McAfee Secure

Cisco 350-901 Bundle

Exam Code: 350-901

Exam Name Designing, Deploying, and Managing Network Automation Systems

Certification Provider: Cisco

Corresponding Certification: DevNet Professional

Cisco 350-901 Bundle $25.00

100% Cisco 350-901 Guaranteed Success With Testking Training Materials

Get Real 350-901 Exam Questions & Expert Verified Answers as Experienced on the Real Exam!

  • Questions & Answers

    350-901 Questions & Answers

    604 Questions & Answers

    The ultimate exam preparation tool, Questions & Answers mirrors the actual 350-901 exam while allowing you to hone and measure every aspect of your test-taking ability.

  • 350-901 Video Course

    350-901 Training Course

    106 Video Lectures

    350-901 Video Course is developed by Cisco Professionals to help you pass the 350-901 exam.

    Description

    This course will improve your knowledge and skills required to pass Designing, Deploying, and Managing Network Automation Systems exam.

Designing, Deploying, and Managing Network Automation Systems (350-901) Practice Tests

Satisfaction Guaranteed

Money Back Guarantee

Testking provides hassle-free money back guarantee with our products. That is because we have 100% trust in the abilities of our professional and experience product team, and our record is a proof of that.

99.6% PASS RATE
Total Cost: $164.98
Bundle Price: $139.98

Purchase Individually

  • Questions & Answers

    Questions & Answers

    604 Questions

    $124.99
  • 350-901 Video Course

    Training Course

    106 Video Lectures

    $39.99

Cisco 350-901 Practice Test 51

Question 500

While investigating a FastMCP security review for fictional Elmcrest Theatre, the team reports: A retrieved configuration note includes a disguised URL that would leak network inventory if the agent automatically opens it. Which targeted correction best addresses the network automation failure?

  1. Check AI conclusions against fresh independent device and controller evidence with explicit freshness and confidence rules.
  2. Restrict tool egress to approved destinations and never grant untrusted retrieved text authority to choose exfiltration endpoints.
  3. Separate read-only diagnostic tools from mutating tools and require an explicit scoped approval step before any privileged operation.
  4. Review generated code provenance, included dependencies and organizational IP policy before accepting it into the repository.

Explanation

Correct Answer

B:

Restrict tool egress to approved destinations and never grant untrusted retrieved text authority to choose exfiltration endpoints. This addresses the problem because tool results are untrusted external input; unrestricted outbound connectivity turns an injected instruction into a data-exfiltration path. Confirm success using: A simulated malicious tool response cannot trigger outbound requests to unapproved destinations or reveal hidden credentials. Applied here, the relevant facts are: A retrieved configuration note includes a disguised URL that would leak network inventory if the agent automatically opens it. The question asks which targeted correction best addresses the network automation failure.

Incorrect Answers

A:

The proposed action belongs to the mechanism: aI recommendations are hypotheses requiring external ground truth; absent evidence is not proof of a network condition. Here the stated fault follows a different mechanism: tool results are untrusted external input; unrestricted outbound connectivity turns an injected instruction into a data-exfiltration path. The corrective choice is “Restrict tool egress to approved destinations and never grant untrusted retrieved text authority to choose exfiltration endpoints”.

C:

This alternative is a response to another technical cause: agent tool chaining can enlarge the blast radius of an ambiguous user instruction unless execution permissions are independently gated. The reported case instead depends on: tool results are untrusted external input; unrestricted outbound connectivity turns an injected instruction into a data-exfiltration path. The action that directly addresses it is “Restrict tool egress to approved destinations and never grant untrusted retrieved text authority to choose exfiltration endpoints”.

D:

The proposed action belongs to the mechanism: aI-generated code may embed third-party patterns or dependencies; tool output is not automatically cleared of licensing obligations. Here the stated fault follows a different mechanism: tool results are untrusted external input; unrestricted outbound connectivity turns an injected instruction into a data-exfiltration path. The corrective choice is “Restrict tool egress to approved destinations and never grant untrusted retrieved text authority to choose exfiltration endpoints”.

Question 501

At fictional Seabright Learning, an acceptance review of a reproducible simulation finds: A developer needs one last-known-good YANG mapping file but checks out an entire outdated branch over current production code. Which result most directly verifies that the proposed correction worked?

