Exam Bundle

AI assistance can accelerate modernization but requires runtime checks. That recommendation is relevant to a different task: An old synchronous HTTP client must be modernized to a supported asynchronous API. In this case the main need is: A team needs to rename a deprecated helper across source and tests.

Synthetic data can provide examples without exposing real customers. That recommendation is relevant to a different task: A data engineer needs a safe example dataset for feature development. In this case the main need is: A team needs to rename a deprecated helper across source and tests.

Clarity improvements are valuable when behavior is preserved. That recommendation is relevant to a different task: A developer wants to make a data access layer clearer without changing SQL semantics. In this case the main need is: A team needs to rename a deprecated helper across source and tests.

Models can invent credible dependency names and versions. That recommendation is relevant to a different task: An assistant generates a package name that does not exist in the approved dependency registry. In this case the main need is: A generated unit-test example includes real email addresses from an internal issue.

Persuasive language is not proof of security properties. That recommendation is relevant to a different task: A team assumes a fluent explanation proves an unfamiliar cryptographic routine is sound. In this case the main need is: A generated unit-test example includes real email addresses from an internal issue.

Turning off TLS verification makes interception possible. That recommendation is relevant to a different task: A tool suggests bypassing TLS verification to fix an integration error. In this case the main need is: A generated unit-test example includes real email addresses from an internal issue.

“Sensitive files remain protected by filesystem and agent tool restrictions.” does not demonstrate the result required here: The sensitive source is inaccessible to the assistant despite client limitations.

“A test workflow cannot access the secret even when the symlink exists.” does not demonstrate the result required here: The sensitive source is inaccessible to the assistant despite client limitations.

“The approval record names applicable exclusions, roles, and open risks.” does not demonstrate the result required here: The sensitive source is inaccessible to the assistant despite client limitations.

Clients may depend on response codes even if behavior looks similar. That recommendation is relevant to a different task: A model-generated patch changes an endpoint's HTTP status contract. In this case the main need is: An assistant reports that a change is deployed even though only the diff was generated.

Correct-looking query code can hide performance defects. That recommendation is relevant to a different task: An assistant generates an ORM query with possible N+1 database calls. In this case the main need is: An assistant reports that a change is deployed even though only the diff was generated.

Language feature support varies by runtime version. That recommendation is relevant to a different task: A Copilot fix passes unit tests but fails the production language version. In this case the main need is: An assistant reports that a change is deployed even though only the diff was generated.

“Invalid rows are flagged and legitimate rows retain correct types.” does not demonstrate the result required here: Valid tokens pass, invalid tokens fail, and approved verification primitives remain in use.

“An injection fixture is treated as data and cannot change query structure.” does not demonstrate the result required here: Valid tokens pass, invalid tokens fail, and approved verification primitives remain in use.

“Invalid tokens fail and a correctly scoped token succeeds.” does not demonstrate the result required here: Valid tokens pass, invalid tokens fail, and approved verification primitives remain in use.

AI review alerts may be false positives. That recommendation is relevant to a different task: A code review bot finds a suspicious pattern but offers no line-specific evidence. In this case the main need is: An engineer asks Copilot to optimize a loop based solely on vague speed impressions.

Concurrency correctness needs testing beyond sequential unit tests. That recommendation is relevant to a different task: Copilot writes a thread-safe cache claim without synchronization evidence. In this case the main need is: An engineer asks Copilot to optimize a loop based solely on vague speed impressions.

Narrow changes reduce regression surface and ease review. That recommendation is relevant to a different task: An agent proposes changing 14 unrelated modules for a one-line defect. In this case the main need is: An engineer asks Copilot to optimize a loop based solely on vague speed impressions.

A separate focused agent can analyze security risk without modifying the primary worktree. That recommendation is relevant to a different task: A repository agent needs a separate security analysis by a specialized reviewer. In this case the main need is: A development team wants Copilot to review a pull request for logic errors.

Actionable review feedback needs context and reproducibility. That recommendation is relevant to a different task: A reviewer receives an AI-suggested fix with no exact file location. In this case the main need is: A development team wants Copilot to review a pull request for logic errors.

Agent session artifacts provide evidence of actual work. That recommendation is relevant to a different task: An agent has produced multiple commits in a cloud session and the user must inspect changes. In this case the main need is: A development team wants Copilot to review a pull request for logic errors.

Access control must be verified against real principals. That recommendation is relevant to a different task: An assistant writes a REST endpoint that returns all database fields to a normal user. In this case the main need is: An engineer adds Copilot-generated infrastructure YAML with unexpected wildcard permissions.

Authentication patches can be subtly insecure despite compiling. That recommendation is relevant to a different task: Copilot produces a patch for a broken OAuth flow with no test evidence. In this case the main need is: An engineer adds Copilot-generated infrastructure YAML with unexpected wildcard permissions.

Concurrency correctness needs testing beyond sequential unit tests. That recommendation is relevant to a different task: Copilot writes a thread-safe cache claim without synchronization evidence. In this case the main need is: An engineer adds Copilot-generated infrastructure YAML with unexpected wildcard permissions.

Invisible profiling can create privacy and fairness harms. That recommendation is relevant to a different task: A generated feature request proposes profiling users using hidden behavioral signals. In this case the main need is: A Copilot recommendation seems to favor one demographic group in a user-ranking algorithm.

Generative assistance is not an accountable final authority for high-stakes determinations. That recommendation is relevant to a different task: A medical claims app asks Copilot to approve coverage without a human decision point. In this case the main need is: A Copilot recommendation seems to favor one demographic group in a user-ranking algorithm.

Persuasive language is not proof of security properties. That recommendation is relevant to a different task: A team assumes a fluent explanation proves an unfamiliar cryptographic routine is sound. In this case the main need is: A Copilot recommendation seems to favor one demographic group in a user-ranking algorithm.

Higher-level settings can restrict repository-level feature use. That recommendation is relevant to a different task: A code review policy is disabled centrally, but a repo owner expects automatic reviews. In this case the main need is: A compliance officer needs to know who changed Copilot policy after a reported incident.

Audit logs help trace administrative actions and times. That recommendation is relevant to a different task: A platform team wants to audit policy changes made by administrators to Copilot settings. In this case the main need is: A compliance officer needs to know who changed Copilot policy after a reported incident.

Central policies control eligible feature availability across accounts and surfaces. That recommendation is relevant to a different task: A company wants to disable agent mode for some Copilot users while retaining chat access. In this case the main need is: A compliance officer needs to know who changed Copilot policy after a reported incident.

“An approved harmless request succeeds while the disallowed output remains suppressed.” does not demonstrate the result required here: The answer reflects the referenced function description with the correct path.

“Changing the relevant comment changes the resulting proposal in an expected way.” does not demonstrate the result required here: The answer reflects the referenced function description with the correct path.

“The response changes appropriately when the precise relevant file is attached.” does not demonstrate the result required here: The answer reflects the referenced function description with the correct path.