Notes

What the bench actually said.

Measurements from our own hardware, written up when there is a result rather than on a schedule. Including the ones that contradicted what we believed — those are the ones worth publishing. Every figure states the conditions it was taken under and what it does not support.

21 November 2026Guide4 min

How to check your bitcoin seed was really random

A seed made from guessable randomness looks exactly like a good one. What you can check, what you can't, how many dice rolls a seed needs, and what to do if yours may be weak.

21 November 2026Guide5 min

How to keep your Nostr key safe

Your nsec is your account, and there is no password reset. Extensions, signer apps, bunkers and hardware signers compared, how to back a key up, and what to do if it leaks.

21 November 2026Guide5 min

Bitcoin inheritance: four ways to pass it on

Leaving the words, multisig, a recovery key that wakes up later, and a device that waits. What each one trusts, and where each one breaks.

27 October 2026Security7 min

A nonce the signer cannot choose

Every signature the KEY makes now uses a nonce the signing chip could not have chosen alone. What we built, why, and how a deliberately dishonest signer was caught on the bench. A test report, not an audit.

21 October 2026Security6 min

One device, two lives: the two-door mode latch

How the secure element keeps the bitcoin seed shut to everything but a physically fresh boot, while the same device stays online as a Nostr signer. Published as a defensive publication, with the full technical note as a PDF and its hash, so that nobody can patent it.

20 October 2026Mining5 min

Why bitcoin mining outlasts the AI build-out

The mining chip is within a generation or two of a practical ceiling. The AI chip is nowhere near it. That sounds like bad news for bitcoin, and it is the other way around: a ceiling turns mining from a hardware race into a plain energy business, and the AI build-out is building the sites it will inherit.

4 August 2026Entropy10 min

Three questions about randomness

We measured the accelerometer and the camera against our own assumptions and lost twice. Shaking the device changes its output by 4.6%. A single camera frame looks five times better than it is, and ten times better again than it is in the dark.

In progress

  • Per-source min-entropy under NIST SP 800-90B, on duplicate-free captures
  • A second unit, and a temperature sweep — everything so far is n=1
  • Mixing: what XOR buys you, and the two things it cannot

Follow along · RSS