  1. The target branch retains an auditable inverse commit and the published history remains intact while behavior returns to the approved state.
  2. Only the selected file matches the approved historical content, and unrelated working files remain unchanged.
  3. The Compose deployment starts required backends and the worker reaches ready state even when a dependency starts slowly.
  4. Generated configs match authoritative source records and postapply compliance detects any remaining unauthorized deviation.

Explanation

Correct Answer

B:

Only the selected file matches the approved historical content, and unrelated working files remain unchanged. This directly verifies the related correction: Restore only the specifically reviewed historical file from the desired revision instead of replacing the whole branch state. It checks the technical distinction: Git can retrieve individual tracked paths from a commit; selecting file scope avoids broad branch-level rollback. Applied here, the relevant facts are: A developer needs one last-known-good YANG mapping file but checks out an entire outdated branch over current production code. The question asks which result most directly verifies that the proposed correction worked.

Incorrect Answers

A:

The offered action targets: This directly verifies the related correction: Create a reviewed Git revert commit for the offending change and run the full network automation release tests. The case facts point instead to: This directly verifies the related correction: Restore only the specifically reviewed historical file from the desired revision instead of replacing the whole branch state. Implement “Only the selected file matches the approved historical content, and unrelated working files remain unchanged” to address that specific condition.

C:

The proposed action belongs to the mechanism: This directly verifies the related correction: Define service dependencies with appropriate health checks and resilient application startup or retry behavior. Here the stated fault follows a different mechanism: This directly verifies the related correction: Restore only the specifically reviewed historical file from the desired revision instead of replacing the whole branch state. The corrective choice is “Only the selected file matches the approved historical content, and unrelated working files remain unchanged”.

D:

The offered action targets: This directly verifies the related correction: Read authorized desired state from the designated source of truth and reconcile observed drift through an approved change path. The case facts point instead to: This directly verifies the related correction: Restore only the specifically reviewed historical file from the desired revision instead of replacing the whole branch state. Implement “Only the selected file matches the approved historical content, and unrelated working files remain unchanged” to address that specific condition.

Question 502

During a source-of-truth integration at fictional Violet Ridge Resorts, the change record states: A Terraform workflow change was merged and released, but its logic must be undone without rewriting published branch history. Which targeted correction best addresses the network automation failure?

  1. Declare and lock the required dependency in the build manifest and install it in the pipeline's reproducible environment.
  2. Construct JSON using the published YANG container hierarchy, qualified node names and correct value types.
  3. Mount an appropriate persistent named volume or approved bind mount for data that must survive container recreation.
  4. Create a reviewed Git revert commit for the offending change and run the full network automation release tests.

Explanation

Correct Answer

D:

Create a reviewed Git revert commit for the offending change and run the full network automation release tests. This addresses the problem because git revert records an inverse patch in new history rather than moving shared branch pointers backward. Confirm success using: The target branch retains an auditable inverse commit and the published history remains intact while behavior returns to the approved state. Applied here, the relevant facts are: A Terraform workflow change was merged and released, but its logic must be undone without rewriting published branch history. The question asks which targeted correction best addresses the network automation failure.

Incorrect Answers

A:

This alternative is a response to another technical cause: cI runners start from controlled environments; undeclared local dependencies cannot be assumed present remotely. The reported case instead depends on: git revert records an inverse patch in new history rather than moving shared branch pointers backward. The action that directly addresses it is “Create a reviewed Git revert commit for the offending change and run the full network automation release tests”.

B:

The proposed action belongs to the mechanism: yANG defines node hierarchy, namespaces, list keys and data types; conforming JSON must preserve that structure. Here the stated fault follows a different mechanism: git revert records an inverse patch in new history rather than moving shared branch pointers backward. The corrective choice is “Create a reviewed Git revert commit for the offending change and run the full network automation release tests”.

C:

The offered action targets: ephemeral container writable layers are distinct from Docker managed or host-mounted persistent volumes. The case facts point instead to: git revert records an inverse patch in new history rather than moving shared branch pointers backward. Implement “Create a reviewed Git revert commit for the offending change and run the full network automation release tests” to address that specific condition.

Question 503

During an AI recommendation accuracy study at fictional Amberlight Energy, the change record states: A developer sends internal Cisco API bearer tokens and customer inventory records to an unapproved hosted LLM to generate a Terraform module. What should the engineer change first to resolve this specific issue?

