流量伪装

代理供应商

首先需要了解代理供应商提供的网络流量特征。

https://datadome.co/bot-management-protection/how-proxy-providers-get-residential-proxies/

https://www.novada.com/blog-ordinary/the-core-difference-and-application-scenarios-between-static-isp-and-rotating-residential-proxies/

静态代理:ip 来源

Most Static Residential IPs are created through a business deal between a Proxy Provider and a Consumer ISP.

How it looks in databases:

静态代理:数据移动链路

Your machine -> proxy server -> target website

When they give you 1.2.3.4:8000:user:pass, here is what is happening on their end:

动态代理:数据移动链路

Residential proxy providers do not give you a direct list of 10 million IP addresses. Instead, they give you a single Gateway Address (e.g., us-geo.provider.com:8000).

Your machine -> gateway / entry node -> exit node -> target website

Here is the chain of events when you make a request:

动态代理:Exit Nodes 来源

设备伪装

Windows VS Mac

Mac 不同产品线都有一个特点,配置确定的情况下,所有机器的硬件都是一样的。比如最低配的 Mac mini m4,世界上所有的机器,硬件都是一样的。

检测移动设备伪装

If you can pass CreepJS with a high trust score, you are beating 99% of detections.

There are absolutely tools and specific techniques to detect this. In fact, distinguishing a "Fake Mobile" (desktop spoofing Android) from a "Real Mobile" is one of the easiest checks for modern anti-fraud systems.

Because your underlying hardware (Intel CPU, NVIDIA GPU, wired power source) is physically different from a phone's hardware (ARM CPU, Adreno/Mali GPU, battery), your browser leaks these differences in dozens of tiny ways.

1 The "Silver Bullet" Detection Tools

You can check your own browser against these tools to see if you are leaking:

2 How They Detect You (The Technical Vectors)

Here are the specific signals security systems use, ranked from easiest to hardest to fix.

各类检测手段对代理流量的检测

IP

静态代理的 IP 一般就是 ISP IP,不用过多担心。

动态代理的 IP 可能各种类型都有,要看供应商的好坏,比如 business isp mobile。同时不同 ip 检测商给出的检测结果也不一样。

移动代理是一种特殊的动态代理,ip 是手机运营商分配的,只可能是 mobile 类型。

business 类型的 ip 最好不要用。

TCP/IP fingerprint

这个问题实际上没那么重要。

动态代理的真实情况

首先说一下真实家庭中路由器的作用。路由器起 NAT 作用,只改 ip,对 tcp 特征基本不改变。

动态住宅代理网络的本质,就是由全球成千上万个真实用户的设备组成的 P2P 网络。这些 IP 的来源主要有三种:

我们进行的模拟

TLS fingerprint

target website 得到的 TLS fingerprint 是链路开头的 fingerprint。链路开头就是指纹浏览器或者 BAS 用的 CEF。

另外安全公司一般只能从 TLS fingerprint 中获取浏览器的种类和版本(通过 SSL/TLS library 判断),不能获取操作系统。

使用指纹浏览器时,user agent 里的 chrome 版本要和实际浏览器用的 chrome 引擎版本对上。不然很容易就露馅了。

Latency detection

这个问题同样棘手。TCP timing 是链路中的最后一环到网站的时间。Websocket timing 是整条链路的时间。

静态代理的链路一般是:my machine -> proxy server -> target website。我们可以缩短 machine 到 proxy server 的物理距离,或者直接在 proxy server 上进行软件操作。

但是静态代理有一个非常大的问题:因为代理服务器在数据中心中,网络设施比较好,TCP Timing 非常小而且基本不变。这和普通用户不符。普通用户因为各种原因,TCP timing 会来回波动。

动态代理的链路一般是:my machine -> gateway -> exit node -> target website。我们可以缩短 my machine 到 exit node 的物理距离,让 my machine gateway exit node 都在一个区域(尽量不要距离太远,小国家跨国可以,但是比如美国,从美西到美东,距离太远了,会有明显 latency)。

移动代理的链路一般是:TODO。

如果 my machine 在 CN,还要在链路中加一个代理,那情况更是雪上加霜。

Javascript fingerprint

通过 bablosoft 动态指纹或者指纹浏览器,一般很好解决。

静态 ip 解决方案

买来的静态 ip 基本 tcp/ip 指纹全是 linux,加上 tcp timing 没有波动,模仿不了真实用户。

动态 ip 解决方案

浏览器热身

"Building a small history of browsing" (often called "Cookie Profile Building" or "Account Warm-up") is the process of making your browser look "lived-in" before you visit your target link.

When you open a fresh Chrome instance in your automation software, it is "sterile." It has no cookies, no cache, no history, and no "Ad Interest" profile.

Ad networks (like Google Ads, Taboola, Outbrain) hate sterile browsers. A user with zero history who appears out of nowhere, clicks an ad, and disappears is the #1 signature of a bot.

Here is the technical breakdown of what "building history" actually does and why your 1-minute plan is risky.

  1. The "Cookie Sync" (The Most Important Part)

Ad networks track users across the internet.

  1. Chrome's Local "Topics API"

Modern Chrome doesn't just rely on cookies; it builds an internal profile of you called the Privacy Sandbox (Topics API).

  1. The "Referrer" Chain

"Building history" also means fixing how you arrive at the page.