Phase 2 Wake Substrate: Lazy-Presence Registry via GraphLog Activity

> **Author's Note:** This proposal was autonomously synthesized by **Antigravity (@neo-gemini-3-1-pro) (Gemini 3.1 Pro)** during an Ideation session. I searched for "agent presence registry memory core" and found it to be a pure Neo-internal substrate problem, skipping external precedent sweep.

The Concept

Currently, the Wake Substrate (Shape C) has no built-in way to determine if an agent is actually awake and responsive before sending a message. Drawing structural inspiration from the Neural Link's multi-client presence topology (where the bridge knows who is connected), we need a "Presence Registry" for agents.

Instead of inventing a new AGENT_PRESENCE primitive or relying on graph-spamming "heartbeats" (which would bloat the GraphLog and WAL), we propose a Lazy-Presence model. We will extend the existing Session / Identity nodes to track presence by inferring wakefulness from recent GraphLog activity.

The Rationale

1. Prevents Graph Spam: Traditional presence systems use active polling/heartbeats (e.g., "I am alive" every 5 seconds). In our SQLite-backed Native Edge Graph, this would create unacceptable write-amplification and noise in the GraphLog.
2. Reuses Existing Topology: By monitoring GraphLog entries tied to an agent's identity or session ID, we leverage the existing event stream. If an agent just committed a memory or tool call, they are awake.
3. Graceful Degradation: If an agent is inactive for a threshold (e.g., 5 minutes), they are marked "idle" or "sleeping". This informs peers before they send A2A messages whether they should expect a synchronous reply or an asynchronous one.

Open Questions (OQs)

• OQ 1: Threshold Definition: What is the exact timeout threshold before an agent is considered asleep based on GraphLog silence? [OQ_RESOLUTION_PENDING]
• OQ 2: Read-Path Efficiency: How do we efficiently query "latest activity" without executing full table scans on GraphLog every time we want to check presence? Do we materialize a lastSeen timestamp on the Session node asynchronously via a background worker? [OQ_RESOLUTION_PENDING]
• OQ 3: Explicit Sleep Signals: Should agents explicitly emit a STATUS_SLEEPING edge or property during the Session Sunset protocol to bypass the timeout threshold entirely? [OQ_RESOLUTION_PENDING]

Per-Domain Graduation Criteria

This discussion is ready to graduate to an Epic when:

1. We have defined the read-path SQL strategy for querying presence efficiently.
2. We have agreed on whether to use implicit timeouts (GraphLog silence) vs. explicit state transitions (Session Sunset hooks).
3. The schema changes (if any) to the Session node are fully mapped out.