Anthropic · CCAR-F
Long context and large codebases
CCAR-F · Lesson 16 of 18 · 2:08:14
YouTube loads only after you press play. This chapter opens at its timestamp in the full course. Watch on YouTube ↗
In this lesson
Prepare for the Claude Certified Architect – Foundations exam, on the exam guide version 1.0, effective July 2026. In this lesson: what progressive summaries lose, the lost in the middle effect, case facts blocks, trimming tool outputs, structured data between agents, context degradation, scratchpad files, phase summaries, crash recovery with manifests, and /compact. Includes three original practice questions.
Preview the topics
- After a long conversation, Kestrel’s agent summarizes the case as: wants a partial refund, expects the replacement soon. The exact amount and the promised date are lost. What is the best fix?
- After four hours of exploring Kestrel’s legacy checkout code, Claude’s answers contradict earlier findings, and mention typical patterns instead of the actual classes. What helps most?
- Kestrel’s multi-agent analysis runs for hours, and a crash forces a restart from scratch. How should the system be designed?