PUBLIC BETA OFFER

Save 50% off on an Eligible Learning PlanValid for 1-month plan, 3-month plan, 12-month plan · Code LPXBETA50 · Ends Nov 4, 2026

Sign inChoose a certification

Claude Certified Developer · Anthropic

Turn mistakes into focused review

Identify the key concept to reinforce, validate it with a similar question, and return with clearer evidence.

Personal review evidence within your learning scope
Open your free workspace

How Smart Review works

Turn practice into a focused review loop

Smart Review turns a completed test into a short, repeatable learning loop. The method is public; your evidence remains private.

  1. 1

    Identify the mistake

    Separate an incorrect answer from uncertainty, guessing, or a recurring gap.

  2. 2

    Review one concept

    Return to the most relevant learning topic instead of repeating the whole subject.

  3. 3

    Validate with a similar question

    Use related practice to confirm whether the reasoning is now reliable.

  4. 4

    Recheck readiness

    Use the next result to decide whether to continue, reinforce, or change focus.

Review example

From a mistake to a focused next step

See how a completed practice result becomes one clear review action. This is a public example, not a personal record.

1

Practice result

An incorrect or uncertain answer points to a concept that needs another look.

2

Why review is recommended

Smart Review connects the answer, confidence, and related learning topic.

3

Focused next action

Review the concept, then validate the reasoning with a similar question.

Build your review evidence

Start with free practice, then use your results to choose a focused next step.

Subject review guide · published content

Capture five review signals

Combine correctness with confidence, speed, recency, and repeated mastery.

Fixed subject guidance from the published Readiness package—not your personal result.
01

Incorrect answer

The learner selected a mechanism that does not satisfy the scenario.

  • Record the binding requirement and the mistaken rule.
  • Separate knowledge gaps from reading errors.
02

Low confidence

The answer was correct but the explanation or confidence was weak.

  • Treat correct guesses as unresolved.
  • Practice one contrasting scenario.
03

Slow response

The learner took too long to identify the relevant domain or discriminator.

  • Create a short decision cue.
  • Practice retrieval under a two-minute limit.
04

Stale topic

A previously strong topic has not been retrieved recently.

  • Schedule spaced mixed review.
  • Use a fresh scenario rather than rereading notes.
05

Low mastery

Repeated evidence remains below the internal domain threshold.

  • Return to official objectives and a hands-on artifact.
  • Delay graduation until fresh performance is stable.

Weight by risk and exam importance

Prioritize unresolved errors in high-weight domains while maintaining all-domain coverage.

1

Priority 1

  • Errors or low-confidence evidence occur in Agents and Workflows.
  • Apply the official 14.7% weight when ranking otherwise similar gaps.

Recommended action

  • Review the missed objective and the exact discriminator.
  • Implement or trace a minimal example, then answer a new scenario.
2

Priority 2

  • Errors or low-confidence evidence occur in Applications and Integration.
  • Apply the official 33.1% weight when ranking otherwise similar gaps.

Recommended action

  • Review the missed objective and the exact discriminator.
  • Implement or trace a minimal example, then answer a new scenario.
3

Priority 3

  • Errors or low-confidence evidence occur in Claude Code.
  • Apply the official 3.1% weight when ranking otherwise similar gaps.

Recommended action

  • Review the missed objective and the exact discriminator.
  • Implement or trace a minimal example, then answer a new scenario.
4

Priority 4

  • Errors or low-confidence evidence occur in Eval, Testing, and Debugging.
  • Apply the official 2.6% weight when ranking otherwise similar gaps.

Recommended action

  • Review the missed objective and the exact discriminator.
  • Implement or trace a minimal example, then answer a new scenario.
5

Priority 5

  • Errors or low-confidence evidence occur in Model Selection and Optimization.
  • Apply the official 16.8% weight when ranking otherwise similar gaps.

Recommended action

  • Review the missed objective and the exact discriminator.
  • Implement or trace a minimal example, then answer a new scenario.
6

Priority 6

  • Errors or low-confidence evidence occur in Prompt and Context Engineering.
  • Apply the official 11% weight when ranking otherwise similar gaps.

