What
Claude Code reports how full the context window is, meaning how much of the model's working memory the conversation uses. That report can now include after_boundary, the ID of the compact-boundary message at the start of the conversation. A compact boundary marks where an earlier part of the conversation was replaced by a summary.
- An estimate taken after a boundary is skipped until a measured value has been written
- The context meter uses an estimate only if it belongs to the same boundary as the measured value, and keeps the measured value's other fields
after_boundaryis left out for sessions withnoHistoryBackfillorforeignset
Why
The context meter is less likely to mix a figure from before a compaction with one from after it.
Something disagreesSomething we can check disagrees with this entry, or the writer said they could not settle it.
The writer flagged doubt
The finding does not say what a `foreign` or `noHistoryBackfill` session is from a user's point of view.