Sigilcore · the seal

The key that is nowhere

A radio's identity seed is split across the other radios in its formation. To sign it must assemble a quorum that is actually present, and it holds the reconstituted seed for one operation before wiping it. A captured radio cannot even read its own message history.

At a glance
For
Signals branches, COMSEC authorities
Replaces
Fill devices, KEK hierarchies, net-wide re-key drills
Removes
The long-term key on the device
Root of trust
Held by the customer
The problem it removes

One captured radio should not mean a net-wide re-key

Today the private key is on the device. Encrypt it, wrap it in a TPM, demand a PIN: it is still there, so the adversary's best play is not cryptanalysis but capture. The response is a net-wide re-key measured in days and run under contact, which is exactly why units delay reporting a loss.

Sigilcore makes losing a radio one revocation entry plus one epoch refresh. In our assessment that single change is worth more to a signals branch than any cryptographic improvement, because it removes the reason to delay reporting.

How it works
  1. Split

    At enrolment the identity seed is Shamir-shared t-of-n across custodians in the formation, then wiped. No device holds it.

  2. Assemble

    To sign, the radio requests shares from custodians that are present. Only a quorum of current-epoch shares reconstitutes the seed.

  3. Sign and wipe

    The seed exists in memory for one operation, then is overwritten. The vault key for message history is quorum-derived the same way.

  4. Refresh

    Every epoch each custodian deals a fresh sharing of zero. The secret is unchanged; every share becomes new, so slow capture campaigns gain nothing.

IN FORMATION NODE holds nothing 3 of 6 answer req shares SEED in memory only SIGN one operation WIPE seed gone lifetime of the key ≈ microseconds CAPTURED — REMOVED FROM THE FORMATION NODE in enemy hands 0 of 6 answer req ✕ QUORUM UNAVAILABLE The correct passphrase is not sufficient, and never was. The key material is in six other rucksacks. None are here. Message history: unreadable — the vault key is quorum-derived too.
Identity exists only while the formation does. Upper path: a quorum answers, the seed is reconstituted in memory, one operation is performed, the seed is wiped. Lower path: capture. No zeroization routine, no timer, no race, no dependency on the loss having been reported.
Capabilities

What the formation gets

Inert on capture

Powered on, with the real passphrase, the device still holds nothing usable: the key material is in other rucksacks.

Patience stops working

An adversary must take t custodians inside one epoch. A radio a month for a year yields twelve shares that never combine.

Duress is expressible

A second credential opens a decoy corpus, revokes the node across its custodians and flags the net, while the captor sees a working radio.

Revocation, not re-key

Losing a radio is one revocation entry and one epoch refresh. Nothing else in the net changes.

Customer-held roots

Enrolment authority, PKI and every key ceremony belong to the customer. Irythra holds no master key.

One piece of hardware

The single sovereign root of trust in the stack, so the supply chain to establish is one component.

Specifications

Technical summary

IdentityEd25519 / ML-DSA-87 signing seed
SharingShamir t-of-n over GF(2⁸); threshold and formation size set by doctrine
RefreshProactive, per epoch; stale shares do not combine
DuressSecond credential: decoy corpus, silent revocation, net flag
Isolation policySOLO_GRACE_EPOCHS 0–3: bounded, logged, self-expiring
StatusProtocol real Hardware not started

The cost, stated plainly

A radio genuinely cut off from its formation cannot originate traffic. The mitigation is a bounded solo grace period: zero for strategic nets, two or three for dismounted infantry who routinely lose contact. A radio cannot tell isolation from capture, so a captured radio inherits that same window.

And taking t custodians inside one epoch, as a section-level ambush might, compromises the identities they hold shares for. Formation size and threshold are doctrine decisions, and we would rather argue them in a trial than default them quietly.

Read the Sigilcore section of the architecture note →

Get in touch

The binding constraint is a signals regiment willing to run a trial.

Not capital. If you can open that door, or want the architecture note and the runnable prototype to take apart yourself, we would like to hear from you.