← Learn Claude

CCA-F Exam Scenarios — Architecture Decision Framework

Structure: 6 production scenarios → exam randomly selects 460 MCQs (~15 per scenario) spanning all 5 domains.

Prepare for all six. You cannot predict which four appear.

YouTube course map: ../CCA-STUDY-MAP (Episodes 01–10 ↔ scenarios ↔ Youtube-main/claude-certified-architect/ code).


Decision framework (apply to every scenario question)

When two answers look plausible, score each against:

Criterion Prefer
Context Minimum necessary — not maximum
Errors Handled at the owning layer (worker/tool/API), not dumped on supervisor
Prompts More constrained and specific
MCP/tools Correct transport + auth for deployment model
Reliability Token-efficient with preserved accuracy

Scenario 1 — Customer Support Resolution Agent

Tests: Multi-agent support pipeline with escalation and tool use.

Architecture anchors:
- Tier-1 agent handles FAQ/RAG; escalate to specialist with summary only, not full transcript
- Tools: order lookup, account status, ticket create — clear schemas, graceful API-down handling
- Failure: retry transient errors; escalate to human when confidence low

Domain overlap: D1 (orchestration), D4 (tools), D3 (structured escalation prompts), D5 (context trim)

Your repo anchor: Task-Manager-App, CLAUDE-MD agent patterns

YouTube: Ep 01, 06–07

Prepared: [ ] Can sketch supervisor + tier-1 + escalation flow on paper


Scenario 2 — Multi-Agent Research Pipeline

Tests: Orchestrated research agents with retrieval and synthesis.

Architecture anchors:
- Decompose: search → fetch → summarise → synthesise (sequential where dependent, parallel for independent sources)
- Workers return compressed findings, not raw pages, to supervisor
- Citation/provenance tracked for final output

Domain overlap: D1 (parallel vs sequential), D5 (context), D3 (structured synthesis)

Your repo anchor: CLAUDE-MD/AgentTeamsDemo

YouTube: Ep 02–04 (capstone_project.py)

Prepared: [ ] Can explain parallel search + sequential synthesis


Scenario 3 — Claude Code Dev Workflow

Tests: CLAUDE.md, hooks, skills, automated coding workflows.

Architecture anchors:
- Project-scoped CLAUDE.md for repo conventions; global for personal prefs only
- PreToolUse blocks destructive ops (exit 2); PostToolUse validates/format (exit 0/1)
- Skills for repeatable workflows; MCP for external tool access

Domain overlap: D2 (primary), D4 (MCP in Claude Code)

Your repo anchor: p4-claude-code-workflow

YouTube: Ep 05, 08–10

Prepared: [ ] Verified P4 build steps complete


Scenario 4 — Enterprise RAG Assistant

Tests: Production RAG with context management and structured output.

Architecture anchors:
- Ingestion pipeline → vector DB → retrieval with top-k + rerank → grounded answer with citations
- Structured output schema for answers; evaluation on citation accuracy
- Long docs: chunk strategy + context window management (not dump full corpus)

Domain overlap: D3, D5 (primary), D1 (optional agent wrapper)

Your repo anchor: DailyFlow backend patterns

Prepared: [ ] Can defend chunk size vs retrieval quality trade-off


Scenario 5 — MCP Tool Integration System

Tests: MCP server design, authentication, enterprise integration.

Architecture anchors:
- 3+ tools with precise inputSchema descriptions
- Auth: API key for service accounts; OAuth for user-delegated access
- Transport: stdio for local Claude Code; HTTP/SSE for shared remote server
- Partial failures: structured error responses, not silent empty results

Domain overlap: D4 (primary), D1 (agent calling tools)

Your repo anchor: p5-custom-mcp-server

YouTube: Ep 08 (compare with P5 implementation)

Prepared: [ ] Ran P5 server and connected from Claude Code


Scenario 6 — Agentic Automation Pipeline

Tests: End-to-end autonomous workflow with multi-step orchestration.

Architecture anchors:
- Ordered subtasks with checkpoints; supervisor validates each step before next
- Parallel only when steps have no shared mutable state
- Abort vs retry vs fallback explicitly defined per step type

Domain overlap: D1 (primary), D5 (reliability), D3 (step contracts)

Your repo anchor: Highest gap — use P3 build steps below if not yet built

Prepared: [ ] Can diagram failure handling when step N fails


Scenario simulation drill

Pick 4 scenarios at random (dice, or shuffle). For each, answer in 5 minutes:

  1. What agents/components exist?
  2. What is sequential vs parallel?
  3. Where do errors get handled?
  4. What context crosses each boundary?
Session Scenarios selected (pick 4) Time Done
1 [ ]
2 [ ]
3 [ ]

Optional P3 build (Multi-Step Agent Workflow)

If Scenario 6 is your weakest area, build:

  1. Design supervisor + specialist workers
  2. Implement sequential task chain
  3. Add parallel execution for independent tasks
  4. Define failure handling when worker B fails
  5. Test end-to-end goal completion

Maps to D1 + D5 — see study site Projects → P3.