Question 100
A tool returns an authorization-denied error. Which retry policy is appropriate?
- Remove authorization headers to bypass the denial.
- Resolve the permission or identity problem before retrying the protected operation.
- Retry rapidly because all errors are transient.
- Switch to another user's credentials from prior context.
Explanation
Correct Answer
B:Authorization failures require a valid access path, not merely more attempts. The assistant should preserve the error and seek a supported remedy.
Incorrect Answers
A:Dropping credentials does not create permission and may change the failure mode.
C:An unchanged unauthorized request will generally remain unauthorized and can create unnecessary load.
D:Another identity cannot be assumed authorized for this user's task.
Question 101
A human approval screen shows only an agent's confidence score before a destructive operation. What information is missing for meaningful review?
- A default-selected approve button to reduce the operator's number of clicks.
- The exact target, proposed change, relevant evidence, and expected consequences.
- The total token count and generation duration, without details of the proposed operation.
- The model version and its general benchmark score, without the actual proposed changes.
Explanation
Correct Answer
B:Reviewers need actionable context to approve the actual operation. A confidence score alone cannot establish informed authorization.
Incorrect Answers
A:Convenient interaction does not supply the missing basis for judgment. The operator still needs to see what action and consequence they are approving.
C:These metrics describe computation, not the target or consequence of the destructive action. They do not enable an informed approval decision.
D:General model performance does not establish whether this particular operation has the correct target and scope. The review must expose the action itself.
Question 102
A workflow saves an approved job in a database and then publishes its task to a queue. A crash between those operations leaves saved jobs that are never dispatched. The database can atomically commit multiple local records but cannot share a transaction with the queue. Which design closes this loss window?
- Wrap only the job update in a database transaction and publish after commit.
- Publish to the queue first and then save the job in the database.
- Commit the job and an outbox entry together, then relay pending entries with duplicate-safe delivery.
- Retry queue publication only inside the process that accepted the job.
Explanation
Correct Answer
C:The atomic local commit preserves both the job and the obligation to publish it. A relay can recover pending entries after a crash; duplicate-safe consumption handles a relay failure after publication but before acknowledgment.
Incorrect Answers
A:The database transaction protects the job record but not the subsequent queue operation. This is the same gap described in the scenario.
B:Reversing the order creates the opposite inconsistency: a worker can receive a task whose database transaction never commits. Two separate writes still leave a crash window.
D:An in-memory retry cannot run after that process crashes. The missing publication must remain discoverable independently of its lifetime.
Question 103
A booking coordinator uses one global lock for all customers. Correctness holds, but throughput collapses. What refinement can preserve needed exclusion?
- Acquire the global lock twice for every request.
- Remove all locking despite conflicting bookings.
- Acquire a separate lock for each worker regardless of the resource being modified.
- Scope concurrency control to the resource whose updates can conflict.
Explanation
Correct Answer
D:Independent resources can proceed concurrently while competing updates remain serialized or version-checked. The lock boundary should match the actual invariant.
Incorrect Answers
A:Additional locking does not improve concurrency and may introduce deadlock risks.
B:This restores concurrency by reintroducing the race the lock prevented.
C:Worker-specific locks do not exclude conflicting updates by other workers to the same resource.
Question 104
Two workers need locks on resources A and B. One acquires A first and the other B first. What policy reduces deadlock risk?
- Use a consistent acquisition order and bounded failure handling.
- Let each worker choose an order based on its name.
- Retry immediately while retaining each first lock, without coordinating acquisition order.
- Assign each worker a preferred resource order independently and require it to keep its first lock until its preferred second lock becomes available.
- Increase both lock timeouts but retain the opposing acquisition orders and held locks.
Explanation
Correct Answer
A:A shared lock order removes a common circular-wait pattern. Timeouts and cleanup still help recover from failures.
Incorrect Answers
B:Different ordering rules can preserve the circular-wait condition.
C:Immediate retry does not break a circular wait if both workers keep the resource the other needs. A consistent order or deliberate release policy is needed.
D:Independent orders preserve the A-then-B versus B-then-A cycle. Holding the first lock indefinitely prevents either worker from releasing the other's dependency.
E:A longer wait extends the same circular dependency. It does not remove the condition that lets each worker block the other.
Question 105
The agent's conversation says an order is unpaid, but the payment service records a completed payment. Which state should govern a new charge decision?
