soft3/mudra/specs/identity.md

identity

Cyber uses private proofs of programmable authority. a default authority commits to a secret with Hemera; Zheng proves that the owner satisfies the policy for an exact action. proof-based authority and Mudra's encryption/key agreement have separate roles.

authority, subject and profiles

authority_commitment = profile_hash(secret)
authorization = private_proof(policy(authority, action, state, witness) = accept)

the authority commitment is bound to a versioned policy. a stable neuron subject and a rotatable authority key are separate objects: rotation updates authorized policy state rather than silently renaming the subject.

native secret generation uses 256 bits of entropy. its KDF, hash domain, encoding and output size belong to the selected profile. field width, digest length and entropy are separate parameters.

existing identifiers and compatibility profiles retain their defined meaning. this specification does not reinterpret a secp256k1-derived neuron ID as a hash-preimage identity. migrations and recovery follow an explicit authorized transition. see the neuron identity contract.

exact authorization statement

the verified statement binds at least:

protocol / cryptographic profile and version
network / genesis / execution environment
expected program and policy identity
canonical action, inputs and outputs
accepted state root and applicable policy revision
nonce / operation identity and validity interval
required public result

an execution proof must bind the expected relation and these values. proving preimage knowledge without action binding is insufficient for message authorization. a valid proof for one action cannot authorize another action, network, root, program or policy revision.

the witness contains the secret, policy credentials and authenticated state openings. private authority requires a zero-knowledge proof profile; a public execution trace containing the secret cannot fulfill this contract.

verification and replay

a verifier checks the expected profile/program, the proof, the action's domain and current policy/state rules. replay and double-spend prevention use authenticated nonce/nullifier state. durable admission and retries retain the same operation identity; proof bytes do not define that identity.

knowledge soundness, statement binding and zero knowledge are requirements of the selected proof construction. preimage resistance protects secret ownership; collision resistance protects relevant commitments. transcript and recursive verification assumptions also remain explicit. the full system is not reduced to collision resistance alone.

programmable policies

the same proof interface supports single owners, thresholds, delegation, timelocks and recovery:

policy required relation
single owner secret opens the authorized commitment
threshold sufficient distinct authorized credentials
delegation valid scope, revision, expiry and delegate authority
timelock authority plus authenticated chain time/height
recovery authorized recovery quorum and policy transition

policy configuration and transitions are authenticated state. wall-clock assertions from an untrusted host do not satisfy a chain timelock. performance, proof sizes and recursive costs are measured for the selected program/backend; they are not fixed by this abstract interface.

private authority

a public permanent subject identifier links its actions. the private profile proves membership and policy satisfaction against a committed authority set while keeping the member, ownership opening and private values hidden.

nullifiers bind their intended scope and prevent replay without exposing the authority commitment or discovery key. private output values and individual contributions remain private witnesses. plaintext per-action weights cannot be labeled hidden merely because authorship is hidden.

public roots and authorized aggregate releases have an explicit leakage contract. exact aggregate changes can reveal an isolated contribution. the enclosing BBG/graph protocol must analyze this disclosure together with transaction timing and side information.

selective disclosure uses a purpose-bound proof or scoped credential. it must not reveal root secrets, reusable viewing keys or derivation material that exposes unrelated actions.

encryption and recovery

  • seal encapsulates a fresh secret for a published recipient key.
  • stealth derives pairwise secrets by a validated commutative action, including per-payment ephemeral delivery.
  • veil supplies profile-specific private computation.
  • quorum supplies the selected threshold distribution protocol.

a sender knows a delivery shared secret. spending authority therefore uses an independent owner secret and verifier-checkable policy binding. static NIKE or a channel MAC alone does not provide transferable public authorization.

Private recovery defines discovery, authenticated history coverage, spend-state verification and key lifecycle. a spell reconstructs keys; the recovery service must also preserve the required ciphertext history/state.

protocol qualification

each deployed profile fixes action encodings, key derivation, hash/proof parameters, privacy disclosures and classical/quantum resource assumptions. the abstract choice of proof-based authority is independent of the measured latency or current build status of an implementation.

source/version comparisons and measurements live in the architecture assessment.

Homonyms

identity
identity a neuron's identity is the Hemera hash of its public key — no registration, no authority, collision-free across the planet. see cybergraph/reference/identity. identity is uniform across every regime in which a neuron operates. the mechanism that authenticates a message varies with the…
cyber/identity
cybergraph/reference/identity

Graph