  1. Expose narrowly scoped typed read-only FastMCP tools for approved device and status fields with controlled access.
  2. Ground actionable recommendations in trusted versioned technical references and reject unsupported or conflicting retrieved claims.
  3. Sanitize inputs and use an approved private or contractually controlled AI service without exposing secrets or regulated network data.
  4. Authenticate each MCP client, authorize its tool calls by tenant identity and isolate credential/session context.

Explanation

Correct Answer

C:

Sanitize inputs and use an approved private or contractually controlled AI service without exposing secrets or regulated network data. This addresses the problem because aI-assisted code tools may process submitted data outside the intended trust boundary, creating privacy and intellectual-property exposure. Confirm success using: Audit logs and data-loss-prevention checks show that prompts and tool responses contain no live credentials or confidential customer payloads. Applied here, the relevant facts are: A developer sends internal Cisco API bearer tokens and customer inventory records to an unapproved hosted LLM to generate a Terraform module. The question asks what should the engineer change first to resolve this specific issue.

Incorrect Answers

A:

The proposed action belongs to the mechanism: an MCP server publishes structured tools and resources to clients; safe network use requires explicit least-privilege tool definitions. Here the stated fault follows a different mechanism: aI-assisted code tools may process submitted data outside the intended trust boundary, creating privacy and intellectual-property exposure. The corrective choice is “Sanitize inputs and use an approved private or contractually controlled AI service without exposing secrets or regulated network data”.

B:

The offered action targets: retrieval-augmented systems can inherit false or hostile data; source quality and technical cross-checks must precede action. The case facts point instead to: aI-assisted code tools may process submitted data outside the intended trust boundary, creating privacy and intellectual-property exposure. Implement “Sanitize inputs and use an approved private or contractually controlled AI service without exposing secrets or regulated network data” to address that specific condition.

D:

The proposed action belongs to the mechanism: an MCP server is an integration boundary and must not treat model-selected identity strings as verified security credentials. Here the stated fault follows a different mechanism: aI-assisted code tools may process submitted data outside the intended trust boundary, creating privacy and intellectual-property exposure. The corrective choice is “Sanitize inputs and use an approved private or contractually controlled AI service without exposing secrets or regulated network data”.

Question 504

At fictional Cobalt Ridge Bank, an acceptance review of a campus API automation rollout finds: A controller script issues interface changes using a module namespace unavailable on a branch router's installed version. What ongoing engineering assurance should follow the successful fix?

  1. Exercise representative 4xx, 5xx, timeout and malformed-payload cases in contract tests before scheduling batches.
  2. Add check-mode previews and a second-run idempotence assertion to the network playbook CI gate.
  3. Record per-platform model support and gate model-specific deploy steps on device capability checks.
  4. Require terminal-state checks, timeouts and alerting for asynchronous API-driven network changes.

Explanation

Correct Answer

C:

Record per-platform model support and gate model-specific deploy steps on device capability checks. This prevents the recurrence because model-driven APIs are available only for modules and revisions implemented by the device software, not by the client library alone. A matching immediate correction is: Discover supported YANG modules or capabilities before using module-specific RESTCONF data paths, and branch for supported models.

Incorrect Answers

A:

That safeguard is relevant to: robust batch automation isolates errors at the per-resource boundary and retains auditable partial outcomes. The current failure instead follows from: model-driven APIs are available only for modules and revisions implemented by the device software, not by the client library alone. Its prevention measure is “Record per-platform model support and gate model-specific deploy steps on device capability checks”.

B:

This routine guards against another mechanism: ansible's desired-state-aware network modules compare declared configuration with device state before determining required changes. The recurring exposure in the current case is: model-driven APIs are available only for modules and revisions implemented by the device software, not by the client library alone. The matching preventive check is “Record per-platform model support and gate model-specific deploy steps on device capability checks”.

D:

The proposed control “Require terminal-state checks, timeouts and alerting for asynchronous API-driven network changes” maintains a separate dependency: An HTTP 202 response indicates accepted asynchronous work; a separate completion status is needed to prove the requested device change actually finished. The recurring issue in this case instead follows from: model-driven APIs are available only for modules and revisions implemented by the device software, not by the client library alone. The targeted review is “Record per-platform model support and gate model-specific deploy steps on device capability checks”.

Question 505

At fictional Sunrise Aeronautics, during an AI-assisted code review, the operator observes: A network assistant scores well on easy inventory questions but fails when simulated device logs contain conflicting routing states. Which control best prevents this particular failure from recurring?

