DEV Community

OnePassApps
OnePassApps

Posted on

What the Accept Header Decides on Our Own Site: Cloudflare's Beacon Tag Rides In or Out, and the API Host That Ignores It

I work on OnePassApps - this is our own site, and every number below was measured today, 11 November 2026, from outside the browser, signed out, over plain HTTPS, with no cookies and no JavaScript.

An earlier post of ours measured that the byte length of one of our pages moves with the Accept header the client sends, and left the mechanism open. This post opens it: which part of the request decides it, what is added and removed, and what the API host does with the same header.

What I measured

  • 13 Accept header shapes on the app host home page, 3 fetches each: 39 requests
  • 2 more app-host routes, 2 shapes each: 4 requests
  • 7 API routes x 6 Accept values: 42 requests

85 requests carry the numbers in this post, every one of them to our two hosts, from one network.

On the app host, one header decides whether the beacon ships

Across all 13 shapes the home page came back in exactly two lengths, and every shape returned the same length on all three fetches: 25,011 bytes or 24,644 bytes.

Accept header sent Bytes Beacon tag
(no header sent) 25,011 present
text/html 25,011 present
text/html;q=0.9 25,011 present
text/html,*/* 25,011 present
text/html,application/xhtml+xml 25,011 present
application/xhtml+xml,text/html 25,011 present
text/plain;q=1.0,text/html;q=0.1 25,011 present
application/xhtml+xml 24,644 absent
image/avif,image/webp,*/*;q=0.8 24,644 absent
*/* 24,644 absent
application/json 24,644 absent
text/plain 24,644 absent
image/png 24,644 absent

In all 13 shapes the beacon tag is in the document exactly when the header names text/html - at q=0.9 and at q=0.1, listed behind text/plain, listed after application/xhtml+xml, and also when no header is sent at all. It is gone for a wildcard, for application/json, for text/plain, for image/png, and for application/xhtml+xml on its own.

The 367-byte difference is one element: Cloudflare's own analytics beacon tag, 366 bytes of markup pointing at static.cloudflareinsights.com, which is a host that is not ours. It is in the 25,011-byte document and not in the 24,644-byte one. An earlier post of ours named that tag and hashed the 31,853 bytes Cloudflare serves for it; what that post did not do was show what decides whether it is in the page at all.

Two more routes behave the same way, with the same 367-byte gap: /guides came back 61,760 bytes for text/html and 61,393 for application/json, and /apps/paperkit 19,054 against 18,687.

One more measured fact worth having next to this: our pages carry no Vary header - not on the three routes above, and an earlier post of ours found none on any of the 56 pages. Nothing on the wire tells a cache that one URL can answer with two different documents.

On the API host, the header changes nothing

7 routes x 6 Accept values - application/json, text/html, application/xml, text/plain, image/png and a wildcard - is 42 requests. Every one of them answered HTTP 200 with Content-Type: application/json, and on each route the body was identical across all six header values once the per-request id and the timestamp are masked. One digest per route, seven routes, no exceptions.

The route lengths: /api/products 15,650 bytes, /api/voting 4,146, /api/plans 4,020, /api/config/feature-flags 1,771, /api/health 410, /api/runtime 471, /api/auth/config 371.

Not one 406. Not one HTML body. Asking for XML, for plain text or for a PNG gets JSON, and gets the same JSON.

The API host does send a Vary header, and it names the dimensions that actually change the answer: Accept-Encoding, Origin on /api/products, /api/config/feature-flags and /api/plans, Origin on /api/runtime, /api/auth/config and /api/health, and Authorization, Cookie, Accept-Encoding, Origin on /api/voting. Accept is not in any of them, which is consistent with the answer never depending on it.

What I did not measure

  • I did not open a browser, so I make no claim about what a browser does with the beacon tag.
  • I sent no cookies and no Authorization header, so the signed-in and cross-origin paths were not exercised.
  • Two fetches of the same shape are not byte-identical to each other - a per-request key moves inside the document - so every comparison here is by length, or by digest with that region masked.
  • These are 13 header shapes, not every header a client can send. The rule I can support is the one above: the beacon rides along when the header names text/html.

Earlier measurements of the same site, for the record: the language layer post that first measured the byte-length effect, the two Cloudflare scripts post that named the beacon tag itself, and the post on why one URL never returns the same bytes.

The apps are at app.onepassapps.com, and the guide for each one is at app.onepassapps.com/guides.

I work on OnePassApps, so this is our own site and our own API host measured today.

Top comments (0)