AI agents

Memory expiry, invalidation, and staleness

• A Freshness Gate Before the Prompt: Fields and Refusals
• A Quarantine Table for Suspect Chunks: Schema and Retention
• Contract Tests Between a Policy Engine and Its Memory Store
• Freshness SLOs: A Staleness Budget per Source
• Invalidation Hooks for Business Data
• Memory Expiry and Decay: Scheduling What an Agent Forgets
• When to Cache Agent Memory: Trade-offs and Strategies
• A Freshness Gate for Agent Memory, Not Just Documents
• A Staleness Budget per Source: TTLs, Cutoffs and What They Cost
• A Staleness Dashboard: Age, Hit Rate and Cost per Source
• A Staleness SLO per Source: Threshold, Owner, the Page That Fires
• Adding a Freshness Check to Retrieval: Staleness and Source Age
• Approval Time Budgets for Held Chunks: When Waiting Costs More
• Caching Rewritten Facts per Query Class: Keys and TTLs
• Deciding Which Memories Need a Human Review Before They Expire
• Designing the Staleness Budget: Thresholds, Owners and Cost
• Detecting Stale Memory Before It Answers
• Expiring Memory by Source: TTLs That Match How Fast Facts Change
• Invalidating Memory on a Model Swap
• Invalidation Events That Bust Both Layers: A Publisher Contract
• Logging the Checker's Decision: Fields That Make a Row Replayable
• Publisher Contract for Deletion Events: When a Source Must Vanish
• Retracting a Memory After a Bad Answer: Find, Delete, Replay
• Routing Change Events to the Right Cache Keys
• Routing Fallbacks When a Source Blows Its Staleness Budget
• Simulating Eviction on a Replayed Access Log
• Sizing a Reconciliation Job From Change Rate and Cost
• Sizing TTLs From Change Frequency: A Source Census
• Source Age in the Log: Fields to Reconstruct a Stale Answer
• What to Log for Invalidation: A Schema
• When a Recalled Fact Contradicts the Rewrite: A Precedence Rule