2026 research and the added practice set
Reviewed 21 October 2026. This update adds 100 original scenarios, P081–P180, to the existing 80 learning questions. The two reserved mocks remain unchanged at 60 questions each. Total: 300 questions.
What the research established
Google's certification page links the current exam guide, official sample questions and recommended training. The guide currently allocates approximately 20%, 25%, 18%, 25% and 12% to the five domains. The added set follows those proportions exactly across 100 questions and touches all 19 numbered objectives. Coverage does not establish the difficulty or completeness of an actual exam. Certification page, current linked guide.
Searches also surfaced commercial and user-published listings labeled “2026 questions” or “dumps.” Their titles and sales claims did not establish their accuracy, authorization, freshness or likelihood of appearing on an exam. I did not purchase those banks or copy their question sets. Google explicitly includes unauthorized brain-dump materials among prohibited preparation materials. Google's misconduct policy.
The new questions are independently written teaching scenarios based on published concepts and documentation. They are not recalled or leaked exam questions, and this pack is not endorsed by Google. A 2026 label is a review date, not proof that a question will appear. Your six-month timeline may put your exam in 2027; recheck the guide before booking.
What was added
| Domain | New questions | Representative practice |
|---|---|---|
| Organization and infrastructure | 20 | Inherited IAM grants, overlapping networks, Terraform state, Packer/kpt, build dependencies, maintenance |
| CI/CD and testing | 25 | Required gates, artifact trust, ML evaluation, canary calculations, schema compatibility, secret rotation |
| SRE practices | 18 | Budget arithmetic, SLI denominators, HPA calculations, capacity, failover, incident decisions |
| Observability and troubleshooting | 25 | Log-based metrics, audit logging, routing, PromQL, traces, image pulls, startup and memory failures |
| Performance and cost | 12 | Profiles, database contention, cold starts, commitment arithmetic, Spot tradeoffs, billing credits |
Every new question includes an answer explanation and primary documentation links. Numerical pricing examples explicitly use hypothetical rates. They teach calculations without claiming to quote current product prices.
Several implementation details were checked against current primary pages: Cloud Build step order, secret use in builds, Cloud Deploy verification placement, log-based metric behavior, GKE image-pull identities, Cloud Run listener requirements, OpenTelemetry collector routing and Terraform sensitive data. Other references appear with the relevant question.
How to use the new questions
- Open the quiz. Select a domain you have studied and a session size of 10 or 20.
- Choose Added scenario set (100) to use only P081–P180. Start practice draws from all 180 learning questions. Both buttons honor the domain and size controls.
- Answer before reading the explanation. Mark answers you guessed, including correct guesses.
- For every miss, state the controlling requirement: identity, network reachability, artifact identity, measurement denominator, compatibility, capacity or cost.
- Explain why each alternative fails under the stated conditions. Change one condition and ask whether the answer would change.
- Reproduce applicable concepts in the labs, then revisit errors after 2, 7 and 21 days.
If a question is unfamiliar, use it as a lesson and follow its source. There is no benefit in rushing through all 180 before you understand the foundations. Questions alone do not provide the operational experience that the labs are designed to begin building.
What would justify confidence
Finishing this bank or recognizing its answers is not enough evidence to spend your only voucher. Look for first-attempt results on previously unseen questions, clear explanations without notes, and successful lab demonstrations. Use the readiness checklist, including its personal mock targets and hands-on criteria. These targets are study rules, not an official passing threshold or a statistical pass prediction.
Keep Mock A and Mock B unseen until the planned checkpoints. Repeated scores become revision evidence. Preserve exported results and your error log so weak domains remain visible. The update keeps existing question IDs and answer ordering intact for saved progress in the same browser and file location; moving the pack or clearing browser storage can still lose that browser's history.