Season 3: Memory + Guardian

Theme

Upgrade Seraph from stateful assistant to long-horizon guardian.

Why this season matters

This season turns Seraph's memory and observer systems into a richer human world model, which is what ultimately separates a guardian from a merely capable agent runtime.

Win condition

By the end of this season, Seraph should understand more than messages and retrieval snippets. It should be reasoning over commitments, projects, patterns, and intervention outcomes with better timing and better personalization.

Batches

• S3-B1 Human World Model
• S3-B2 Observer Deepening
• S3-B3 Learning Loop

Related sections

• Memory + Guardian Intelligence
• Trust + Capability

Exit criteria

• memory is more relational and less purely vector-based
• observer signals carry more behavioral depth
• proactivity adapts based on outcomes, not only heuristics