Faymas · Practical guide

Why Do the Faymas Domains Give Two Different Answers About HTTP/2?

Because the protocol is picked from the client's offer list, not dictated by the server. On 16 November 2026 all three Faymas apex domains served HTTP/2: offer h2 in the TLS handshake and the edge selects h2, and nine real HTTP/2 requests returned 200 on faymas.in, 301 on faymas.net and 200 on faymas-in.com. The curl build on this Windows host offers only http/1.1 in ALPN, so it reports HTTP/1.1. The site never refused h2. It was never asked. HTTP/3 is advertised in a response header and was not measured here.

Updated November 16, 2026 · Prompt-text analysis

This guide analyses published prompt text. No generation tests were run.

Why one site reports two protocols

ALPN is decided by the client first. A client sends the list of application protocols it can speak. The server picks one from that list, or picks none. It does not get to volunteer a protocol the client never offered. That is the whole mechanism, and it explains most of the confusion around this question.

Here is curl 8.21.0 on this machine, verbatim from curl -vI against faymas.in: ALPN: curl offers http/1.1, then ALPN: server accepted http/1.1, then HEAD / HTTP/1.1. One offer. One acceptance. Nine repeat requests with curl -w against the three apexes all reported 1.1. Nothing about the edge refused anything. It was never asked.

Change the offer and the answer changes. With openssl s_client -alpn h2 the same host reports ALPN protocol: h2. With no ALPN extension at all it reports No ALPN negotiated. Same host, same port, same minute. Three different outcomes, determined entirely by what the client put in its list.

An earlier audit of these same domains, published on 20 October 2026, recorded 106 responses over HTTP/1.1 and explicitly left HTTP/2 untested, because the curl build on this machine has no --http2 option. That was a limit of the tool, not a finding about the edge. This page closes that one gap with two different tools: openssl s_client negotiates ALPN, and Node's built-in http2 client actually speaks it.

The offer matrix: 90 handshakes, one rule

We ran six different ALPN offer lists against five hostnames, three times per cell. Ninety handshakes. Every cell was identical on all three attempts. The five hostnames were the three apex domains plus www.faymas.net and cdn.faymas.in as controls; the controls behaved exactly like the apexes.

Client ALPN offerServer selection (all five hostnames)
no ALPN extensionNo ALPN negotiated
h2h2
http/1.1http/1.1
h2,http/1.1h2
fooNo ALPN negotiated
h3No ALPN negotiated

Three things fall out of that table. The edge prefers h2 when it is available. It falls back to http/1.1 when that is the only real offer. And it declines to invent a protocol: an unknown identifier like foo produces no negotiation at all, not a downgrade and not an error. h3 behaves the same way, which is correct. HTTP/3 runs over QUIC on UDP, so it is not negotiated inside a TCP TLS handshake.

We also forced TLS 1.2 with -tls1_2 -alpn h2 on the three apexes, three times each. All nine reported Protocol: TLSv1.2 and ALPN protocol: h2. The protocol selection is independent of the TLS version in use.

Negotiation is not delivery

ALPN reports what the handshake agreed to. It does not prove a page was actually transferred over that protocol. To close that gap we opened a real HTTP/2 session with Node's built-in http2 client, which always offers h2, and requested the root path on each apex three times.

HostnameAttemptsNegotiated ALPNHTTP statusServer header
faymas.in3h2 (3/3)200 (3/3)cloudflare
faymas.net3h2 (3/3)301 (3/3)cloudflare
faymas-in.com3h2 (3/3)200 (3/3)cloudflare

The 301 on faymas.net is not a failure. That apex redirects to its www host, a behaviour an earlier audit of this site already recorded. An HTTP/2 response can carry any status. Every one of the nine responses came back through an h2 session, and every CF-RAY in the run ended in -NRT, which matches the edge location recorded on 14 November 2026.

HTTP/3 is advertised, not measured

Response headers are a different kind of evidence again. We fetched the root path with curl -sI on five hostnames, three times each, and read the alt-svc header.

HostnameStatus (3/3)alt-svc header (3/3)
faymas.in200h3=":443"; ma=86400
faymas-in.com200h3=":443"; ma=86400
www.faymas.net200h3=":443"; ma=86400
faymas.net301not present
cdn.faymas.in404not present

Three hosts carrying a 200 response advertise h3 on port 443 with a one-day lifetime. The two non-200 responses carry no such header. An earlier audit counted the same header on 106 of 106 successful responses across four domains, so this is a narrower hostname-level count that also includes two non-200 replies, not a contradiction. We did not open a QUIC connection, we did not test UDP port 443, and we did not verify that an h3 upgrade succeeds. An alt-svc header is an invitation, not a measurement. HTTP/3 support here is therefore advertised and untested, and we record it as not checked rather than as working or broken.

Four levels of evidence

These results are easy to misread if they are all called "support". They are four different claims with four different strengths, and the strongest honest statement uses the strongest level only.

Advertised. The alt-svc header says h3 exists on port 443. Nothing was sent over QUIC. Negotiated. Ninety ALPN handshakes show the edge will select h2 when offered. Delivered. Nine real HTTP/2 sessions returned responses. Client-observed. curl reports 1.1 because of its own offer list. A report that says this site is HTTP/1.1 only is describing the last level and silently discarding the other three.

This also separates the layers cleanly from earlier work on this site. The certificates differ, the mail records differ, and the DNS zones differ. The protocol profile does not. All five hostnames returned the same handshake parameters on all three repeats: TLS 1.3, TLS_AES_256_GCM_SHA384, the hybrid post-quantum key-exchange group X25519MLKEM768, and ecdsa_secp256r1_sha256 signatures. That is one edge configuration, whatever the zones behind it do.

Where the evidence stops

Fifteen -status handshakes returned OCSP response: no response sent. That describes what the edge stapled during these handshakes; it says nothing about whether the certificates are valid or revoked, and we make no claim either way. A separate -reconnect attempt produced fresh handshakes each time, but TLS 1.3 session tickets arrive after the handshake and this client does not carry them across reconnects, so session resumption is recorded as not determined rather than as unsupported.

We requested only the root path. We did not test TLS 1.0 or 1.1, because OpenSSL 3.5.7 on this host refuses those versions locally and the failure cannot be attributed to the server. We did not test 0-RTT, QUIC, or any other vantage point. We did not consult registrant records, certificate transparency logs, or an origin server. No traffic figures, ranking data, or revenue estimates appear anywhere on this page.

Protocol support is not a property of a domain by itself. It is a property of a domain plus the client that knocks. On 16 November 2026 these three domains answered h2 when asked for h2, and http/1.1 when asked for http/1.1. Both answers are true. Only one of them tends to get quoted.

FAQ

Why did an earlier Faymas audit leave HTTP/2 untested?

Because that run used a curl build without --http2. This audit used openssl s_client and Node's http2 client instead.

What does the edge do when a client offers only h2?

It selects h2. All five hostnames did so on three repeats each.

What does the edge do when a client offers an unknown ALPN identifier?

It negotiates nothing. All five hostnames reported No ALPN negotiated for an offer of foo.

Is HTTP/3 negotiated inside the TLS handshake?

No. An offer of h3 produced No ALPN negotiated; the h3 advertisement lives in the alt-svc response header.

Did the earlier HTTP/1.1 result mean the edge refused h2?

No. curl offered only http/1.1, so the edge accepted the single protocol on the list.