Recommended action

  • Review the missed objective and the exact discriminator.
  • Implement or trace a minimal example, then answer a new scenario.
7

Priority 7

  • Errors or low-confidence evidence occur in Security and Safety.
  • Apply the official 8.1% weight when ranking otherwise similar gaps.

Recommended action

  • Review the missed objective and the exact discriminator.
  • Implement or trace a minimal example, then answer a new scenario.
8

Priority 8

  • Errors or low-confidence evidence occur in Tools and MCPs.
  • Apply the official 10.6% weight when ranking otherwise similar gaps.

Recommended action

  • Review the missed objective and the exact discriminator.
  • Implement or trace a minimal example, then answer a new scenario.

Review one domain through decisions

Each unit links concepts, common mistakes, retrieval prompts, and observable success criteria.

01Agents and Workflows

Review points

  • Choose between fixed workflows and autonomous agents, and structure manager, supervisor, and subagent relationships for reliable task execution.
  • Build Claude agents with the Agent SDK, custom loops or harnesses, appropriate hosting, and deterministic hooks.
  • Apply tool-use loops, memory, context management, subagents, and suitable agentic frameworks to multi-step work.

Common mistakes

  • An agent is always an upgrade over a fixed workflow.
  • Adding subagents automatically improves quality without coordination cost.
  • Claude executes client tools without application code.
  • A prompt instruction provides the same enforcement as a hook.

Retrieval prompts

  • Explain the central decision rules for Agents and Workflows without notes.
  • Describe a scenario where the most tempting alternative would fail.
  • Name the observable evidence that would validate the selected approach.

Success criteria

  • At least 85% on fresh Agents and Workflows items.
  • Explanations identify requirement, mechanism, evidence, and rejected alternatives.
02Applications and Integration

Review points

  • Translate business needs into functional behavior and infrastructure constraints before choosing an implementation.
  • Apply planning, development, testing, deployment, operations, and maintenance discipline to Claude systems.
  • Use messages, tools, streaming, vision, thinking, caching, third-party providers, data access, and batch processing correctly.
  • Apply REST, JSON, async programming, version control, code review, testing, and refactoring practices to Claude integrations.
  • Design maintainable Claude applications across Code, Desktop, claude.ai, API, and SDK interfaces with clear content and session boundaries.
  • Version and scope CLAUDE.md, settings.json, model identifiers, prompts, and plugin dependencies for reproducible behavior.

Common mistakes

  • Model choice should precede requirements analysis.
  • A feature list is sufficient to define a production system.
  • Probabilistic software does not benefit from conventional SDLC practices.
  • A successful prototype is ready for production deployment.

Retrieval prompts

  • Explain the central decision rules for Applications and Integration without notes.
  • Describe a scenario where the most tempting alternative would fail.
  • Name the observable evidence that would validate the selected approach.

Success criteria

  • At least 85% on fresh Applications and Integration items.
  • Explanations identify requirement, mechanism, evidence, and rejected alternatives.
03Claude Code

Review points

  • Use Rules, Skills, Commands, Agents, Agent Memory, session modes, CLAUDE.md hierarchy, initialization, and settings.json correctly.

Common mistakes

  • Headless mode and auto mode are synonyms.
  • A more specific CLAUDE.md always replaces broader guidance.

Retrieval prompts

  • Explain the central decision rules for Claude Code without notes.
  • Describe a scenario where the most tempting alternative would fail.
  • Name the observable evidence that would validate the selected approach.

Success criteria

  • At least 85% on fresh Claude Code items.
  • Explanations identify requirement, mechanism, evidence, and rejected alternatives.
04Eval, Testing, and Debugging

Review points

  • Diagnose Claude applications by error type, trace evidence, stop reasons, and the boundary where the first deviation occurs.

Common mistakes

  • Every malformed response is a model failure.
  • Retrying all failures improves reliability.

Retrieval prompts

  • Explain the central decision rules for Eval, Testing, and Debugging without notes.
  • Describe a scenario where the most tempting alternative would fail.
  • Name the observable evidence that would validate the selected approach.

