Fingerprintscore Chart form FS-1 Rubric 1.0.0 Priors: assumed · Counts: real Donut Browser ↗

The version is a property of the paper

One version exists so far: rubric 1.0.0, the table every score on this site is computed from. There is no small tracked version string parked in a footer here — a different rubric prints a different form, with different envelopes, and you can see which ones moved by putting the two stocks on top of each other.

That matters because a uniqueness score is only comparable to another score computed the same way. A number quoted without the version of the table behind it is not checkable a year later, which is why the version is carried by the drawing, printed on every stub, and served in the JSON rather than mentioned in a footnote.


Released versions

Rubric 1.0.0

First public rubric. All priors are reasoned assumptions; none are measured. Groups and discount published alongside.

Basis
prior-assumption
Measured
false
Columns
13
Band edges
12 · 20 · 28 · 36 bits
Modal total
23.61 bits, band “distinctive”

One version exists. That is not a gap in this page — it is the state of the table, and inventing a 0.9 to make a changelog look busier would be the first lie on a site built against exactly that.

What a diff will look like

When a second version ships, this page overlays its stock on the current one. An envelope that grew means the rule gained outcomes; an envelope that shifted up means the rule now treats a value as less likely than it did; a modal line that moved means the assumption about the commonest case changed.

That is the whole reason the version is carried by the drawing rather than by a number. A reader can see what moved without trusting a release note about it.

What happens to old stubs

Every published stub records the rubric version it was scored under, and every row on it carries its own assumed probability. It therefore stays checkable after the table changes: you can re-derive the same total from the numbers printed on the stub itself, without this site's cooperation.

What would force a version bump

  • Any change to a prior, however small.
  • Adding or removing a rule, or moving one between correlation groups.
  • Changing the 0.35 discount, or the clamp, or a band edge.
  • Replacing the assumptions with measured frequencies — which would also change the basis field from prior-assumption, and would add a sample count to every prior.

Chart form FS-1 — rubric 1.0.0

024681012.0012 COMMON20 DISTINCTIVE28 RARE36 SINGULAROSSCREENDPRDEPTHCORESMEMORYZONELANGGPUCANVASAUDIOFONTSTOUCHONE OUTCOMEBITS / SIGNALTOTAL 0–40FORM FS-1 · PRIOR TABLE, RUBRIC 1.0.0 · PRINTED BEFORE THE RUN · PRIORS ASSUMED 024681012.0012 COMMON20 DISTINCTIVE28 RARE36 SINGULAROSSCREENDPRDEPTHCORESMEMORYZONELANGGPUCANVASAUDIOFONTSTOUCHONE OUTCOMEBITS / SIGNALTOTAL 0–40FORM FS-1 · PRIOR TABLE, RUBRIC 1.0.0 · PRINTED BEFORE THE RUN · PRIORS ASSUMED

This is the baseline every future diff is drawn against: 13 columns, their envelopes, and the reference reading of 23.61 bits printed with the form. When a second version ships, its stock is drawn over this one.


Which parts of a version are load-bearing

Four things travel with a version number, and all four have to move together. The priors decide what each outcome costs. The correlation groups decide which readings are treated as restatements of one fact. The discount of 0.35 decides how much a non-dominant member still contributes. And the four band edges decide what the resulting total is called.

Change any one of them and every previously published total means something slightly different. That is why they are versioned as a set rather than tuned individually, and why a stub records the version instead of only the number.

The clamp is versioned too

Every prior is held between one in 4096 and 0.999 before the logarithm is taken, which caps any single column at 12.00 bits. That ceiling is why the left-hand scale on the form stops where it does. Moving the clamp would change the shape of the paper more visibly than moving any single prior.

What measured priors would and would not change

The optimistic version of this page has a rubric 2.0.0 in which the priors are replaced by frequencies with sample counts beside them. Even then, three things on this site would not change: the derivation would still be printed row by row, the correlation discount would still have to be disclosed, and the question of whether the sample represents anything would become more important rather than less.

A measured table taken from visitors to a page about browser uniqueness would be a measurement of people interested in browser uniqueness. Publishing it without saying so would be a worse failure than the assumptions it replaced, because it would borrow the authority of counting.

The current table, rule by rule · file a challenge that could force a version bump · fetch the versioned table as JSON


Questions about rubric versions

Basis prior-assumption, measured false, correlation model: grouped: strongest member at full weight, others at 35%.

How many rubric versions exist?
One. Rubric 1.0.0 is the first public table and nothing has been superseded yet. A changelog with a single entry is the honest state of a table published once, and inventing an earlier version to make this page look busier would be the first lie on a site built against exactly that.
What forces a version bump?
Any change to a prior, however small. Adding or removing a rule. Moving a rule between correlation groups. Changing the 0.35 discount, the clamp at one in 4096, or any of the four band edges. All of those change what a score means, so all of them reprint the form.
Will an old derivation still make sense after a change?
Yes, because a published stub carries the version it was scored under and the assumed probability for every one of its own rows. You can re-derive its total from the numbers printed on it without this site agreeing with them any more.
Why show version changes as overlaid forms instead of release notes?
Because a release note is a claim about what moved and an overlaid form is the movement itself. An envelope that grew means the rule gained outcomes; one that shifted upward means the rule now treats a value as less likely; a modal line that moved means the assumption about the commonest case changed.
What would the first measured version look like?
The basis field would stop reading prior-assumption, the measured flag would stop reading false, and every prior would gain a sample count alongside the frequency that replaced it. Until all three of those happen at once, no page here may describe a score as measured.

Where the pens move

A versioned table tells you what changed on paper. Only the engine changes what your browser reports.

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.

Download Donut Browser Source on GitHub — AGPL-3.0 ↗