[Cryptography] ScrambNet: trading x n bandwidth for an information-theoretic content floor (seeking cryptanalysis)
ScrumbNet
scrambnet at proton.me
Wed Nov 11 06:19:28 EST 2026
Hello,
I'd value this list's scrutiny of a design wager, stated plainly so it can be
attacked. (I raised the networking side of this on tor-dev; here I want the
cryptographic core examined.)
Premise: every deployed anonymity system protects message content with a
computational assumption — break the cipher, now or with a future quantum
computer, and the content is readable. ScrambNet's content confidentiality is
instead information-theoretic. A message is split with Shamir secret sharing
into n shares (strawman n=4, k=3) and the shares travel over disjoint network
paths. An adversary holding fewer than k shares has, by Shamir's theorem, a
view that is independent of the secret: there is no ciphertext to attack below
the threshold, and no cryptanalytic or quantum progress erodes that. Onion
encryption is retained but demoted to a second belt; if it ever falls, the
content floor still stands.
I am aware of Schneier's Law and am not claiming a new primitive. The claim is
narrow and, I hope, boring at its base: Shamir (1979) composed with onion
routing so that the unconditional property, not the cipher, carries the
confidentiality load. The whole value — and the whole risk — is in that
composition and its honest accounting. So let me state the costs, not hide them:
- It buys content confidentiality only. Linkage/metadata is the same hard
problem it is everywhere, and in low-latency mode I make no unconditional
claim there.
- It costs x n bandwidth (x4 at the strawman): every share is as long as the
secret. This is Shamir's price, paid in full.
- The floor is only *operationally* real if the n paths are genuinely disjoint
(node, AS, jurisdiction). Verifying disjointness from a client's own data,
against relays that lie about everything except their signatures, is the
hardest open problem and only drafted.
- Share-set membership must be unlinkable to anyone but the recipient; I use a
recipient-keyed PRF tag. That construction deserves attack.
The design is public, RFC-style, and the load-bearing parts are a runnable
dependency-free proof-of-concept: the k-of-n dispersal with a brute-force
demonstration that k-1 shares leave all 256 candidate secret bytes equally
consistent, plus a k-path-adversary simulator. Everything is CC0.
Spec, PoC, threat model, numbered open problems:
https://github.com/ScrambNet/scrambnet
If the wager is wrong — if the composition leaks something the parts do not, or
the unconditional claim is weaker than I state — I would rather learn it from
this list than after someone relies on it. Thank you for reading.
More information about the cryptography
mailing list