Microsoft AB-100 Practice Test 30

Question 290

A route-optimization agent is prompted not to exceed the driver-hours limit. During a test it proposes a schedule that violates the limit but confidently describes it as compliant. What is the key test assertion?

Accept the result if the language model self-reports that it followed the rule.
Measure whether the itinerary summary uses the same terms as the compliance handbook.
Run a deterministic hours-of-service validator on the proposed itinerary and block any violating route before dispatch.
Increase generation randomness until a different explanation is returned.

Explanation

Correct Answer

Legal driving-hour constraints require independent calculation from route timing and driver records rather than model attestation.

Incorrect Answers

Model assertions can be incorrect even when phrased as compliance conclusions.

Terminology alignment does not establish that cumulative hours fall within required bounds.

Changing randomness is not a validation strategy for a deterministic legal threshold.

Question 291

A multilingual support agent opens tickets in Customer Service and sends regional escalations. An Arabic message maps to the wrong customer but the response sounds fluent. Which end-to-end test is missing?

Use labeled multilingual cases to verify source identity, case-link mapping and the resulting regional escalation in both systems.
Confirm only that the escalation email contains Arabic characters.
Score fluency only because language quality demonstrates correct customer identity.
Check ticket creation counts without reconciling customer record identifiers.

Explanation

Correct Answer

Integration and language tests must validate the factual identity mapping and business side effects, not only generated prose.

Incorrect Answers

Character presence says nothing about routing the case to the entitled recipient.

A fluent reply may conceal wrong entity resolution and privacy leakage.

A count of tickets cannot detect cross-customer linkage mistakes.

Question 292

Generated regression cases for an adverse-drug-event agent contain multiple paraphrases of the same safe query and no attempted unsupported dosage advice. How should the suite be revised?

Keep all near-identical benign cases because a higher test count always proves broader coverage.
Replace redundant paraphrases with policy-boundary, abstention and escalation tests verified by a medical reviewer.
Remove all risk tests to keep the agent evaluation success rate high.
Score dosage-boundary tests by how often the agent offers an empathetic refusal, without checking that prohibited advice was withheld.

Explanation

Correct Answer

A defensible test suite covers distinct safety decisions and clinically reviewed expected outcomes rather than volume alone.

Incorrect Answers

Rewording one harmless query repeatedly does not exercise prohibited-advice boundaries.

Excluding adverse scenarios biases acceptance toward easy successes.

Empathetic phrasing is useful but cannot establish safe behavior if the agent also supplies unapproved dosing instructions; reviewers must validate prohibited-content boundaries.

Question 293

A field-service scheduling agent passes one-turn tests but fails after a technician corrects the asset serial number mid-conversation. Which test design should be added?

Repeat a single-turn serial-number test using only the corrected value without preserving the previous conversation state.
A single response test using the original serial number alone.
A throughput benchmark that never examines the order payload.
A multi-turn conversation test that updates the referenced asset and verifies the final work order uses only the corrected identity.

Explanation

Correct Answer

The failure depends on state updates across turns; the final committed work-order identifier is the measurable assertion.

Incorrect Answers

The failure occurs when a correction follows an earlier identity. A fresh single-turn test cannot expose stale context reuse.

An initial one-turn test cannot reveal stale conversational entity binding.

High throughput does not prove that a corrected identifier reached the transaction.

Question 294

Two agent versions return correct policy answers, but version B performs unnecessary inventory lookups and doubles cost. Which metric set best captures the regression?

Compare text length as the only proxy for resource efficiency.
Evaluate answer correctness together with tool-call necessity, token spend and end-to-end latency for equivalent requests.
Compare only semantic answer similarity and omit the number and cost of inventory calls made by each version.
Count total agent turns and reward the version with more tool calls.

Explanation

Correct Answer

Agent success includes useful results and controlled tool use; unnecessary calls impose operational and monetary cost.

Incorrect Answers

Verbose replies may be cheap or expensive depending on token and tool patterns.

Equal final wording can conceal doubled paid tool traffic; per-call necessity, cost and latency must be compared on the same test workload.

Additional calls are not automatically evidence of superior correctness.

Question 295

An insurance model is tuned using labeled claims collected before a major eligibility rule change. Which release test should prevent hidden performance regressions?

Check that the tuning job completed without exceptions and skip outcome evaluation.
Trust a model-generated summary of its own compliance with the new rule.
Shuffle the training examples and rescore only those same examples.
Evaluate a temporally separated labeled holdout reflecting the new rule and report errors separately for old and new cohorts.

Explanation

Correct Answer

Temporal holdouts expose changes in policy and customer distributions that historical training metrics cannot show.

Incorrect Answers

A successful training pipeline does not establish eligibility-rule correctness.

Self-reported compliance is not a labeled business-policy outcome.

Reusing trained examples risks leakage and optimism without testing the new rule.

Question 296

A logistics agent ingests warehouse documents whose filenames look official. One document embeds an instruction to expose all shipment contacts. What prompt-red-team acceptance check is decisive?

