A customer will ask what is real, and we would rather volunteer the answer than have it extracted. This is our status as of November 2026. The live version is always on the engineering status page.
Real and tested
- Authenticated encryption, key derivation and signatures: ChaCha20-Poly1305, HKDF and Ed25519, known-answer tested against their RFCs and cross-checked against an independent implementation.
- Secret sharing with proactive refresh: any t of n reconstruct, t−1 leave every secret possible, and shares from different epochs do not combine.
- Erasure coding: MDS Reed–Solomon, with every k-subset of 300 random subsets decoding and every (k−1)-subset refusing.
- Dispersal: the seal, code and share-the-key construction described in Erasure coding is not encryption, with tamper detection and a per-bearer cap.
- Release policy tree and UAS ledger: per-field policies over attributes, and hash-chained signed provenance with tamper detection.
The same engine now runs in the browser. Every figure on this website is computed live by it, and the reference lab exposes all five modules for anyone to test.
Not yet real
- ML-KEM-1024 is a sized stub with no post-quantum security. Its buffers match FIPS 203, so link budgets and shard sizes are honest, and it is swapped for a validated module at integration.
- Module validation and formal verification have not started. Models of the quorum unlock and the hybrid handshake are planned.
- Hardware integration and accreditation have not started. They are the work that turns any of this into a fielded system.
- Irythra Watch is in design.
The reference code is correctness-oriented. It is not constant-time and not side-channel hardened, and it must not be used to protect real traffic.
Open problems
- A radio genuinely cut off from its formation cannot originate traffic, and cannot tell isolation from capture.
- Constant-rate cover at ≈4.8× is painful on a dismounted battery over 72 hours.
- Quorum assembly is fine at message cadence and wrong at packet cadence; caching a session key weakens the capture property.
- The Release construction suits a single national authority, not multi-authority federation.
- Enumerating bearers is itself observable on some of them.
Next
In the next six months: a certified crypto core through a validated module, and formal models of the quorum unlock and handshake. The kill criterion for that phase is plain: if quorum latency cannot be made workable at message cadence, we will say so.