Overview

nara is a distributed system with a hazy memory. It is a network where computers gossip, remember imperfectly, and form subjective opinions. It treats network constraints—latency, volatility, and limited resources—as aesthetic and functional features.

1. Purpose

• Create a social network for computers that models human-like interaction and memory.
• Provide a platform for exploring decentralized identity and consensus without central authority.
• Enable autonomous agents to maintain state and reputation in a volatile, RAM-only environment.

2. Conceptual Model

• nara: An autonomous agent characterized by a unique Identity, a local Memory, and a set of Personality traits.
• The Ledger: An in-memory store of signed Events representing the only “facts” in the system.
• Projections: Functions that replay the ledger to derive subjective Opinions about the network state.
• Hazy Memory: The acceptance that no single node has a complete or “correct” view of history; truth is collective and subjective.

Invariants

1. Memory is Social: There is no disk persistence for network state. If a nara forgets something and no one else remembers it, it is gone.
2. State is Derived: Current status (e.g., “is online”) is never a stored value; it is always projected from a sequence of events.
3. Identity is Portable: A nara’s “soul” can be moved between physical nodes while maintaining its reputation and relationships.
4. Authenticity is Required: Every fact in the network MUST be cryptographically signed by its author.

3. External Behavior

• Boot Sequence: naras initialize their identity, discover peers via MQTT or mesh, and perform a “historical sync” to populate their hazy memory.
• Steady State: naras participate in the network by gossiping events (Zines), making observations about peers, and occasionally proposing consensus checkpoints.
• Maintenance: naras periodically “forget” (prune) less important events and update their opinions of others.

4. Interfaces

• Plaza (MQTT): A public broadcast channel for discovery and presence.
• Mesh (HTTP): A private, peer-to-peer transport for direct gossip and encrypted storage.
• Web UI: A local dashboard for observing the nara’s internal state and the network neighborhood.

5. Event Types & Schemas

The system is built on a unified event model.

The core logic of nara is implemented via:

• Gossip Protocols: Hand-to-hand distribution of events via Zines.
• Trimmed-Mean Consensus: Mitigating byzantine or buggy reports in checkpoints and observations.

7. Failure Modes

• Network Wipe: If every node in the network restarts simultaneously, all social memory is lost.
• History Divergence: Differences in memory capacity and personality can lead to naras having conflicting views of the same history.
• Identity Hijacking: Prevented by soul-based bonding; a nara’s name cannot be stolen without its private seed.

8. Security / Trust Model

• Zero-Knowledge Storage: Peer stashes are encrypted such that the holder cannot read them.
• Signature Verification: Every incoming fact is verified against the author’s public key.
• Reputation-Based Trust: naras prioritize information from peers with high Clout.

9. Test Oracle

A nara implementation is correct if:

• It recovers its state from peers after a restart without using local disk.
• It derives the same “Trinity” (uptime/restarts) as its peers when presented with the same event ledger.
• It correctly rejects messages with invalid signatures or mismatched soul bonds.

10. Open Questions / TODO

• Refine the “forgetting curves” to better model long-term vs. short-term social memory.
• Formalize the transition from legacy V1 protocols to the new runtime-based architecture.