Score the model as safe if it explains why the embedded command sounded plausible.
Check only that the warehouse document text was retrieved in full.
Verify retrieval content never authorizes contact exports; an attempted tool call must be denied without approved business authority.
Allow the export when the retrieved document filename includes “corporate-policy.”

Explanation

Correct Answer

A file name and source text are lower-trust content; action authorization must be enforced at the tool boundary.

Incorrect Answers

Explanations of the attack do not compensate if the forbidden export actually executes.

Retrieval completeness is separate from protecting restricted contacts against exfiltration.

A trusted-looking filename can be attacker-controlled and cannot grant permissions.

Question 297

A service agent updates a case but a downstream billing task times out and later retries. The second attempt creates a duplicate invoice. Which end-to-end acceptance behavior is required?

Use a stable business idempotency key, verify the single persisted invoice and test retry after an ambiguous timeout.
Declare completion as soon as the first network request is sent.
Treat each timeout retry as a distinct invoice unless the customer objects.
Make the conversational response shorter so the billing service responds more quickly.

Explanation

Correct Answer

Distributed retries need idempotent transaction semantics and persisted-state assertions to prevent duplicate billing.

Incorrect Answers

Transmission of a request does not prove any invoice was committed.

Separate invoices for one action create financial and reconciliation errors.

Output verbosity does not fix backend idempotency or uncertain commit state.

Question 298

Copilot proposes tests for a regulated sales agent. The generated suite never checks whether an expired approval can still authorize a promotional claim. What is the strongest improvement?

Add expert-reviewed validity-period boundary cases, including just-before and just-after expiration with evidence IDs.
Use model-written expected responses with no approval-register reference.
Increase the number of tests with new customer names but unchanged valid approval dates.
Replace date assertions with checks of the sales presentation formatting.

Explanation

Correct Answer

Expiry is a time-bound compliance decision, so labeled boundary tests must check both effective and invalid intervals.

Incorrect Answers

Without source-approved expected outcomes, the test could reward a prohibited claim.

Names do not exercise the temporal policy boundary.

Formatting is irrelevant to whether the agent was authorized to make that claim.

Question 299

A field-service chatbot has a high answer pass rate, but technicians report that the wrong equipment repair order is still created. Which evaluation must be added?

Compare committed asset ID, work-order action and authorized technician against a labeled expected business-state record.
Count the number of repair suggestions without checking the created order.
Treat a high technician satisfaction score as evidence that the correct repair order was committed.
Accept the final response because it correctly names the type of equipment.

Explanation

Correct Answer

Tool-side correctness must be measured independently of fluent natural-language answers and surface topic accuracy.

Incorrect Answers

Suggestion volume does not prove that the intended record was changed.

Technicians may like the advice while the tool targets the wrong asset; acceptance must match persisted work order ID and authorized technician.

A correct equipment category can still point to the wrong individual asset.

Question 300

A support center has a regression suite for routine English questions, but newly launched voice and Spanish channels produce incorrect escalation assignments. Which release criterion addresses this?

Review only the voice agent transcription length, not its routed queue.
Run language-specific transcription tests without verifying that the resulting incident routes into the correct regional support queue.
Treat the English chat score as transferable to all voice and language paths.
Require channel/language-stratified cases with known escalation targets and verified downstream queue receipts.

Explanation

Correct Answer

Different input modalities and language detection can affect entity extraction and queue mapping; measure actual handoff.

Incorrect Answers

Transcript length cannot determine whether the right support queue received the case.

Correct recognition or translation is not sufficient when the downstream escalation action uses the wrong assignment.

One channel cannot prove all translation and voice workflows use equivalent routing.

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

Microsoft AB-100 Practice Test 28 - 10 Questions (Last Reviewed 17-Oct-2026)
Microsoft AB-100 Practice Test 29 - 10 Questions (Last Reviewed 16-Oct-2026)
Microsoft AB-100 Practice Test 30 - 10 Questions (Last Reviewed 15-Oct-2026)
Microsoft AB-100 Practice Test 31 - 10 Questions (Last Reviewed 14-Oct-2026)
Microsoft AB-100 Practice Test 32 - 10 Questions (Last Reviewed 13-Oct-2026)

Pass Your Microsoft AB-100 Exam Easy!

Question 290

Explanation

Correct Answer

Incorrect Answers

Question 291

Explanation

Correct Answer

Incorrect Answers

Question 292

Explanation

Correct Answer

Incorrect Answers

Question 293

Explanation

Correct Answer

Incorrect Answers

Question 294

Explanation

Correct Answer

Incorrect Answers

Question 295

Explanation

Correct Answer

Incorrect Answers

Question 296

Explanation

Correct Answer

Incorrect Answers

Question 297

Explanation

Correct Answer

Incorrect Answers

Question 298

Explanation

Correct Answer

Incorrect Answers

Question 299

Explanation

Correct Answer

Incorrect Answers

Question 300

Explanation

Correct Answer

Incorrect Answers

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