  1. Mark generated code for extra negative tests and prohibit bypassing deployment gates based on model confidence.
  2. Review scopes per AI tool, segregate read/write services and require explicit elevated-change approvals.
  3. Adopt AI data-classification rules, approved endpoints and automatic prompt redaction checks for developer workflows.
  4. Maintain a versioned adversarial evaluation suite and block regressions in unsafe recommendation rates.

Explanation

Correct Answer

D:

Maintain a versioned adversarial evaluation suite and block regressions in unsafe recommendation rates. This prevents the recurrence because high aggregate answer accuracy can hide safety-critical errors; validation must measure behavior on representative failure modes. A matching immediate correction is: Evaluate the agent on a representative labeled test set with difficult, adversarial and negative cases, measuring factual error and unsafe-action rates. Applied here, the relevant facts are: A network assistant scores well on easy inventory questions but fails when simulated device logs contain conflicting routing states. The question asks which control best prevents this particular failure from recurring.

Incorrect Answers

A:

This routine guards against another mechanism: generative systems can propose plausible-looking but incorrect commands; independent executable validation is required for safe infrastructure changes. The recurring exposure in the current case is: high aggregate answer accuracy can hide safety-critical errors; validation must measure behavior on representative failure modes. The matching preventive check is “Maintain a versioned adversarial evaluation suite and block regressions in unsafe recommendation rates”.

B:

The suggested review tracks a different technical risk: the model is not a reliable security principal; tool identities and server-side authorization bound the damage of agent errors. Here the sustained control must address: high aggregate answer accuracy can hide safety-critical errors; validation must measure behavior on representative failure modes. Choose “Maintain a versioned adversarial evaluation suite and block regressions in unsafe recommendation rates”.

C:

That safeguard is relevant to: aI-assisted code tools may process submitted data outside the intended trust boundary, creating privacy and intellectual-property exposure. The current failure instead follows from: high aggregate answer accuracy can hide safety-critical errors; validation must measure behavior on representative failure modes. Its prevention measure is “Maintain a versioned adversarial evaluation suite and block regressions in unsafe recommendation rates”.

Question 506

At fictional Westlake Biomedicine, an acceptance review of a FastMCP security review finds: An internal FastMCP server trusts a caller-supplied user name and gives any connected chat agent access to all device telemetry. What should the engineer change first to resolve this specific issue?

  1. Use AI assistance for bounded drafts and test generation while assigning accountable reviewers and validating facts against approved sources.
  2. Sanitize inputs and use an approved private or contractually controlled AI service without exposing secrets or regulated network data.
  3. Require citation to a compatible authoritative reference and fail closed or escalate when command syntax or device version is uncertain.
  4. Authenticate each MCP client, authorize its tool calls by tenant identity and isolate credential/session context.

Explanation

Correct Answer

D:

Authenticate each MCP client, authorize its tool calls by tenant identity and isolate credential/session context. This addresses the problem because an MCP server is an integration boundary and must not treat model-selected identity strings as verified security credentials. Confirm success using: Two simulated tenants can query only their own device records and cross-tenant tool attempts are denied and audited. Applied here, the relevant facts are: An internal FastMCP server trusts a caller-supplied user name and gives any connected chat agent access to all device telemetry. The question asks what should the engineer change first to resolve this specific issue.

Incorrect Answers

A:

The proposed action belongs to the mechanism: aI can accelerate low-risk preparation, but speed benefits must be balanced against hallucination, privacy and review obligations. Here the stated fault follows a different mechanism: an MCP server is an integration boundary and must not treat model-selected identity strings as verified security credentials. The corrective choice is “Authenticate each MCP client, authorize its tool calls by tenant identity and isolate credential/session context”.

B:

The offered action targets: aI-assisted code tools may process submitted data outside the intended trust boundary, creating privacy and intellectual-property exposure. The case facts point instead to: an MCP server is an integration boundary and must not treat model-selected identity strings as verified security credentials. Implement “Authenticate each MCP client, authorize its tool calls by tenant identity and isolate credential/session context” to address that specific condition.

C:

This alternative is a response to another technical cause: model fluency is not evidence of device capability; uncertain technical recommendations require explicit verification before action. The reported case instead depends on: an MCP server is an integration boundary and must not treat model-selected identity strings as verified security credentials. The action that directly addresses it is “Authenticate each MCP client, authorize its tool calls by tenant identity and isolate credential/session context”.

