Claude Certified Architect - Professional (CCAR-P) Flashcards¶
43 cards from this guide's fact sheet and notes. Click a question to reveal the answer.
To study with spaced repetition, download flashcards.csv and import it into Anki or Quizlet (comma separated, first row is the header). Back to the study guide.
Exam logistics¶
How long is the Claude Certified Architect - Professional (CCAR-P) exam?
120 minutes
How many questions are on the Claude Certified Architect - Professional (CCAR-P) exam?
63 multiple-choice and multiple-response
What is the passing score for Claude Certified Architect - Professional (CCAR-P)?
720 / 1000
What does the Claude Certified Architect - Professional (CCAR-P) exam cost?
$175 USD
How long is Claude Certified Architect - Professional (CCAR-P) valid for?
12 months
How is the Claude Certified Architect - Professional (CCAR-P) exam delivered?
Pearson VUE - online proctored or test center
Governance safety and risk management¶
Training use (governance safety and risk management)
whether your data improves future models (off by default for API traffic)
Retention (governance safety and risk management)
whether your data is stored at all after the request is served
Violation rate (governance safety and risk management)
harmful content that got through
False refusal rate (governance safety and risk management)
legitimate requests that were blocked
Jailbreak (governance safety and risk management)
the user tries to talk the model out of its constraints
Direct prompt injection (governance safety and risk management)
the user embeds instructions that override the system prompt
Indirect prompt injection (governance safety and risk management)
untrusted content the agent retrieves (a web page, a document, a tool result) carries instructions aimed at the model
First-party API (governance safety and risk management)
your application must emit the audit trail; Anthropic provides request IDs to correlate
Bedrock (governance safety and risk management)
CloudTrail records every InvokeModel call natively; enable model invocation logging for payloads if policy requires
Vertex (governance safety and risk management)
Cloud Audit Logs record API activity; Cloud Logging for payloads if enabled
Autonomous (governance safety and risk management)
low-impact, reversible actions (draft a reply, summarize a document)
Approve-before-execute (governance safety and risk management)
the agent proposes, a human confirms (send the email, issue the refund)
Human-performs (governance safety and risk management)
the agent only recommends; a human executes (medical decisions, legal filings)
Stakeholder communication and lifecycle management¶
Value (stakeholder communication and lifecycle management)
what outcome improves, by how much
Cost (stakeholder communication and lifecycle management)
what it costs to run, per outcome
Risk (stakeholder communication and lifecycle management)
what can go wrong, likelihood, blast radius, mitigation
Time (stakeholder communication and lifecycle management)
when it ships, what phases, what gates
Direct savings (stakeholder communication and lifecycle management)
handle-time reduction, deflection rate, throughput per head
Revenue effects (stakeholder communication and lifecycle management)
faster response, longer coverage hours, higher conversion
Cost of the system (stakeholder communication and lifecycle management)
inference, engineering, evaluation, ongoing operations (not just the API bill)
Sensitivity (stakeholder communication and lifecycle management)
which assumption breaks the case first; usually adoption rate or escalation rate
Quality (stakeholder communication and lifecycle management)
eval score at or above baseline; task success rate at or above the phase target
Safety (stakeholder communication and lifecycle management)
red-team suite pass; violation and false-refusal rates within bounds
Operations (stakeholder communication and lifecycle management)
alerting live, runbook written, on-call named, rollback tested
Architecture decision records (ADRs) (stakeholder communication and lifecycle management)
what was chosen, what was rejected, why; the future team inherits your reasoning, not just your code
Runbook (stakeholder communication and lifecycle management)
alert-by-alert response procedures: quality drift, cost spike, provider outage, guardrail trip, model deprecation notice
System card / model use documentation (stakeholder communication and lifecycle management)
purpose, data flows, limitations, evaluation evidence; doubles as EU AI Act and procurement documentation
Escalation matrix (stakeholder communication and lifecycle management)
who is paged for what, and who can authorize a rollback or kill switch
Change log (stakeholder communication and lifecycle management)
every prompt, model, and tool change with its eval delta
End users (stakeholder communication and lifecycle management)
what the assistant can and cannot do, how to phrase requests, how to escalate; short and example-driven
Support and operations teams (stakeholder communication and lifecycle management)
how to read traces, common failure signatures, when to page engineering
Reviewers (human-in-the-loop) (stakeholder communication and lifecycle management)
what they are approving, what to look for, how their decisions are logged
Leadership (stakeholder communication and lifecycle management)
the dashboard: quality trend, cost per outcome, adoption, incident count
Weekly (early life) (stakeholder communication and lifecycle management)
quality metrics, escalation rate, cost per outcome, top failure categories; feed real failures back into the eval set
Monthly (stakeholder communication and lifecycle management)
stakeholder review against the original ROI case; adoption and satisfaction trends
Quarterly (stakeholder communication and lifecycle management)
model version review (anything deprecating?), risk re-tiering, guardrail effectiveness (both violation and false-refusal rates), and a decision: invest, maintain, or retire
Incident-driven (stakeholder communication and lifecycle management)
every user-visible failure gets a lightweight postmortem; recurring patterns become eval cases and runbook entries