Split a secret into pieces SHARD

Any three of five, and two reveal nothing at all

The problem with a single copy of a secret is that it has two failure modes pulling in opposite directions. One copy can be lost, so you make more; more copies can be stolen, so you make fewer. Splitting solves both at once: choose how many shares exist and how many are needed, and the two numbers are independent.

Three of five means any three shares rebuild the secret, so two can be lost without consequence, and it means any two shares together are worthless. Not hard to break: worthless. Every possible secret of the same length remains exactly as likely as every other, so there is nothing to attack. That property is information-theoretic: it does not depend on how much computing power somebody has, and it does not weaken over the next thirty years the way a cipher might.

How it works, briefly

Each byte of the secret gets its own random polynomial of degree one less than the threshold, with that byte as the constant term. A share is that polynomial evaluated at the share's index. Any number of points equal to the threshold determines the polynomial and therefore the constant term, which is the byte; any fewer leaves every constant term equally possible.

The arithmetic is in GF(2⁸), modulo the same polynomial AES uses, so a byte in gives a byte out and nothing grows or rounds. This is real Shamir's Secret Sharing, not a scheme that splits the text into pieces; splitting a passphrase into three chunks and handing them out is not this, and leaks a third of it per chunk.

Where the real difficulty is, and it is not the mathematics

Shares that all live in the same drawer are one copy with extra steps. The scheme's value comes entirely from the shares being genuinely separated (different people, different buildings, different jurisdictions if it comes to that), and separation is a logistical problem you have to actually solve.

Think about it from the other end. Write down who holds each share, and ask what happens when you are not available to explain any of this. Does the person holding share two know what it is, that the others exist, who has them, and what threshold was used? A three-of-five split where only you know it was three of five is a secret you have successfully destroyed. Store the threshold, the total, and the list of holders somewhere separate from the shares and somewhere your executor will find it.

Then be honest about the failure you are protecting against. Splitting is right for a recovery key, a master passphrase, a root credential or a seed phrase: things that are catastrophic to lose and catastrophic to leak, and rarely needed. It is wrong for anything you use weekly, because reassembly is deliberately inconvenient and inconvenience is what makes people keep a copy of the whole secret on a laptop.

Everything happens in this tab; no share is ever transmitted. For the neighbouring pieces, SHADOWKEY says what a passphrase is actually worth, CIPHER encrypts a message for a channel you do not trust, and AIRGAP moves a share to a machine that is not on any network.