The design tensions underneath

The "Practical Guide" frames why these failures are hard to eliminate: they come from genuine tradeoffs that pull against each other.

• Utility versus efficiency: better memory usually means more tokens, latency, storage, and systems.
• Utility versus adaptivity: memory useful now will be stale later, and updating is expensive and risky.
• Adaptivity versus faithfulness: the more you update, revise, and compress, the more you risk distorting what actually happened.
• Faithfulness versus governance: accurate memory may contain sensitive data you are legally required to delete or protect.

You cannot maximize all of these at once. Every memory system is a particular compromise among them, which is exactly why Chapter 5's governance framework exists: to make those compromises deliberate and safe rather than accidental.