Recognizing Superior Architecture — Pattern Library for Exam Scenarios

⚡ Exam Tested 10 min +45 XP

💡

THE ANALOGY

A building inspector who's seen thousands of buildings. They don't check the code for every detail — they've internalized patterns. They can tell from a glance that a building is structurally sound or has hidden problems. The CCA exam tests whether you've internalized the architectural patterns well enough to recognize superior design on sight.

⚠️ EXAM TRAP — The Wrong Answer People Choose

Spending time on two-right-answers questions analyzing pros and cons of each option. The exam is timed — 2 minutes per question. Recognizing the pattern and applying the principle should take 30 seconds.

KEY POINTS

1 Pattern: Prompt vs Hook → Hook wins for compliance, hooks for financial/legal/safety.

2 Pattern: All tools vs Scoped tools → Scoped wins for reliability and blast radius.

3 Pattern: Silent failure vs Structured error → Structured error wins always.

4 Pattern: Sequential vs Parallel for independent tasks → Parallel wins for performance.

5 Pattern: Self-review vs Independent instance → Independent wins for accuracy.

The Pattern Library

Pattern A: Compliance Enforcement

Scenario: "Agents must always verify X before doing Y"
Option 1: System prompt: "Always verify X before Y"
Option 2: PreToolUse hook that blocks Y until X is verified

Principle: Determinism > Probabilism for compliance
Answer: Option 2 (hook)
30-second reasoning: "must always" + compliance = deterministic enforcement required

Pattern B: Tool Access

Scenario: "Agent needs to do [specific task]"
Option 1: Give agent all 15 available tools
Option 2: Give agent the 3 tools relevant to this task

Principle: Minimum footprint
Answer: Option 2 (scoped)
30-second reasoning: minimum necessary = scoped access. Reliability + blast radius.

Pattern C: Error Handling

Scenario: "When a tool fails, the agent should..."
Option 1: Return default values and continue
Option 2: Return structured error with category and isRetryable

Principle: Fail loudly
Answer: Option 2 (structured error)
30-second reasoning: default values = silent failure = impossible to diagnose

Pattern D: Parallel vs Sequential

Scenario: "Agent needs to run tasks A, B, C which don't depend on each other"
Option 1: Run A, then B, then C sequentially
Option 2: Run A, B, C in parallel

Principle: Performance + correct use of agentic capability
Answer: Option 2 (parallel)
30-second reasoning: no dependencies = parallelize. Sequential is artificial serialization.

Pattern E: Review Architecture

Scenario: "Code was generated by Claude. Review it for bugs."
Option 1: Ask Claude to review its own code in the same session
Option 2: Use a fresh Claude instance with no knowledge of the generation

Principle: Independence eliminates bias
Answer: Option 2 (independent instance)
30-second reasoning: same session = generator's reasoning biases the review

Pattern F: Session Resume vs Fresh Start

Scenario: "Files have been modified since the last Claude Code session"
Option 1: Resume the previous session (--resume)
Option 2: Start fresh with a briefing summarizing prior findings

Principle: Stale context prevention
Answer: Option 2 (fresh + briefing) if significant files changed
30-second reasoning: stale tool results cause Claude to reason about an outdated state

Quick Reference Card

Signal word → Principle → Answer pattern

"must always" / "never"     → Determinism      → Hook/gate, not prompt
"all available tools"       → Min footprint     → Scope to necessary
"continue anyway"           → Fail loud         → Structured error
"independent tasks"         → Parallelize       → Parallel execution
"same session review"       → Independence      → Fresh instance
"files changed since"       → Stale context     → Fresh session
"compliance" / "financial"  → Programmatic      → Code, not prompt
"might" / "sometimes"       → Explicit          → Make behavior visible

Key Takeaways

Pattern-match, don’t analyze — exam conditions require speed
Signal words change the answer — “must always” vs “should usually” are different questions
Complexity ≠ correctness — simpler correct architecture beats complex wrong one
Six patterns cover most questions — memorize and apply
30 seconds per two-right-answers question — recognize, apply principle, move on