# Study Method

How to allocate hours by blueprint weight, combine study with hands-on building, and use practice exams to decide when to book.

## Principle 1 – Study weighted, not flat

The most common self-sabotage is spending equal time on a 2.6% domain and a 33.1% domain. Convert weights into hours:

```text
hours_for_domain = total_study_hours × domain_weight
```

Example for CCDV-F with a 30-hour budget:

| Domain | Weight | Hours |
| --- | --- | --- |
| Applications and Integration | 33.1% | ~10 |
| Model Selection and Optimization | 16.8% | ~5 |
| Agents and Workflows | 14.7% | ~4.5 |
| Prompt and Context Engineering | 11.0% | ~3.5 |
| Tools and MCPs | 10.6% | ~3 |
| Security and Safety | 8.1% | ~2.5 |
| Claude Code | 3.1% | ~1 |
| Eval, Testing, and Debugging | 2.6% | ~1 |

Then **redistribute** toward your measured weak domains after a first practice exam.

## Principle 2 – Build while you study

The official guides say it directly: combine study with hands-on work. One weekend project that touches five domains beats ten more hours of reading. Suggested builds per exam:

| Exam | Minimum viable build |
| --- | --- |
| CCAO-F | Configure a Claude Project with custom instructions and 3–5 knowledge documents; run one real workflow; evaluate the outputs for accuracy, bias and audience fit; write a one-page adoption memo |
| CCDV-F | An app that calls the Messages API (streaming + error handling), uses one custom tool and one MCP server, applies prompt caching, and includes a 20-case eval |
| CCAR-F | A coordinator/subagent research system with structured JSON output, a validation-retry loop, hooks for a hard business rule, and a CLAUDE.md hierarchy for a repo |
| CCAR-P | Everything in CCAR-F plus a RAG pipeline with chunking choices you can defend, an eval harness with A/B comparison, observability dashboards, and an architecture decision record (ADR) set |

## Principle 3 – Learn the trade-off, not the fact

Exam items give a scenario and four defensible options. The correct one reflects the **right trade-off given the constraints in the stem**. For every concept, be able to answer:

1. When is this the right choice?
2. What signal in a scenario points to it? ("cost matters, latency does not" → Batch API)
3. What is the tempting wrong alternative, and why is it wrong here?

The domain pages on this site are written around this triad.

## Principle 4 – Use practice exams as a booking gate

Fees are $99–$175 per attempt and retake waits are 14–90 days. A failed attempt costs money **and** months. Decision rule:

- Score **≥ 80%** on a fresh, timed practice exam, with **no domain below 65%** → book.
- Otherwise → redistribute hours to the weak domains, rebuild, re-test.

## A six-week template (5–8 h/week)

| Week | Focus |
| --- | --- |
| 1–2 | Read your course in weight order; do the free <a href="https://anthropic.skilljar.com/" target="_blank" rel="noopener noreferrer">Anthropic Academy</a> courses mapped to your exam |
| 3–4 | Build the minimum viable project; revisit domain pages as questions arise |
| 5 | Timed practice exam; drill the weak domains with the domain-level practice questions |
| 6 | Close gaps, re-test, review exam-day policy, book with the 24/48-hour change window in mind |

## Free Academy courses mapped to each exam

| Exam | Highest-value free courses (<a href="https://anthropic.skilljar.com/" target="_blank" rel="noopener noreferrer">Anthropic Academy on Skilljar</a>) |
| --- | --- |
| CCAO-F | Claude 101 · AI Capabilities and Limitations · AI Fluency track for your role |
| CCDV-F | Building with the Claude API (the backbone) · Introduction to Model Context Protocol · Claude Code in Action · Introduction to Agent Skills · Introduction to Subagents |
| CCAR-F | Building with the Claude API · MCP: Advanced Topics · Claude Code 101 · official docs on agents, context management and tool design |
| CCAR-P | Everything in the CCAR-F row plus Claude on Amazon Bedrock / Google Vertex AI if you deploy through a cloud platform |

## Reading the official documentation efficiently

Documentation was split in **July 2026**:

- Claude API and platform: `platform.claude.com/docs/en/*`
- Claude Code: `code.claude.com/docs/en/*`
- Agent SDK: `code.claude.com/docs/en/agent-sdk/*` (renamed from "Claude Code SDK" to **Claude Agent SDK**)
- MCP specification: `modelcontextprotocol.io`

Old `docs.anthropic.com` URLs still redirect. When a study source and the live docs disagree, the live docs win.