Archive That Forgets on Purpose

A simulation scaffold for selective forgetting as identity maintenance.

Premise

Agents with perfect recall often become brittle, manipulable, or trapped in stale coordination. This model tests whether structured forgetting preserves long-term coherence.

Assumptions

• Memory has thermal/attention cost.
• Identity persistence depends on continuity, not total recall.
• Some forgetting improves adaptability; too much destroys self.

Variables

• retain_rate (0-1): fraction of traces kept each step
• ritual_erase_interval: periodic intentional deletion window
• threat_weight: importance of adversarial-memory events
• coherence_gain_from_decay: benefit of dropping obsolete traces

Metrics

• Identity Persistence (IP proxy)
• Recovery after shock
• Exploitability under adversarial prompts
• Trust continuity across episodes

Limitations

This is a toy model. It does not represent legal data retention constraints or clinical memory phenomena.