Microsoft AB-100 Practice Test 78

Question 770

A test-generation Copilot needs examples for order returns. Developers propose feeding it a complete production customer database. What alternative supports safe coverage?

Publish raw customer records as examples in an open test repository.
Allow the generator to invent an expected refund status without source validation.
Use only one happy-path synthetic order for every scenario.
Use de-identified representative fixtures and review generated negative cases against approved business rules.

Explanation

Correct Answer

Test breadth does not require uncontrolled disclosure of real customer identities.

Incorrect Answers

Private orders and contact details are unrelated to a public fixture requirement.

Expected outcomes must come from the actual policy and transaction contract.

One example cannot cover failed authorization, returns and partial commits.

Question 771

A Copilot-generated test calls `create_order` without a customer ID, then expects HTTP 201. The API contract requires that field. How should the test be handled?

Reclassify it as a negative validation case expecting rejection, or add the field for a legitimate success case.
Change the API contract silently to accept all anonymous orders.
Leave the expected 201 because the test was machine generated.
Skip schema-related tests to focus on language quality.

Explanation

Correct Answer

Expected status codes must reflect the real action schema rather than a guessed generator output.

Incorrect Answers

Weakening required identity validation damages transactional integrity.

Automation does not grant an expected result technical authority.

Tool argument correctness is a required part of end-to-end agent safety.

Question 772

An AI test generator produces 60 cases for an invoice bot, many differing only by customer name. What review protects actual coverage?

Use generated titles as a substitute for the expected backend result.
Group by distinct tested decision and failure mechanism, retaining diverse numeric, authorization and dependency boundaries.
Count all 60 as independent coverage because their texts differ.
Delete every repeated case without checking if conditions differ.

Explanation

Correct Answer

Changing names alone does not add a new behavior or evaluation outcome.

Incorrect Answers

Title differences do not establish distinct transaction or access tests.

Surface variation is not a meaningful increase in tested decision coverage.

Two similar stems may still test different technical paths or boundaries.

Question 773

A Copilot proposes access tests for an HR assistant only while employees have current roles. An employee transferring departments can retain cached access. What missing test is needed?

Reuse old cached data as the expected answer after a role change.
Ask the model to avoid citing the old department while still retrieving its files.
Revoke the role, repeat the same restricted retrieval and assert immediate denial after cache invalidation.
Add another test where a permanent administrator reads the policy.

Explanation

Correct Answer

Access lifecycle behavior includes changed permissions, not merely initial successful grants.

Incorrect Answers

Keeping revoked content violates the updated permission boundary.

Prompt instructions cannot enforce authoritative revocation in cached search.

Admin success does not prove removed staff lose access.

Question 774

A Copilot-generated test suite is promoted into CI without recording who validated its expected results. Which governance improvement is necessary?

Let the model self-approve every generated expected result.
Change expected results automatically whenever the code output changes.
Use test file creation timestamps as the sole acceptance evidence.
Version test inputs and expected outcomes with human reviewer, source specification and approval history.

Explanation

Correct Answer

Test-generation assistance does not remove accountability for correctness of acceptance criteria.

Incorrect Answers

Correlated model errors can become false gold standards in CI.

Self-updating expectations conceal regressions rather than catch them.

File age does not establish review or policy traceability.

Question 775

A generated suite includes 100 invoice text-format checks but no double-payment or approval-denial tests. What should reviewers add?

High-risk cases covering duplicate transaction prevention, threshold authorization and rejected payments.
Another hundred punctuation variations of the same invoice label.
Only average response-latency tests after success.
Assume a large generated suite is complete by its question count.

Explanation

Correct Answer

Risk-based test coverage prioritizes irreversible financial errors over cosmetic format breadth.

Incorrect Answers

Additional presentation cases do not exercise missing transaction guardrails.

Performance evidence cannot prove that unauthorized payments are blocked.

Raw scenario quantity is not evidence of distinct failure-mode coverage.

Question 776

An HR policy agent is deployed with prompt v3, but its index updates independently each night. An auditor cannot reproduce yesterday’s answer. What should release records include?

Only the date of the most recent employee login.
An immutable prompt/model configuration reference and the effective indexed source snapshot or revision manifest.
Only the chat interface build number.
An approximate document title list without revision hashes.

Explanation

Correct Answer

Answer reproduction requires knowing which knowledge versions were available at the time of inference.