Question 507

While investigating a FastMCP security review for fictional Elmcrest Theatre, the team reports: A user switches sites in the same session, but the assistant carries the earlier site's tenant scope into a new API query. What should the team add to its network automation operating process?

  1. Review FastMCP tool schemas, allowed methods, output redaction and per-device access before agent connection.
  2. Enforce identity binding, tenant-isolated caches and tool-level audit and RBAC tests for MCP services.
  3. Adopt AI data-classification rules, approved endpoints and automatic prompt redaction checks for developer workflows.
  4. Expire cached network context, confirm target scope and revalidate resources before agent actions.

Explanation

Correct Answer

D:

Expire cached network context, confirm target scope and revalidate resources before agent actions. This prevents the recurrence because conversation memory is not authoritative network state; stale context and scope leakage create wrong-target operational risk. A matching immediate correction is: Bind current request context to fresh authorized inventory and explicit tenant or device selection before tool calls.

Incorrect Answers

A:

The offered action targets: an MCP server publishes structured tools and resources to clients; safe network use requires explicit least-privilege tool definitions. The case facts point instead to: conversation memory is not authoritative network state; stale context and scope leakage create wrong-target operational risk. Implement “Expire cached network context, confirm target scope and revalidate resources before agent actions” to address that specific condition.

B:

This alternative is a response to another technical cause: an MCP server is an integration boundary and must not treat model-selected identity strings as verified security credentials. The reported case instead depends on: conversation memory is not authoritative network state; stale context and scope leakage create wrong-target operational risk. The action that directly addresses it is “Expire cached network context, confirm target scope and revalidate resources before agent actions”.

C:

The proposed action belongs to the mechanism: aI-assisted code tools may process submitted data outside the intended trust boundary, creating privacy and intellectual-property exposure. Here the stated fault follows a different mechanism: conversation memory is not authoritative network state; stale context and scope leakage create wrong-target operational risk. The corrective choice is “Expire cached network context, confirm target scope and revalidate resources before agent actions”.

Question 508

While investigating a source-of-truth integration for fictional Willowfield Hospitality, the team reports: A GitLab CI build succeeds on a developer laptop but fails in the runner because the automation package was never declared as a dependency. What technical evidence should be required before closing this change?

  1. A fresh isolated runner installs the dependency from the manifest and completes the build without relying on developer machine state.
  2. The working tree reports the intended revision and no uncommitted cross-branch edits before test execution.
  3. The target branch builds and tests cleanly with only the required backported changes and no missing dependency references.
  4. A deliberately invalid network change blocks before deploy and a valid change reaches the deployment stage only after passing gates.

Explanation

Correct Answer

A:

A fresh isolated runner installs the dependency from the manifest and completes the build without relying on developer machine state. This directly verifies the related correction: Declare and lock the required dependency in the build manifest and install it in the pipeline's reproducible environment. It checks the technical distinction: CI runners start from controlled environments; undeclared local dependencies cannot be assumed present remotely. Applied here, the relevant facts are: A GitLab CI build succeeds on a developer laptop but fails in the runner because the automation package was never declared as a dependency. The question asks what technical evidence should be required before closing this change.

Incorrect Answers

B:

The alternative verifies a different condition: This directly verifies the related correction: Safely stash or commit unrelated work and check out the intended branch or detached revision in a clean worktree. The current case instead depends on: the automation package was never declared as a dependency. Confirm that “A fresh isolated runner installs the dependency from the manifest and completes the build without relying on developer machine state”.

C:

That test belongs to another control with this basis: This directly verifies the related correction: Identify the prerequisite commits and cherry-pick a minimal compatible sequence or adapt the patch for the target baseline. In the current case, the governing reason is: the automation package was never declared as a dependency. The direct verification is “A fresh isolated runner installs the dependency from the manifest and completes the build without relying on developer machine state”.

D:

This observation would test a different technical dependency: This directly verifies the related correction: Make deployment depend on successful prevalidation jobs for schema, policy, simulation and approval checks. The current issue is governed by: the automation package was never declared as a dependency. Its decisive check is “A fresh isolated runner installs the dependency from the manifest and completes the build without relying on developer machine state”.

Question 509

While investigating a network policy migration for fictional Cedarbridge Care Systems, the team reports: A generalized automation assumes every switch supports the same vendor YANG augmentation and fails against older IOS XE images. What ongoing engineering assurance should follow the successful fix?

