Statement

An Encrypted Semantic Island is a space resolvable only through its exact derivation. Its audience is not a permission a custodian grants — it is a cryptographic set, composable by union or intersection, independent of where the ciphertext physically resides.

Definition

I = (path, ciphertext, T, A, C)

A = A₁ ∪ A₂ ∪ ... ∪ Aₙ     (any member resolves independently)
A = A₁ ∩ A₂ ∩ ... ∩ Aₙ     (all members required, jointly)

Same tuple as the base Island. The addition: A is no longer atomic — it is itself a set expression over kernel identities.

Composition

An Encrypted Semantic Island is what these three produce composed: a space (manifest) whose audience (_) is a set expression (union/intersection) over independent kernel identities.

Core Relation

T ⊥ A

Topology and audience are independent variables. The party holding ciphertext at a given T is not, by that fact, a member of A. Holding the bytes and deriving their meaning are separate capabilities.

Contrast: Policy Custody vs. Derivation

Policy custody (ToS)Encrypted Semantic Island
readabilityf(custodian intent)f(key derivation)
T and Asame partyindependent (T ⊥ A)
enforcement pointexternal checkpoint, bypassablefused into decryption itself
failure modesilent policy violationdecrypt failure
what changes behaviortrust in a promisepossession of derived key material

Scope — What This Does Not Resolve

T is not a kernel data structure; where ciphertext physically lives remains external infrastructure. Constructing a union or intersection A requires each principal's P256_pub in advance — a key-distribution problem the kernel does not solve. Removing a member from a union A requires re-encryption for backward secrecy; intersection A requires out-of-band coordination to reassemble shares. None of this is promise-based — all of it is a protocol dependency, not a trust dependency on a custodian who already holds the plaintext.

For why this distinction matters in practice, not just in formalism: manifesto.html and The Byzantine Prompt.