TLS and HTTP/3 Diagnostic Guide

doc

Learn how to read TLS handshake fields and HTTP/3 negotiation results to detect downgrade, proxy policy, and path issues.

TLS was designed to protect the content of a connection, not to hide the shape of it. Before any certificate is validated, the client announces what it supports: protocol versions, cipher suites, extensions, and the order they appear in. That announcement is unique enough to fingerprint, and it is visible to every middlebox on the path.

What the handshake exposes

ElementWhat it revealsStability
Protocol versionHow old the client stack is. TLS 1.0 or 1.1 signals an unmaintained system.Changes only on upgrade
Cipher suite listLibrary and version. The set and its order differ between OpenSSL, BoringSSL, Schannel and NSS builds.Stable per build
Extension orderFiner than the suite list. Two browsers on the same library still differ here.Stable per build
ALPN offerWhich application protocols the client will accept: h2, http/1.1, h3.Stable per build
Supported groupsElliptic curves offered for key exchange. Narrows the library generation.Stable

Hashed together, the client hello becomes a compact identifier — the well-known JA3 form, and the newer JA4 which separates the pieces so that one changed field does not invalidate the whole. The identifier is not a person. It is a build of a TLS stack, which is why thousands of users share one.

Why middleboxes care

Filtering by content requires decrypting, which is expensive and legally constrained in many places. Filtering by handshake shape is cheap: it needs only the first packet, and it can discriminate between a browser and a proxy client without touching the payload. Two practical consequences:

  • Clients that do not look like browsers get challenged. A custom TLS client from a proxy tool has an unusual suitelist. Blocking or throttling that fingerprint affects every user of that tool, regardless of destination.
  • Fingerprint-based classification is coarse and unstable. It produces false positives on new browser releases and on clients that mimic browser fingerprints deliberately. Treating a match as proof of intent is a category error.

HTTP/3 changes the transport, not the privacy

HTTP/3 runs over QUIC, which is UDP rather than TCP. Three differences matter for diagnosis:

  • UDP 443 must be reachable. Networks that block all UDP — common on corporate Wi-Fi and some hotel networks — force a fallback to HTTP/2 over TCP. A page that loads slowly but loads is often this, not a broken server.
  • QUIC hides the handshake inside encrypted packets. Unlike the TLS client hello, the QUIC initial packet is encrypted with a key derived from the connection ID. Passive fingerprinting of the TLS layer does not apply, which removes one classification channel — and introduces others based on packet timing and size.
  • Connection migration. QUIC can survive an IP change mid-session. This is useful on mobile and confusing when you are diagnosing why an address changed without a reconnection.

HTTP/3 being available does not mean it is being used. The check has to observe an actual negotiation — an ALPN h3 response — rather than infer from browser version.

What a TLS and HTTP/3 check measures

  1. Negotiated version. What the server and client actually agreed on, which may be lower than what both support.
  2. Negotiated cipher. The suite in use, distinct from the suite list offered.
  3. ALPN result. Which protocol was selected: h2, http/1.1 or h3.
  4. Certificate chain basics. Issuer, validity window, and whether the chain validates. Not the same as a trust judgement.
  5. Handshake fingerprint. The client's announcement hash, comparable across sessions from the same machine.

What you can and cannot change

You can change the version floor, disable legacy suites, and prefer encrypted transports. You cannot, from a normal browser, change the extension order or the ALPN set — those are compiled into the library. This is the honest limit: a browser fingerprint at the TLS layer is largely fixed by the browser build you run, and the meaningful decisions are about which build you run and whether the network can see it at all.

Encrypted Client Hello, where deployed, hides the destination from the middlebox by encrypting the SNI. It does not change the fingerprint of the client hello fields, and it is supported by a minority of endpoints, so its practical effect depends on the specific path.

Common misreads

  • "The fingerprint identifies me." It identifies the TLS stack build. Many users share it; it is a class, not an individual.
  • "HTTP/3 means I am private." It changes the transport. It does not hide destination or content from the endpoint you connect to.
  • "TLS 1.2 means the connection is weak." With modern AEAD suites, TLS 1.2 is still sound. The version number alone is not the security measure.
  • "A fingerprint mismatch means interception." A new browser release changes the fingerprint legitimately. Interception is suggested by altered certificates, not by a new hash.

FAQ

Is a TLS fingerprint the same as a browser fingerprint?

No. Browser fingerprinting uses JavaScript-observable properties. TLS fingerprinting uses the handshake, which is visible without running any code on the client.

Does the handshake reveal the destination?

The SNI field usually does, in plaintext, unless Encrypted Client Hello is in use. That is a separate problem from the fingerprint.

Should I disable HTTP/3?

Only for diagnosis. If HTTP/3 works, it is generally faster. Disabling it to "improve privacy" changes little, because the endpoint still sees the same session.