Success criteria

  • At least 85% on fresh Eval, Testing, and Debugging items.
  • Explanations identify requirement, mechanism, evidence, and rejected alternatives.
05Model Selection and Optimization

Review points

  • Understand tokens, context windows, sampling, non-determinism, next-token generation, thinking modes, and prompting modes.
  • Apply SDK, REST API, websocket, async, and integration fundamentals that support AI application development.
  • Choose among Claude model tiers and reasoning capabilities using quality, latency, cost, and compatibility evidence.
  • Budget, observe, model, and reduce token cost with appropriate caching and workload execution choices.

Common mistakes

  • The same prompt always produces identical output.
  • Extended thinking is required for every task.
  • Using an SDK removes the need to understand the underlying request contract.
  • Websockets are the default choice for every streamed workload.

Retrieval prompts

  • Explain the central decision rules for Model Selection and Optimization without notes.
  • Describe a scenario where the most tempting alternative would fail.
  • Name the observable evidence that would validate the selected approach.

Success criteria

  • At least 85% on fresh Model Selection and Optimization items.
  • Explanations identify requirement, mechanism, evidence, and rejected alternatives.
06Prompt and Context Engineering

Review points

  • Control what enters, remains in, and leaves context through pruning, compaction, isolation, and deliberate memory scope.
  • Write clear instructions, examples, placement, constraints, and sanitization rules, then refine from observed failures.
  • Produce, validate, parse, and consume Claude output defensively using schemas and semantic checks.

Common mistakes

  • More context always improves performance.
  • Compaction preserves every detail automatically.
  • Longer prompts are inherently more reliable.
  • A system prompt alone can enforce application authorization.

Retrieval prompts

  • Explain the central decision rules for Prompt and Context Engineering without notes.
  • Describe a scenario where the most tempting alternative would fail.
  • Name the observable evidence that would validate the selected approach.

Success criteria

  • At least 85% on fresh Prompt and Context Engineering items.
  • Explanations identify requirement, mechanism, evidence, and rejected alternatives.
07Security and Safety

Review points

  • Mitigate prompt injection, jailbreaks, leakage, and PII risks while preserving authentication, authorization, confidentiality, privacy, and integrity.
  • Layer content policy, screening, authorization, and secure-by-design controls at the points where harm can occur.
  • Use hooks as deterministic enforcement and audit points around Claude tool activity.
  • Manage identities, API keys, access approvals, environment separation, rotation, and monitoring across development and production.

Common mistakes

  • A stronger system prompt makes an agent immune to injection.
  • A trusted user implies all retrieved content is trusted.
  • One output filter covers input manipulation and unauthorized actions.
  • Monitoring after an action is equivalent to preventing it.

Retrieval prompts

  • Explain the central decision rules for Security and Safety without notes.
  • Describe a scenario where the most tempting alternative would fail.
  • Name the observable evidence that would validate the selected approach.

Success criteria

  • At least 85% on fresh Security and Safety items.
  • Explanations identify requirement, mechanism, evidence, and rejected alternatives.
08Tools and MCPs

Review points

  • Define focused tool sets, clear schemas, execution boundaries, approval patterns, and explicit error returns.
  • Author, deploy, and integrate MCP tools, resources, and prompts over suitable client-server communication patterns.
  • Choose among built-in tools, custom tools, Skills, and MCP servers by capability ownership, reuse, maintenance, and context cost.

Common mistakes

  • Claude directly executes every tool it selects.
  • Exposing more overlapping tools always improves capability.
  • MCP is useful only when one application consumes the capability.
  • Transport choice and configuration scope are the same decision.

Retrieval prompts

  • Explain the central decision rules for Tools and MCPs without notes.
  • Describe a scenario where the most tempting alternative would fail.
  • Name the observable evidence that would validate the selected approach.

Success criteria

  • At least 85% on fresh Tools and MCPs items.
  • Explanations identify requirement, mechanism, evidence, and rejected alternatives.
Check exam readiness