Claude Certified Architect Foundations Playbook
Pass the CCA-F Exam in 30 Days with 60 Scenario Drills and 8 Build Projects
The CCA-F is a scenario-based architect exam, not trivia. This playbook is a 30-day study system built around the 5 judgment patterns the exam keeps testing: scale-failure, wrong-tool routing, silent failure, cross-machine portability, and trust-boundary edge cases. By Day 9 you have a working multi-agent coordinator. By Day 15 a portable MCP setup that survives a teammate clone. By Day 25 silent-failure detection you can defend in an interview. By Day 30 you sit the 60-question scenario mock.
The CCA-F syllabus dropped, then the shelf filled up with thin online guides and 460-page bricks that explain AI basics instead of preparing you for the exam you actually have to pass. This is dense, scenario-driven prep with a 30-day execution path and concrete artifacts you can explain after the exam.
What You'll Build
The 5-question diagnostic: know your weakest domain before you start.
Partner gate, format, scoring, the 3-4 question pass/fail margin.
Recognize scenario archetypes on sight. The mental model the rest of the book builds on.
Hub-and-Spoke coordinators with subagents. Domain 1 is 27% of the exam.
Batch API routing, JSON schemas, 3-attempt fallback loops with exponential backoff.
Portable MCP that survives a teammate clone: config placement and prompt-injection defense.
The 3-Level CLAUDE.md hierarchy, slash commands, hooks, and scope traps.
Schema validation, Pydantic post-validation, 20-trial test set with 100% schema-valid outputs.
Silent-failure detection with assertions, golden-set regression, and LLM-as-judge review.
Operator/user boundaries, least-privilege MCP access, audit logging, prompt-injection defense.
Cert positioning, salary anchors, and pairing the badge with visible prep projects.
The 60-question scenario mock weighted across all 5 domains. Score yourself domain by domain.
Free Articles from this Book
Claude Agent Silent Failure Detection: 3 Mechanisms
Claude agent silent failure detection for output that passes schema but is wrong: assertion validators, a pinned golden set, and an LLM-as-judge stack.
from: Claude Certified Architect Foundations Playbook
Claude Batch API Cost Savings: It Saves 50%, Not 30%
Claude Batch API cost savings are 50% off input and output, not 30%. The synchronous-vs-cache-vs-batch decision tree, verified limits, and the exam trap.
from: Claude Certified Architect Foundations Playbook
The 5 Judgment Patterns the CCA Exam Actually Tests
The Claude Certified Architect exam grades pattern recognition, not trivia. The 5 scenario archetypes, their stem cues, and the one-line fix for each.
from: Claude Certified Architect Foundations Playbook
Claude Chatbot Hallucinating Fix: The Hub-and-Spoke
Claude chatbot hallucinating fix: a Coordinator, Researcher, and Synthesizer with scoped tools and a logged refusal path stop confident invention at the source.
from: Claude Certified Architect Foundations Playbook
MCP Works on My Machine, Breaks on Teammate's: 5 Fixes
MCP works on my machine breaks on teammate clones for 5 reasons. The portability gotchas, the .mcp.json project-scope rule, and the tool-naming convention fix.
from: Claude Certified Architect Foundations Playbook