How it works

Safeparts is built around three simple goals:

1. Split a secret so no single share reveals it.
2. Reconstruct only when enough shares show up.
3. Fail loudly when inputs are wrong (wrong shares, typos, wrong passphrase).

Threshold sharing (k-of-n)

You choose a threshold (k of n):

• With fewer than k shares, reconstruction is impossible.
• With any k shares, reconstruction succeeds.

Under the hood, this is Shamir-style secret sharing over a finite field (GF(256)), applied byte-wise.

What a share is

Conceptually, each share is one point on a polynomial of degree k-1. With k points you can reconstruct the polynomial, and therefore the original secret. With fewer than k points, you learn nothing useful about the secret.

Why GF(256)

Safeparts works on bytes. GF(256) lets the math happen on 8-bit values, so secret sharing applies cleanly to arbitrary binary data.

Integrity

Secret sharing gives you confidentiality (until k shares), but by itself it does not reliably catch user mistakes. Safeparts adds an integrity tag so combine can detect incorrect inputs.

• If shares are corrupted or from different sets, combine fails.
• If you typed a share wrong, combine fails.

There are also extra checks at the encoding layer:

• Base58Check includes a checksum.
• Mnemonic shares include CRC16 to catch many transcription mistakes.

Optional passphrase protection

Safeparts can encrypt the secret before splitting it.

High level:

• Derive a key from the passphrase (Argon2id).
• Encrypt the secret (ChaCha20-Poly1305).
• Split the encrypted bytes into shares.

This means an attacker needs both k shares and the passphrase.

Note

Passphrases add a second requirement; they do not replace good share separation.

Encodings

Shares can be encoded for different workflows:

• machine-friendly (copy/paste): base64url, base58check
• human-friendly (transcription): mnemonic formats

See Encodings.

Packet metadata (why shares must match)

Shares are not just chunks. Each share carries enough metadata to prove it belongs to the same set:

• threshold (k) and share count (n)
• set identity / versioning
• integrity data

That is why mixing shares from different splits should fail loudly.

What Safeparts does not do

• It does not keep shares safe for you. Storage choices are the security boundary.
• It does not protect you from someone who legitimately holds k shares.

Note

For a deeper security-focused explanation (data flow, algorithm choices, and recommended procedures), see Technical design.