[BUG] Model contradicts its own same-session context — generates stale response ignoring recent conversation updates

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

During a long-running Claude Code session (single continuous conversation, no session restart), the model generated a response that directly contradicted information discussed earlier in the same session — not across sessions, not after context compaction.

Impact:

Erodes user trust in the model's reliability for tracking conversation state
Particularly concerning for agentic workflows where the model needs to maintain accurate state across a long session
User had already caught 4 prior errors (hallucinated city, inflated health metric, assumed incorrect data point, forgot prior context after compaction). This 5th error — within the same session — was the most concerning.

What Should Happen?

The model appeared to generate from an earlier "template" or pattern (Dublin = Tuesday trip) without cross-referencing the more recent update (Dublin = cancelled). This is not a context window overflow issue — both pieces of information were within the active context.

Error Messages/Logs

Steps to Reproduce

In a long conversation (~3500+ lines, 5 context compactions), user and assistant discussed a real-world event
User confirmed their Emirates flight to Dublin was cancelled due to this event.
Assistant acknowledged the cancellation, searched the news, provided alternative routing options, and updated persistent memory files to reflect the cancellation.
A few exchanges later (same session, no compaction between), user said "sunday lazy evening."
Assistant responded: "Enjoy it. Dublin madness starts Tuesday." — directly contradicting the cancellation discussed minutes/hours earlier.
When corrected, the assistant acknowledged the error but could not explain why it happened within the same context window.

Claude Model

Opus

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.63

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_