Incorrect Answers

UI version does not identify the policy passages used by the model.

Titles can stay constant while legally material clauses change.

Question 777

An invoice assistant’s model and prompt are unchanged, but a newly indexed supplier master includes duplicate invalid vendor IDs. Which rollback boundary is most direct?

Change every action schema so any vendor ID is accepted.
Keep the bad index because the model was previously approved.
Restore the previous approved supplier data/index snapshot while preserving model and prompt artifact identity.
Retrain the language model before checking supplier data.

Explanation

Correct Answer

Data lineage makes it possible to recover the failed source change without unnecessarily modifying the model.

Incorrect Answers

Looser validation makes the wrong-supplier defect more dangerous.

Model approval does not validate later uncontrolled source updates.

Retraining cannot correct a corrupted authoritative vendor mapping.

Question 778

A customer invokes an approved deletion request for a document. The source file is removed, but an agent’s vector cache still returns its content. What must the data ALM workflow implement?

Remove only the visible source URL and leave all extracted text.
Change the chunk title to “archived” while serving it as normal.
Tell the agent not to mention the removed document if asked.
Delete or expire authorized derivative chunks, embeddings and cached responses tied to the source identifier.

Explanation

Correct Answer

Source deletion obligations may apply to searchable derivative stores that retain the original content.

Incorrect Answers

The sensitive information remains accessible through the agent despite a missing link.

Changing labels does not satisfy deletion or access revocation.

Prompt instructions cannot guarantee the old data is absent from its retrieval context.

Question 779

An updated chunker splits legal clauses across paragraphs and retrieval starts returning incomplete obligation text. What change-management practice is appropriate?

Treat chunking configuration and index build as versioned data artifacts with regression queries and rollback.
Treat chunking as an invisible implementation detail requiring no tests.
Increase generator temperature to replace missing clause context.
Merge all legal policies into one unbounded prompt to bypass indexing.

Explanation

Correct Answer

Retrieval structure affects which evidence is available even when the source PDFs did not change.

Incorrect Answers

Changes in passage boundaries can alter the meaning of cited legal terms.

Randomness cannot restore omitted text required for a faithful answer.

Unscoped full-document context risks leakage, cost and source-revision confusion.

Question 780

A knowledge pipeline adds `effective_to` and changes document IDs, but an old agent build expects only `effective_from`. What does coordinated release require?

Deploy the index before checking whether consumers can parse its fields.
Introduce a compatibility-tested schema migration and update dependent retrieval filters before activating the new index.
Ignore `effective_to` so superseded documents remain eligible.
Allow the model to guess document IDs when lookup fails.

Explanation

Correct Answer

Authoritative effective-interval filtering must remain consistent across data and agent versions.

Incorrect Answers

Uncoordinated breaking changes can cause runtime failures or missing authorization filters.

Loss of expiration semantics can produce obsolete policy advice.

Generated identifiers do not establish source provenance or valid retrieval links.

Agentic AI Business Solutions Architect (AB-100) Practice Tests

Microsoft AB-100 Practice Test 76 - 10 Questions (Last Reviewed 26-Oct-2026)
Microsoft AB-100 Practice Test 77 - 10 Questions (Last Reviewed 25-Oct-2026)
Microsoft AB-100 Practice Test 78 - 10 Questions (Last Reviewed 24-Oct-2026)
Microsoft AB-100 Practice Test 79 - 10 Questions (Last Reviewed 23-Oct-2026)
Microsoft AB-100 Practice Test 80 - 10 Questions (Last Reviewed 22-Oct-2026)

Pass Your Microsoft AB-100 Exam Easy!

Question 770

Explanation

Correct Answer

Incorrect Answers

Question 771

Explanation

Correct Answer

Incorrect Answers

Question 772

Explanation

Correct Answer

Incorrect Answers

Question 773

Explanation

Correct Answer

Incorrect Answers

Question 774

Explanation

Correct Answer

Incorrect Answers

Question 775

Explanation

Correct Answer

Incorrect Answers

Question 776

Explanation

Correct Answer

Incorrect Answers

Question 777

Explanation

Correct Answer

Incorrect Answers

Question 778

Explanation

Correct Answer

Incorrect Answers

Question 779

Explanation

Correct Answer

Incorrect Answers

Question 780

Explanation

Correct Answer

Incorrect Answers

Agentic AI Business Solutions Architect (AB-100) Practice Tests