CCAR-F 4-Week Blueprint Study Plan
Independent study resource. Not affiliated with or endorsed by Anthropic.
A structured, self-paced 4-week preparation plan mapped directly to the 30 official Anthropic blueprint task statements. Requires 45 to 60 minutes per day with progressive diagnostics and full-length timed mock exams.
Curriculum Milestone Roadmap
Systematic progression from baseline calibration to full exam readiness.
W1 27% WEIGHT
Agentic Orchestration
Master autonomous loops, all 6 stop reasons, state machines, and safety circuit breakers.
W2 38% WEIGHT
MCP & Claude Code
4-category MCP error taxonomy, tool annotations, CLAUDE.md hierarchy, and plan mode.
W3 35% WEIGHT
Prompts & Context
JSON schema validation-retry loops, prompt caching cost optimization, and RAG grounding.
W4 SIMULATION
Timed Mocks & Gaps
Two full 120-minute 60-question mock exams, targeted remediation drills, and test day prep.
Week 1 • Domain 1 (27% Weight)
Study Autonomous Loops GuideAgentic Architecture & Orchestration
Tasks 1.1 through 1.6 • Focus: Autonomous Loops, Continuation Turns, State TransitionsDaily Study Schedule
- Day 1: Baseline Diagnostic (30m)Take the free 15-question diagnostic to identify prior knowledge and establish your pre-study domain score.
- Day 2: Tasks 1.1 & 1.2 — Multi-Turn Agent Loop MechanicsStudy all 6 API stop reasons (tool_use, end_turn, max_tokens, stop_sequence, pause_turn, refusal). Learn continuation turn prompting.
- Day 3: Task 1.3 — Stopping Mechanisms & Runaway Loop BoundsWhy MAX_TURNS must only be a safety net, not a primary completion trigger. Implement explicit terminal conditions.
- Day 4: Tasks 1.4 & 1.5 — Routing, State & Agent SDK HooksState machine transitions with deterministic condition guards. Intercepting model tool calls via Agent SDK hooks.
- Day 5: Task 1.6 — Multi-Agent Subagent CoordinationOrchestrator-subagent hierarchy, isolated context passing, and shared memory synchronization.
Week 1 Required Readings & Code Labs
- Claude Platform Docs: Agentic Orchestration Patterns
- Lab: Build an autonomous agent loop in TypeScript that handles all 6 stop reasons without crashing on max_tokens.
- Drill: Practice 15 Domain 1 scenario questions in Qurtesy Practice Mode.
Week 2 • Domains 2 & 3 (38% Combined Weight)
Study MCP Error Isolation GuideTool Design, MCP Integration & Claude Code
Tasks 2.1–2.6 (18%) & Tasks 3.1–3.6 (20%) • Focus: MCP Architecture, CLAUDE.md HierarchyDaily Study Schedule
- Day 6: Tasks 2.1 & 2.2 — Tool Definitions & 4-Category Error TaxonomyDefine tools with JSON Schema. Master transient, validation, permission, and business errors with isRetryable flags.
- Day 7: Tasks 2.3–2.6 — Model Context Protocol (MCP) Integrationstdio vs SSE transports. Resource subscriptions, tool annotations, empty result vs access error distinction.
- Day 8: Tasks 3.1–3.3 — Claude Code Configuration & CLAUDE.md HierarchyRoot vs subfolder rules, system prompt layering, gitignored secrets, and permission boundaries.
- Day 9: Tasks 3.4–3.6 — Plan Mode vs Direct & Test ExecutionWhen to use plan mode for multi-file refactors. Automated test execution and git workflow automation.
- Day 10: Mid-Point AssessmentTake 20 practice drill questions combining Domain 2 MCP tools with Domain 3 Claude Code rules.
Week 2 Required Readings & Code Labs
- MCP Specification: Tools, Prompts & Resources
- Lab: Configure an MCP server in TypeScript that surfaces an SQLite database tool with strict 4-category error handling.
- Lab: Write a production CLAUDE.md with architecture guidelines and project command workflows.
Week 3 • Domains 4 & 5 (35% Combined Weight)
Study Structured Output GuidePrompt Engineering, Structured Output & Context Management
Tasks 4.1–4.6 (20%) & Tasks 5.1–5.6 (15%) • Focus: Schema Enforcement, Caching, EscalationDaily Study Schedule
- Day 11: Tasks 4.1–4.3 — Input XML Delimiters vs Output tool_use SchemasPrompt injection defense via XML tags. Forced tool_choice vs auto. Defensive schema typing (nullable fields, enum other).
- Day 12: Tasks 4.4–4.6 — Host Validate-and-Retry Loop ArchitectureDistinguishing syntax errors from semantic domain errors. Feeding validation diagnostics back to Claude for structured repair.
- Day 13: Tasks 5.1 & 5.2 — Context Window Preservation & Prompt CachingStrategic cache breakpoints, 90% cost reduction, context compaction algorithms, and fact preservation across long conversations.
- Day 14: Tasks 5.3–5.6 — Reliability & Human Escalation CalibrationCalibrating confidence thresholds for human-in-the-loop escalation. Graceful degradation when rate limits or API outages occur.
- Day 15: Mock Exam Attempt #1 (120m)Take your first full 60-question timed mock exam under authentic exam conditions. Note your scaled score.
Week 3 Required Readings & Code Labs
- Claude Platform Docs: Prompt Caching Implementation
- Lab: Build a Zod validate-and-retry wrapper that passes JSON Schema validation errors back to Claude up to 3 times.
- Milestone: Complete Qurtesy Timed Mock Exam #1. Target scaled score: >= 720.
Week 4 • Full Simulation & Remediation
Launch 120-Minute MockMock Exams, Gap Analysis & Test Day Readiness
Target: Scaled Score >= 750 / 1000 • Timing: Under 100 Minutes (20m Buffer)Daily Study Schedule
- Day 16: Mock Exam #1 RemediationReview every question missed or flagged in Mock #1. Re-read the corresponding lesson guides in our 30-task curriculum.
- Day 17: Scenario Cross-Mapping ReviewReview all 6 official exam scenarios. Verify you understand which architectural constraints apply to each domain.
- Day 18: Mock Exam Attempt #2 (120m)Take your second full 60-question timed mock exam. Confirm you complete all items within 100 minutes to leave time for review.
- Day 19: Pearson VUE Test Environment CheckRun OnVUE system check utility on the exact laptop and network you will use on exam day. Test webcam, mic, and bandwidth.
- Day 20: Pre-Exam Taper & Blueprint Confidence AuditLight review of anti-patterns and key formulas (scaled score model, stop reasons). Ensure government photo ID is ready.
Test Day Protocol Checklist
- Valid, unexpired government photo ID matching registration name
- Clean desk policy: no second monitor, no paper, no smartwatch
- Check-in begins 30 minutes prior to scheduled appointment
- Closed-book exam: no Claude, no notes, no external aids allowed