnerability claim.
