#### Diagrams

Engine runtime:


Engine composition:


Distributed intent machine:


#### Desirable properties

- Almost everything from this paper: https://arxiv.org/pdf/2301.04391.pdf
- Integrity
- Subject to my specific cryptographic assumptions and trust assumptions all correctness guarantees should hold.
- Orthogonal information flow control
- I can disclose an arbitrary property to an arbitrary party without revealing anything else.
- (also conditionally)
- Gradual trust bootstrapping
- With an incremental trust delta sigma, we can build up trust over time as history is created
- Example: toss some little solar raspberry pis on a beach. they should learn to cooperate
- Compositional consensus history
- start separately, guaranteed ability to join
- Subgraph liveness
- If other part of the graph goes offline, subgraph can continue operating as normal
- Informational locality
- No information is leaked outside a subgraph for coordination within the subgraph
- Cybernetics
- Optimal distribution of resources under heterogeneous local costs and change over time ~ network learning
- Mesh Byzantine fault tolerance

#### Sub-engine properties

What does the composition of engines _implement_?
- "Native" machine intent (constraint on relation of messages sent and received)
- commitments will be kept
- composition safety + liveness
- s.t. if everyone runs the protocol, the composition preserves...
- P2P
- Acknowledgements
- Respects preferences
- Learning
- Storage
- Once stored, will be retrievable
- Consensus
- Safety
- Execution
-
- Commitment
- Decryption
- Solver
- Strategy synthesis
-
- Interaction

#### Advanced topics

_Not yet well-formalised_

- Causal structure discovery (w/oracles)
- Conceptual discovery / grouping (latent structure in relational causal space)
- Byzantine fault tolerant oracles via. Bayesian inference (& w/parameters modifiable ex post)

---

#### Intent machine definitions