The composition lattice
Each primitive plugs a specific gap. Knowing what each one does (and does not do) is what lets you compose them safely.
| Primitive | Hides | Doesn't hide | Pairs well with |
|---|---|---|---|
| TLS | Wire content | Endpoints, sizes, timing | Mixnet, attestation |
| TEE | RAM from OS or cloud | Side channels, vendor backdoor | Attestation, transparency log |
| FHE | Plaintext during compute | Existence and size of inputs | ZK proof of correctness |
| ZK proof | Witness | Proof statement and result | FHE, on-chain verification |
| PIR | Index i |
Database contents | Relays, mixnet for transport |
| Mixnet | Network metadata | Volume over long windows | Sealed sender, padding |
| Onion routing | Per-hop knowledge | Timing correlation | Padding, mixnet |
| Signal Protocol | Per-message keys | Recipient identity (without sealed sender) | Mixnet, MLS for groups |
| Decentralized name | Identity from registrar | Identity from blockchain analysis | Onion routing, PKARR |
| Content-addressed storage | Byte substitution | Availability or confidentiality | Encryption, pinning, decentralized names |
| Self-custody key | From custodian | From your own device compromise | Hardware wallet, multisig |
Common compositions
Confidential AI inference
Decentralized name (.eth / PKARR / .onion)
↓
Attested TLS bound to enclave-generated key
↓
TDX or SEV-SNP confidential VM
↓
Confidential GPU mode with protected CPU–GPU transfers
↓
Open-weights LLM + reproducible build hash
↓
Transparency-logged release (à la Apple PCC)
Files: Confidential-AI-Inference, Private-LLM-Inference-Patterns, NVIDIA-GPU-TEE, Attestation-Architecture.
Private query over public data
Client query
↓
PIR (single- or multi-server)
↓
Optional: ZK proof of correct database state
↓
Transport: Tor or Nym
Files: Overview - Private Information Retrieval, Overview - Mix Networks.
Censorship-resistant payment
Bitcoin keypair (self-custodied)
↓
Lightning channel
↓
Onion-routed HTLC payment
↓
Receiver: BOLT12 or LNURL with attested proof
Files: Overview - Financial Sovereignty, Lightning Network Deep Dive.
Sovereign social account
Nostr keypair (signing key)
↓
NIP-17 gift-wrapped DMs (sender hidden; recipient tag visible)
↓
Blossom for media blobs (SHA-256-addressed, mirrored)
↓
Public-key identity; optional NIP-05 alias via DNS and HTTPS
↓
Multiple relays (transport redundancy)
↓
Lightning zaps for monetization
Files: Overview - Encrypted Messaging, Nostr Deep Dive, Blossom, Overview - Decentralized DNS.
Sovereign off-grid node (Blackbox-style)
Local LLM inference (Ollama or equivalent, no cloud API)
↓
LoRa mesh (Meshtastic / MeshCore / Blackout Comms)
↓
Offline maps (TAK / equivalent, no Google)
↓
Cashu ecash + Bitcoin (blind tokens + self-custody, no bank)
↓
E2E encrypted request/response chunked over mesh (~200B packets)
One laptop + $30 radio. Zero internet, zero banks, zero datacenters. See MeshCore, Cashu and Chaumian Ecash, off-grid files, and MOC - Censorship Resistance. Composes local confidential compute with air-gapped comms and private money.
Durable public artifact
Release artifact or static site
↓
Hash and sign locally
↓
IPFS for content-addressed distribution
↓
Arweave for archival copy
↓
ENS / PKARR / signed index points to both
Files: Overview - Decentralized Storage, IPFS and Filecoin, Arweave, Overview - Decentralized DNS.
Verifiable confidential computation
Client sends encrypted input
↓
Server runs computation under FHE
↓
Server attaches a SNARK proving the function was the agreed one
↓
Client verifies SNARK in milliseconds, then decrypts result
This is the vFHE (verifiable FHE) pattern. Either of these alternatives also works:
- FHE evaluator inside an attested TEE. The TEE attestation replaces the SNARK.
- Threshold FHE with active-secure honest-majority MPC. The protocol detects misbehavior.
Files: Overview - Fully Homomorphic Encryption, Overview - Zero-Knowledge Proofs, Overview - Confidential Computing.
Hardware-protected ecash mint (deployment pattern)
Bitcoin/Lightning reserves (self-custodied or on-chain visible)
↓
Cashu mint logic running inside attested TEE (keys generated in enclave, never exposed to host)
↓
Remote attestation + transparency log (client verifies correct code + measurement)
↓
Blind signatures issued; users hold bearer tokens offline / over Nostr / mesh
↓
Redemption still depends on the mint; evaluate code, reserves, rollback and hardware trust
Direct 01 + 06 composition. See Cashu and Chaumian Ecash and confidential computing attestation files. Moves from "trust the mint runner" to "trust the hardware root + audited enclave code."
Post-quantum migration of everything
For each long-term-secret protocol (TLS, SSH, Signal, age-encrypted backups):
1. Switch to a hybrid KEM (X25519 + ML-KEM-768/1024)
2. Rotate signing keys to ML-DSA or hybrid Ed25519+ML-DSA
3. Audit the symmetric key sizes — AES-256 stays
4. Plan re-encryption for archived ciphertexts under new keys
Files: Overview - Post-Quantum Cryptography, Glossary (Hybrid KEM, Harvest now decrypt later).
Anti-patterns
These look decentralized or private only on the surface:
- Custodial "non-custodial": exchange holds your keys but calls itself non-custodial.
- CDN'd onion service: operator's identity revealed via Cloudflare logs.
- Signal Desktop on a compromised OS: E2EE does not help if the endpoint is malware.
- Nostr without NIP-17: DMs leak the social graph to relays.
- TEE without attestation enforcement: the server runs in a TEE, but the client never checks.
- FHE without verifiability: the server can return garbage and the client cannot tell.
- Self-custody without a backup plan: losing the key is the same as a custodian losing your funds.
Files most relevant
Every Overview in this library. Start with Research Methodology's evaluation criteria, then pick the relevant overviews and build a stack.
Primary sources
- RFC 9334: attestation architecture
- RA-TLS: binding attestation to TLS
- Nostr NIP-05: DNS-based identifiers
- Nostr NIP-59: gift-wrap metadata
- Cashu NUT-00: protocol and mint trust
- Lightning BOLT 4: onion routing
- NVIDIA: Confidential Computing on H100 GPUs
Also in Confidential Computing
Page history
- Updated
- 2026-10-25
- Origin
- Compiled from the ungovernable.tech research notes.
- Improve this page
- Suggest an edit — issues and pull requests are welcome.