Five bands, four edges, no percentile
The edges sit at 12, 20, 28 and 36 bits, and your total lands in one of the five spaces between them. They are printed on the right-hand scale of the chart form before any run, and at run time the alarm pen over-inks in red the single edge your total actually crossed.
There is no percentile anywhere on this site, and the reason is not modesty. A percentile is a position within a measured distribution; this rubric has priors instead of a distribution, so the only honest positional claim it can make is where a total falls between four numbers that were published in advance.
blends in
0 –12Under this rubric's priors, nothing you present is individually surprising. That is not anonymity — it is the absence of a standout signal.
common
12 –20One or two signals carry real weight, but the combination is the sort the priors treat as ordinary.
distinctive
20 –28Where a stock desktop browser with a hardware GPU normally lands. The modal browser under this rubric sits in this band.
rare
28 –36Several high-surprisal signals at once, or a value the priors treat as tail on more than one axis.
singular
36 +The rubric's priors have run out of room. This is the top band because 36 is where the edges stop, not because anything was counted.
What a band does not tell you
It does not tell you how many other browsers are in it, because nobody here counted them. It does not tell you whether a specific site can identify you, because that depends on what that site collects and how it correlates it. And it does not improve because you moved from the top of one band to the bottom of the next — eight bits is a long way, and the edge is just a printed line.
What it is useful for
Comparing a configuration to itself. Take a pass, change one thing, press ADVANCE, and read the per-signal delta. A change that moves you across an edge moved something real; a change that moves the total by 0.4 bits mostly moved a value the table barely prices.
The reference point
A browser sitting at every rule's modal outcome — the likeliest value the table knows for each signal — totals 23.61 bits after discount and lands in “distinctive”. That is a property of the rubric, printed with the form. If your own total sits below it, the table's assumptions are treating you as more ordinary than its own idea of an ordinary browser.
Why the top band is where it is
“Singular” starts at 36 bits because that is where the printed edges stop, not because anything was counted at 36. With every prior clamped at 1/4096, thirteen signals cannot exceed 156 bits even in principle, and no real browser comes close — the total is bounded by how coarse the table is, and the top band is the honest name for “the assumptions have run out of room”.
What it takes to cross an edge
Eight bits. That is a specific quantity of change, not a vibe, and it is worth knowing what
it buys. Eight bits is roughly one loud column appearing or disappearing: a GPU renderer
string becoming readable is 8.64, a canvas hash is 8.97, and a three-language list
collapsing to en-US is 6.82 before any discount.
It is emphatically not a stack of small corrections. Normalising your colour depth, your touch points and your device memory together moves under two bits in most passes, and moves less than that once the capability and display groups discount their non-dominant members to 35%. If a change was supposed to cross an edge and moved 0.4 bits, the change landed on a column the table barely prices.
Why the ceiling is where it is
Every rule at its dearest outcome, discounted the way a real pass is discounted, totals 64.10 bits — which is inside the top band and nowhere near the arithmetic maximum of 156. The gap between those two numbers is the coarseness of the table itself. There is no configuration this rubric can express that goes higher, which is why the top band is named for the assumptions running out of room rather than for anything about you.
Where a band is genuinely useful
As a unit of comparison against yourself. The band survives a rubric version change in a way a raw total does not: if the priors move, both of your passes move with them, and the question you actually care about — did this configuration change do anything — is answered by the delta rather than by either number alone.
It is also the right granularity for a claim you are going to paste into an argument. “I moved from rare to distinctive after switching to engine-level spoofing” is checkable against a torn-off stub that carries every row of the derivation. “My score improved by 11%” is not a sentence this rubric can support at all.
Where a band is useless
Any time you want to know whether a specific site will recognise you. That depends on what the site collects, whether it correlates across sessions, and whether it has a record of you already — none of which this form can see. A rare-but-coherent browser is frequently less interesting to a vendor than a common one whose signals contradict each other.
How the correlation discount changes a total · the prior table each band is computed from · which columns can move you across an edge
Questions about bands and edges
Every figure below is a constant in the shipped scoring code or is derived from the prior table in this page.
- Why a band instead of a percentile?
- A percentile is a position in a distribution, and this rubric has no distribution. A band is a position between published constants, which is a claim the table can actually support.
- Are the edges meaningful, or round numbers?
- They are round numbers, chosen so that eight bits separate each band, and they are constants in the shipped code rather than parameters tuned to make a score look dramatic. That is the honest description of them.
- Which band does an ordinary desktop browser land in?
- Under this table, distinctive. A browser sitting at every rule's modal outcome — the likeliest value the rubric knows for each of the 13 signals — totals 23.61 bits after discount and lands in "distinctive". That is a property of the rubric, not an observation of visitors.
- What is the highest total this rubric can produce?
- Every rule at its dearest outcome, discounted the way a real pass is discounted, comes to 64.10 bits. The clamp at one in 4096 caps any single column at 12.00 bits, so 13 columns cannot exceed 156 even in principle — and the table is far too coarse to get near either figure.
- Does a lower band mean I am anonymous?
- No. A band is a statement about how surprising your values are under a printed set of assumptions. It says nothing about whether a particular site can recognise you, which depends on what that site collects, how it correlates it, and whether it has seen you before.
- How many bits separate one band from the next?
- Eight, at every edge. That spacing is a choice, and it means an edge crossing is roughly the weight of one loud signal appearing or disappearing — a GPU renderer becoming readable, or a three-language list collapsing to one.
Where the pens move
Crossing an edge downward takes a change the browser engine makes, not one a script makes.
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.