Glossary

Context Rot

The degradation of a model's ability to accurately recall and use information as the token count in its context window grows — even in models built for long context. Nothing errors; the model just gets quietly less accurate, which is what makes it easy to miss until output quality has already dropped.

Anthropic's own term for the general pattern behind why longer context isn't free accuracy. It's the reason context management primitives — compaction, tool-result clearing, memory — exist at all: the goal is never fitting more in, it's keeping the smallest set of high-signal tokens that gets the desired behavior.