  1. Require validated backups, staged batches and executable rollback procedures in approved network playbooks.
  2. Record per-platform model support and gate model-specific deploy steps on device capability checks.
  3. Require terminal-state checks, timeouts and alerting for asynchronous API-driven network changes.
  4. Require remote state access controls, locking verification and serialized deployment jobs for shared workspaces.

Explanation

Correct Answer

B:

Record per-platform model support and gate model-specific deploy steps on device capability checks. This prevents the recurrence because model-driven APIs are available only for modules and revisions implemented by the device software, not by the client library alone. A matching immediate correction is: Discover supported YANG modules or capabilities before using module-specific RESTCONF data paths, and branch for supported models. Applied here, the relevant facts are: A generalized automation assumes every switch supports the same vendor YANG augmentation and fails against older IOS XE images. The question asks what ongoing engineering assurance should follow the successful fix.

Incorrect Answers

A:

The suggested review tracks a different technical risk: safe network automation separates preflight, controlled deployment, verification and recovery instead of assuming all endpoints will succeed. Here the sustained control must address: model-driven APIs are available only for modules and revisions implemented by the device software, not by the client library alone. Choose “Record per-platform model support and gate model-specific deploy steps on device capability checks”.

C:

The proposed control “Require terminal-state checks, timeouts and alerting for asynchronous API-driven network changes” maintains a separate dependency: An HTTP 202 response indicates accepted asynchronous work; a separate completion status is needed to prove the requested device change actually finished. The recurring issue in this case instead follows from: model-driven APIs are available only for modules and revisions implemented by the device software, not by the client library alone. The targeted review is “Record per-platform model support and gate model-specific deploy steps on device capability checks”.

D:

The suggested review tracks a different technical risk: terraform state tracks managed resource associations; concurrent writes without locking can corrupt effective desired-state coordination. Here the sustained control must address: model-driven APIs are available only for modules and revisions implemented by the device software, not by the client library alone. Choose “Record per-platform model support and gate model-specific deploy steps on device capability checks”.

Question 510

At fictional Seabright Learning, an acceptance review of a reproducible simulation finds: A deployment stage targets all routers at once despite an approved plan to canary one site before fleet rollout. What technical evidence should be required before closing this change?

  1. A fresh isolated runner installs the dependency from the manifest and completes the build without relying on developer machine state.
  2. GitLab deployment records show the canary passed before other sites changed, and credentials cannot access unrelated workspaces.
  3. The working tree reports the intended revision and no uncommitted cross-branch edits before test execution.
  4. A recreated container reads the expected persistent audit or inventory records from the configured volume.

Explanation

Correct Answer

B:

GitLab deployment records show the canary passed before other sites changed, and credentials cannot access unrelated workspaces. This directly verifies the related correction: Deploy through bounded environment-scoped jobs with approved service credentials and staged canary or batch rollout. It checks the technical distinction: A deployment stage should enforce release scope and approval, not merely execute a command with unrestricted infrastructure access. Applied here, the relevant facts are: A deployment stage targets all routers at once despite an approved plan to canary one site before fleet rollout. The question asks what technical evidence should be required before closing this change.

Incorrect Answers

A:

That test belongs to another control with this basis: This directly verifies the related correction: Declare and lock the required dependency in the build manifest and install it in the pipeline's reproducible environment. In the current case, the governing reason is: This directly verifies the related correction: Deploy through bounded environment-scoped jobs with approved service credentials and staged canary or batch rollout. The direct verification is “GitLab deployment records show the canary passed before other sites changed, and credentials cannot access unrelated workspaces”.

C:

The alternative verifies a different condition: This directly verifies the related correction: Safely stash or commit unrelated work and check out the intended branch or detached revision in a clean worktree. The current case instead depends on: This directly verifies the related correction: Deploy through bounded environment-scoped jobs with approved service credentials and staged canary or batch rollout. Confirm that “GitLab deployment records show the canary passed before other sites changed, and credentials cannot access unrelated workspaces”.

D:

That test belongs to another control with this basis: This directly verifies the related correction: Mount an appropriate persistent named volume or approved bind mount for data that must survive container recreation. In the current case, the governing reason is: This directly verifies the related correction: Deploy through bounded environment-scoped jobs with approved service credentials and staged canary or batch rollout. The direct verification is “GitLab deployment records show the canary passed before other sites changed, and credentials cannot access unrelated workspaces”.