Domain 5: Context Management & Reliability

Domain 5 is the smallest by weight and the one that punishes vagueness hardest. Its recurring insight is that context is lossy in specific, predictable ways — summarization eats numbers and dates, long inputs lose their middles, verbose tool results crowd out what matters, and attribution disappears at the moment findings are merged. Every answer in the domain is a structure that survives those losses: a case-facts block outside the summarized history, key findings at the top of an aggregated input, a claim-source mapping that downstream agents must preserve, a scratchpad file that outlives the context window.

What the guide measures here

1. Manage conversation context to preserve critical information across long interactions
2. Design effective escalation and ambiguity resolution patterns
3. Implement error propagation strategies across multi-agent systems
4. Manage context effectively in large codebase exploration
5. Design human review workflows and confidence calibration
6. Preserve information provenance and handle uncertainty in multi-source synthesis

The order to learn them in