What changed from the original material
The original 220-question AI bank and cram sheet are archived unchanged in archive/original-2026-10-21/. They are not included in the current quiz. The active pack contains 300 newly written questions: 180 learning questions (including 100 added scenarios) and two disjoint 60-question mocks. The initial replacement pack contained 200 questions; the expansion preserves those questions and both mocks. This replaces unreliable content rather than claiming that every original item was individually repaired and certified correct.
Important conceptual corrections
| Original issue | Correct approach |
|---|---|
| SEC1-035 treated VPC Service Controls as a Docker Hub hostname egress filter | Separate supported Google API perimeter controls from internet network/proxy egress |
| SEC4-054 claimed tail sampling preserves errors discarded by earlier head sampling | A downstream sampler can select only received data |
| Guide mixed up burn rates and short/long windows | Use AND within a window pair and OR across alternative burn branches; understand the calculation |
| Guide said Cloud Run has no CUDs | Check eligible Compute flexible commitments and current SKU coverage |
| Guide treated a generic blue/green pattern as a native Cloud Deploy strategy | Distinguish the architecture from supported product configuration |
| Guide said PR deployments and manual actions are always wrong | Evaluate trust boundaries, preview environments and required approvals |
| Guide called vmExternalIpAccess boolean | It is a list constraint |
| Guide treated GKE federation as only a KSA-to-GSA link | Understand direct grants and impersonation patterns |
| Guide suggested budgets universally only notify | Distinguish alerts-only budgets, automation and current eligible-service spend caps |
| Original mock selected questions already used for practice | New mocks have separate question IDs and are excluded from learning mode |
| Correct options heavily favored A/B with no answer shuffling | Options shuffle with their answer mapping on each session |
| Quiz claimed 70% was a typical passing bar | Removed; study targets are explicitly personal and not official thresholds |
Mock state omitted checked while rendering accessed it |
New session handling initializes state consistently and is tested |
Supporting references: VPC Service Controls, sampling, SRE alerting, Cloud Run CUDs, Cloud Deploy strategies, spend caps, exam result FAQ.
Scope and limits of this pack
All 19 numbered objectives in the guide linked by the certification page are mapped to questions. That does not mean every product feature or scenario has been exhausted. The explanations and references support learning; they are not endorsed by Google. Practice questions are more constrained than real production diagnosis. First-attempt mock performance must be considered with labs and independent official samples, especially for a beginner.
The browser's storage is local to a browser/profile and file origin; moving files or clearing storage can lose history. Export results. Mock separation is a study convention, not secrecy: the JSON contains the answers for offline operation. Do not inspect mock data before the first attempt.