跳到主要内容

本页面暂无您所用语言的版本,您正在阅读英文原文。

Glossary

Entropy

It is the idea behind every claim that some readings matter more than others. One bit halves the crowd, and the readings that halve it many times over are the ones worth spending effort on.

Entropy is the measure of how far one reading narrows the crowd a browser could be hiding inside.

What a bit actually buys

A reading that splits all visitors into two equal halves carries one bit. Ten such readings, if none of them followed from another, would leave one visitor in a thousand.

  • Rarity is the whole idea

    A value almost everyone shares narrows nothing, however technical it looks on a report.

  • The figure belongs to a crowd

    Bits describe a population somebody measured, so the number depends on who was measured and when.

  • Nobody sits at zero

    A browser answering ordinary values for everything still carries the readings a site cannot help seeing.

Why readings do not simply add up

Two readings that follow each other carry far less together than apart. The claimed system and the set of installed fonts overlap heavily, so learning the second after the first adds little.

  • Overlap costs you almost nothing

    Readings that agree by nature amount to close to one reading for the purpose of narrowing a crowd.

  • Contradiction costs a great deal

    A pair that cannot occur on any real machine is worth more to a site than either reading on its own.

  • Stability is a separate axis

    A rare value that changes every visit links nothing, and an ordinary one that never changes links everything.

Your readings right now

These values are read in your browser on this page and nothing leaves it. The first three narrow a crowd sharply, and the last one barely at all.

此刻您的设备

Canvas hash
正在您的设备上测量
Graphics renderer
正在您的设备上测量
Time zone
正在您的设备上测量
Colour depth
正在您的设备上测量

在您的浏览器中读取并在此显示,不会发送到任何地方。

Where the word appears in a browser

Browsers split the fields they report about themselves into cheap ones sent with every request and costly ones answered only when a site asks. That split is written in terms of this idea, and it is where the word shows up in the plumbing rather than in an article.

  • The cheap fields travel by default

    The browser name and its coarse version ride along with ordinary requests, so they are always aligned with the profile.

  • The costly ones are requested

    The precise version, the architecture and the device model are answered on demand, and our engine rewrites them only when a request already carries them.

  • Asking is itself a signal

    A site that requests the costly fields has declared its interest, and the request is visible to anything watching the traffic.

What the app does about it

Every reading a profile answers with is derived from one seed belonging to that profile. Two profiles therefore describe two machines, and each profile stays the same machine for itself.

Agreement is the part that can be checked rather than argued about, and the self-test checks it reading by reading against that seed.

  • One seed per profile

    The values are derived at launch from the profile, not stored as a file that can be copied away from it.

  • Readings are compared against each other

    A stand asserts relationships instead of values, because an ordinary reading beside the wrong neighbour is the loudest thing on the page.

  • The free check marks the sharp rows

    On the fingerprint check the rows that tell you apart carry a mark, and the mark is a name rather than a score.

How to read a figure you are quoted

Numbers in bits come from studies of the visitors of one site at one time. They describe that crowd, and your own browser is being compared against a population nobody has measured.

  • Ask who was counted

    A figure from a privacy study describes the people who came to that study, who are not ordinary visitors.

  • Ask when

    Browsers change what they report every few releases, so an old figure describes browsers that no longer exist.

  • Ask what was added up

    Sums taken over readings that overlap are larger than the crowd they claim to narrow.

What this does not do

  • The app does not measure your entropy. Nothing in it counts bits or scores how far you stand out, and any figure quoted for a reading comes from someone else's population rather than from your machine.
  • It does not lower a figure by hiding a value. A browser refusing a reading answers in a way most browsers do not, and refusal is rarer than any answer.
  • It does not decide which account gets closed. Behaviour, address reputation and account history weigh more than every reading on this page together.
  • It does not compare across checkers. Two tools read different sets and weigh them differently, so their results are not the same quantity.

Check it

Every claim above comes from a file you can open.

Where one profile's readings are derived from
packages/shared/src/index.js
The readings the self-test compares against the seed
apps/desktop/src/lib/fpAudit.js
The readings a profile has to keep in agreement
apps/desktop/test/fingerprint-consistency.js
What the free check reads and which rows it marks
apps/web/components/tools/FingerprintCheck.tsx

Questions

How many bits is my browser worth?
We do not put a number on it, because nothing in the product counts one. A figure taken from a public study describes the visitors of that study, not your machine.
Does blocking a reading help?
Often the opposite. Refusing to answer is rarer than answering, so a blocked reading can narrow the crowd around you instead of widening it.
Is an ordinary value always safer?
Only while it agrees with its neighbours. A perfectly ordinary graphics card beside a processor count that machine never had is worse than either value alone.
Do two profiles share the same readings?
No. Each profile derives its answers from its own seed, so the descriptions differ while each one stays steady for itself.
Does a proxy change any of this?
The clock does. A profile takes its time zone and its language list from the country the proxy leaves through, while the canvas hash, the graphics renderer and the colour depth are produced on your own machine.

See what your own profile answers

The self-test runs inside a profile and shows all 29 measurements, green and red alike.