- Charge again and reconcile afterward.
- Average the two states into partially paid.
- Reconcile against the authoritative payment record before acting.
- Prefer the conversation because it was written by the model.
Explanation
Correct Answer
C:Conversation state can become stale. A charge decision must use the service's current transaction evidence to avoid billing twice.
Incorrect Answers
A:A second charge creates the harm that reconciliation should prevent.
B:Discrete transaction states cannot be meaningfully averaged this way.
D:Authorship does not make a remembered state authoritative.
Question 106
A fraud-check tool is unavailable. The agent interprets the absence of a fraud alert as a clean result. What distinction must the workflow preserve?
- The difference between a low-severity alert and a high-severity alert, while assuming every call completed.
- The difference between a cached clean result and a current clean result, while treating an outage as current evidence.
- The distinction between low risk and medium risk within completed checks only.
- The difference between a completed negative check and a check that never completed.
Explanation
Correct Answer
D:Unavailable evidence is not evidence of absence. The workflow needs a failure or escalation path rather than silently granting a clean status.
Incorrect Answers
A:Severity categories describe observed alerts. They do not represent the missing outcome of a check that was unavailable.
B:Freshness matters for completed checks, but an outage supplies no current result at all. Labeling that absence as clean still bypasses the failed dependency.
C:Those categories apply after a check produces evidence; they do not represent an unavailable check.
Question 107
One optional enrichment worker fails while required checks succeed. The product permits answers without enrichment. Which response is appropriate?
- Continue with the required result and accurately disclose the omitted enrichment where relevant.
- Invent enrichment data to keep the response shape unchanged.
- Mark all required checks as failed too.
- Fail every request whenever any optional worker fails.
Explanation
Correct Answer
A:Optional work should not automatically become a hidden hard dependency. The output must still distinguish missing enrichment from completed analysis.
Incorrect Answers
B:Fabrication conceals the worker failure and creates unsupported information.
C:An optional failure does not invalidate independently successful required checks.
D:This sacrifices permitted availability even when all mandatory conditions are met.
Question 108
A release workflow requires both security and compatibility checks. Security passes but compatibility fails. What should the coordinator do?
- Reuse the compatibility pass from a different build because the current security check passed.
- Treat the two required gates as alternatives and release when at least one passes.
- Record the compatibility failure as a post-release follow-up without an authorized change to the gate.
- Block release and resolve the failed required check.
Explanation
Correct Answer
D:A conjunction of prerequisites is satisfied only when every required condition passes. The coordinator must preserve that logic.
Incorrect Answers
A:A result for another build does not establish compatibility of this release candidate. The required gate must apply to the artifact being released.
B:The requirement is conjunctive: both checks must pass. Replacing AND with OR changes the acceptance policy.
C:Tracking the failure does not satisfy the pre-release condition. The coordinator cannot silently move a required check to after deployment.
Question 109
A worker labels a conclusion tentative, but the coordinator rewrites it as confirmed. What has been lost?
- Only the coordinator's preferred level of detail.
- The distinction between a hypothesis and an established result.
- Only the worker's preferred writing style.
- Only the ordering of the worker's findings in the final report.
Explanation
Correct Answer
B:Synthesis must preserve material uncertainty. Removing it can make downstream decisions appear better supported than the evidence allows.
Incorrect Answers
A:Removing a tentative qualification changes the claim's evidential status, even if the summary is otherwise an appropriate length.
C:The qualification describes evidential status, not merely style.
D:Reordering may change emphasis, but the observed error is promotion of an unconfirmed conclusion into a fact.
Question 110
A worker returns malformed JSON where a downstream calculator expects numeric fields. What should happen before calculation?
- Extract every number with a regular expression and assign fields by appearance order.
- Pass the raw text and hope the calculator infers the intended values.
- Validate and repair or reject the worker output against the agreed contract.
- Replace every missing number with zero without a documented rule.
Explanation
Correct Answer
C:Downstream computation needs well-defined inputs. A schema gate prevents malformed data from becoming an apparently precise numerical result.
Incorrect Answers
A:Numeric appearance does not establish which contract field each value belongs to or whether required fields are present.
B:Implicit coercion can conceal missing or misidentified fields.
D:Zero is a substantive value and may distort the calculation.