Show us where this rubric is wrong
A published rubric that nobody can attack at a specific line is not falsifiable — it is just visible. This is the mechanism that makes the difference, and it is deliberately the plainest thing on the site.
What filing does, exactly. It stores your challenge against one of the 13 priors and gives you a permanent address for it. It does not edit the table. When a prior actually moves, the rubric version increments, the printed form changes shape, and the changelog records what moved. Anything else would be an editorial process this site does not have and should not pretend to.
What makes a good challenge
- It names one prior. “The whole table is guesswork” is true and useless; “the fractional device pixel ratio bucket prices 1.25 the same as 1.7639” is actionable.
- It carries a source. A published dataset, a specification section, a vendor's own documentation, or a measurement you took and are willing to describe.
- It says what should replace the assumption. A prior can only be moved to another number, and that number needs an argument behind it.
Weaknesses already admitted
Each prior's own page carries a “where this prior is weakest” section written by us, first, before anyone attacked it. Those are the known problems; a challenge that finds a new one is worth considerably more.
Why counting alone does not settle a prior
The straightforward way to justify a prior is to measure it. This site does count: every column now shows how many of the devices these sites have read share your value. But those devices are visitors of fingerprinting tools, not the web, and the form says so beside every count. A count can embarrass a prior; it cannot replace one.
Falsification is the honest substitute. It cannot produce a frequency, but it can remove a wrong assumption, and a table that has survived being attacked in public is worth more than a table nobody could reach.
The priors are also available as machine-readable JSON, so an argument can be made against the data rather than against a web page.
The three priors most likely to be wrong
Nobody has to find these; they are printed on the columns they belong to, and a challenge against one of them starts from an admission rather than from an argument.
Every fractional pixel ratio in one bucket. The pixel ratio rule prices 1.25 — the value Windows display scaling at 125% produces — identically to a value like 1.7639, at 0.02 and 5.64 bits. Any source that separates ordinary scaling factors from unusual ones would move this immediately.
Twelve time zones sharing one number. The time zone rule assigns 0.05 to twelve named zones and 0.006 to every other zone on earth. Asia/Kolkata and Europe/Amsterdam do not carry the same traffic, and there is no tier between the two buckets.
One prior for every canvas hash. The canvas rule gives every value 0.002 and therefore 8.97 bits, which is the largest single reading in the table handed out identically to everyone. It cannot distinguish a stock configuration from a unique one, and it never claims to.
What a version bump actually changes
A moved prior is not an edit to a number in a database. It reprints the chart form: the envelope for that column changes shape, its modal line moves, the reference reading for a browser at every rule's modal outcome shifts, and the changelog carries the two forms overlaid so the difference is a shape rather than a claim.
It also changes what an old derivation means, which is why every published stub records the version it was scored under and carries the assumed probability for each of its own rows. A stub torn off under rubric 1.0.0 stays checkable after the table moves, because it does not depend on the table any more — the numbers it needs are printed on it.
Where to look before filing
The prior table in full lists every rule with its rationale, and each column has its own page with a section on where that prior is weakest. The page on priors and counts explains what kind of evidence would actually change the basis field. Both are worth a pass before you write.
Questions about challenging a prior
Filing records a challenge against one prior in rubric 1.0.0. It does not change the table, and nothing on this page pretends it does.
- What happens to a challenge after I file it?
- It is stored against one named prior and given a permanent address. It does not edit the table. When a prior actually moves, the rubric version increments, the printed form changes shape, and the changelog records what moved and why.
- Do I need a published source to file one?
- You need something better than an opinion. A specification section, a vendor document, a public dataset, or a measurement you took and are willing to describe in enough detail that somebody else could take it too. A challenge with nothing behind it is a note, and it is treated as one.
- Can I challenge the correlation discount rather than a prior?
- Yes. The discount of 0.35, the clamp at one in 4096 and the four band edges are all published constants of the same rubric version, and all of them are open to the same argument. The discount is the constant we consider weakest, because a single flat factor cannot express two different strengths of dependence.
- Why not just fix the table when someone is right?
- Because a prior that changes quietly makes every stub torn off before the change unreadable. A moved prior is a version bump, and a version bump reprints the form. The delay is the cost of keeping older derivations checkable, and it is a cost worth paying.
- What kind of challenge is most useful?
- One that names a bucket the table treats as uniform and shows it is not. Twelve time zones sharing one prior, every fractional pixel ratio sharing another, and every canvas hash sharing a third are the three places where a real source would change the most numbers at once.
Where the pens move
A prior can be argued with. A signal produced by a real driver has to be changed where it is produced.
Engine level, not JS level. Donut Browser spoofs through CDP injection in its Wayfern Chromium fork, so the value a page reads is the value the engine reports — there is no patched accessor to catch.
It refuses to contradict itself. A profile whose proxy exit geography, time zone or language disagrees with its fingerprint does not launch. That is a pre-launch gate in the app, not advice in a manual.
Open, and quiet. AGPL-3.0, desktop app and sync server both in the repository, and no analytics SDK anywhere in the source. macOS, Windows, Linux.