Mastering the DotCreds Practice Bank for the Claude Certified Architect Foundations Exam

After each practice set, label the miss. Was the problem model selection, request structure, instruction hierarchy, tool definition, tool invocation flow, MCP integration, context management, validation, or security boundary? The label tells you what to review next and prevents broad retakes from hiding the real weakness.

If a model-selection question is missed, revisit the workload requirements. Look for task complexity, desired quality, latency tolerance, cost sensitivity, and context needs. A wrong answer often chooses a model because it sounds more capable without matching the actual design constraint.

A common mistake is treating tool use and MCP as interchangeable. Tool use is about Claude requesting structured tool calls based on declared tool definitions in an application workflow. MCP is about connecting Claude Code to external tools and resources through MCP servers. Practice review should keep those integration patterns separate.

When a tool-related answer is wrong, inspect the schema and boundary. Is the tool name clear? Are required inputs defined? Does the description distinguish it from other tools? Does the tool expose too much power? Tool questions often test precision and least-privilege thinking rather than the idea of tools in general.

Instruction mistakes often come from unclear hierarchy, missing constraints, irrelevant context, or assumptions the model cannot verify. Review whether the request gives Claude the right task, the right information, and the right boundaries. Context management is an architecture choice, not just a prompt-length problem.

A strong Claude workflow plans for invalid tool arguments, failed external calls, uncertain outputs, and structured responses that do not pass validation. If a practice item chooses the wrong recovery path, review when to retry, ask for clarification, block an action, fall back, or require human review.

When two choices sound plausible, verify the underlying claim in Anthropic documentation. Use the model-selection page for model tradeoffs, the tool-definition page for tool schema behavior, and the MCP page for Claude Code integration. Practice is most useful when it sends the learner back to the correct source.