Home · Biometric Entropy Harness
Room · randomness from a body
The Biometric Entropy Harness
Where do keys come from when you do not trust the machine's own randomness? This demo harvests timing from a human hand, measures it, and shows you the audit.
Cryptography has a quiet dependency that most people never think about: every key starts as a random number, and a computer is bad at being random. It is a deterministic machine doing arithmetic. So every serious system harvests randomness from the physical world — disk timing, network jitter, a hardware chip that measures thermal noise — and then measures how much randomness it actually got.
This harness is a demonstration of the human version. A person's hand is a physical process: keystroke intervals, the tiny pause before a decision, the arc of a pointer. None of it is deterministic. The page collects that timing, runs it through a conditioning step, and shows you the audit, so you can see the difference between "we collected some numbers" and "we collected enough entropy to seed a key".
The live demo is the interactive harness. This page is the honest manual, including the part where I tell you what it cannot do.
What it measures
Three streams, all of them timing rather than content. Inter-key intervals while you type a phrase. Pointer path deviations, which carry small motor corrections nobody consciously controls. And pause distributions, which are the closest thing to a signature a human hand has.
There is a reason the harness never asks for a fingerprint and never records what you type (beyond the phrase it gives you). A biometric system that stores the biometric is a liability; a system that measures the timing of a hand and immediately reduces it to a hash is not storing anything about you that survives. That distinction is the difference between an authentication product and a demo worth publishing.
What the audit shows
Raw timing is not uniform. Human pauses cluster; intervals have a heavy tail. Feeding raw measurements into a key derivation function gives you a weak key with a confident face. So the harness runs a conditioning step and then estimates min-entropy — the conservative version of "how many bits do we really have". You can watch the estimate climb as you type and stall when you stop, which is a better intuition pump than any diagram.
The demo also shows the failure mode on purpose. Mash a key with a metronome, or let an automated script type, and the entropy estimate flattens. That is the lesson: the source is only as good as the unpredictability of the process behind it, and a human who is trying to be random is worse than a human who is not.
What it does not prove
It does not prove that a browser is a secure environment. JavaScript timing is coarse, a hostile page can pollute measurements, and no demo should be used to generate a key that protects something real. The demo is a teaching instrument: it makes visible a step that production systems hide.
It is also not a medical device and does not diagnose anything. The word "biometric" makes people assume health, and I would rather say plainly that this measures intervals, not bodies. If a page ever claims to read your physiology from typing speed, close it.
Why publish a security demo at all
Because the alternative is a paragraph of assurance. A page that says "we use strong randomness" is unfalsifiable. A page that shows you the collection, the conditioning, and the entropy estimate can be argued with, and being arguable is the point. This is the same policy as the rest of the workshop: publish the receipt, not the adjective.
The fiction's version of this idea is the seal — a guarantee that can be pointed at later. The world page explains seals and chains; this page is the same argument with numbers instead of dragons.
If you want the general rule, read Systems that break early. If the vocabulary is thick, the glossary defines seal, hash, and dual ledger in one line each.