DictionaryArchitecture
Preserved Thinking
The AI's earlier reasoning is locked to the exact conversation that produced it, so rewriting history breaks it.
Definition
A rule on recent Claude models (Fable 5.1, Opus 5.5, Sonnet 5.5) that binds thinking blocks to the model and conversation that produced them, so editing earlier turns invalidates them. Agent harnesses should keep history append-only.
Example
A harness that trims an old assistant message on Claude Opus 5.5 invalidates that turn's thinking blocks; appending new messages keeps them valid.