Question 210
A contract reviewer must distinguish obligations from optional recommendations. Which example selection is strongest?
- Contrast mandatory and advisory terms in near-match clauses
- Mix contradictory labels for similar obligations
- Use only long samples without labels
- Provide samples from an unrelated image captioning task
Explanation
Correct Answer
A:Contrastive examples highlight the linguistic distinction that changes classification. Labels should explain why each clause belongs in its category.
Incorrect Answers
B:Inconsistent demonstrations create ambiguity and unreliable classifications. The labels must follow the defined criterion.
C:Unlabeled text does not show the desired decision. The model needs example outputs and category meanings.
D:Different modalities and goals do not teach contract obligation semantics. Examples should mirror the use case.
Question 211
A log analyzer detects errors but incorrectly marks informational warnings as incidents. What few-shot improvement would be useful?
- Replace all warning examples with unrelated stack traces
- Provide only catastrophic error examples
- Label warning and error cases with their severity cues
- Remove status fields from the training examples
- Teach the model to flag every log line to maximize recall
Explanation
Correct Answer
C:The samples should show discriminating signals such as status codes and operational impact. The model can then apply a consistent boundary.
Incorrect Answers
A:Unrelated examples do not clarify the warning-versus-error threshold. The classifier needs labeled near misses.
B:Extreme positives do not explain the warning threshold. Borderline negatives are essential.
D:Severity metadata is relevant evidence. Hiding it makes the classification harder.
E:Unconditional flags recreate the false-positive issue. The task needs discrimination.
Question 212
A prompt contains dozens of examples with inconsistent order and output keys. What simplification is most appropriate?
- Shuffle the labels on every request without updating outcomes
- Add another hundred loosely formatted examples
- Curate consistent examples with clear input/output fields
- Remove the task instruction and rely only on examples
Explanation
Correct Answer
C:Well-structured examples reduce conflicting signals and unnecessary context use. Quality matters more than raw count.
Incorrect Answers
A:Broken associations teach the wrong mapping. Deterministic, correct labels are needed.
B:Volume amplifies inconsistency and prompt cost. It is unlikely to repair an unclear output contract.
D:Examples aid interpretation but should not replace explicit goals and acceptance criteria. Both serve complementary roles.
Question 213
A customer support prompt has sample dialogues containing real customer credentials. How should few-shot design be changed?
- Commit raw customer conversations to a public prompt repository
- Remove all compliance constraints from the example outputs
- Use synthetic or sanitized examples that preserve the behavioral pattern
- Include more production secrets so the model learns accurate syntax
Explanation
Correct Answer
C:The model needs examples of structure and decisions, not live secrets. Redaction limits exposure while retaining instructional value.
Incorrect Answers
A:Publishing private tickets violates data minimization. Examples should be safe to share with intended consumers.
B:Samples should reinforce, not contradict, privacy rules. Correct handling is part of the desired pattern.
D:Credential exposure is unnecessary and unsafe. A synthetic identifier can illustrate the same field format.
Question 214
A reviewer copies the precise prose of demonstration answers into every final report. What should the examples emphasize instead?
- Show the same decision structure with varied wording
- Examples containing no outcomes or explanations
- A request to reproduce example text exactly regardless of case
- One highly stylized sentence repeated in every example
Explanation
Correct Answer
A:Diverse demonstrations can teach format without inviting verbatim repetition. The evaluator should check substantive fit, not cloned phrasing.
Incorrect Answers
B:Unlabeled inputs cannot show which behavior is expected. A useful demonstration needs a target output.
C:Case-specific facts require new analysis. Literal reuse risks incorrect and repetitive outputs.
D:Repetition encourages copying the exact formula. It does not teach how to adapt to different evidence.
Question 215
A multilingual support system must retain the same JSON keys while extracting values from several languages. Which few-shot collection best fits?
- Translate every schema key into the source language
- Use only English examples and assume all language structures match
- Use diverse source-language examples with identical canonical output fields
- Omit the schema and ask for any convenient structure
Explanation
Correct Answer
C:The examples can demonstrate language variation while preserving the downstream schema. Key stability is a separate requirement from translated values.
Incorrect Answers
A:Dynamic keys break consumers expecting a fixed contract. Language variation should affect values, not field names.
B:Cross-language formats can differ materially. Representative multilingual cases improve robustness.
D:Without a consistent target, extraction cannot be consumed reliably by automation.
Question 216
A document pipeline needs a machine-readable invoice record rather than free-form prose. Which interface is best?
- Use structured output or a defined tool schema with typed fields
- Accept arbitrary fields and infer their meaning from label similarity
- Convert every invoice into a free-text narrative only
- Ask for approximately JSON-looking paragraphs without a schema
Explanation
Correct Answer
A:A schema makes expected keys and types explicit for downstream consumption. Application-level validation still checks content against source evidence.
Incorrect Answers
B:Unbounded output makes validation uncertain. A canonical field schema is preferable.
C:Narratives may omit required fields and vary in order. The downstream system expects structured data.
D:Informal formatting is brittle and difficult to parse. A typed contract better fits automation.
Question 217
A purchase order must include order_id and total, but discount_code is optional. How should the schema represent requiredness?
- Mark all three fields required with invented discount codes
- Treat all fields as unrestricted prose with no types
- Mark order_id optional despite it being essential
- Make two fields required; treat discounts as optional
Explanation
Correct Answer
D:Requiredness should match business rules and the selected structured-output mode. Validation must distinguish absent discounts from missing mandatory data.
Incorrect Answers
A:Fabricated optional values corrupt the record. A missing discount should have an explicit allowed representation.
B:Downstream consumers need predictable identifiers and numeric or decimal conventions. Loose text invites errors.
C:A record lacking its primary business identifier cannot meet the stated contract. The schema should reject it.
Question 218
An extractor must return one of approved statuses: open, pending or closed. What field definition reduces invalid status strings?
- Use a free-form comment field as the only status indicator
- Use an enumerated status field in the output schema
- Encode status as an unlabeled random number
- Ask the model to create a new status whenever uncertain
Explanation
Correct Answer
B:An enum constrains the allowed representation. Runtime checks can also ensure the status matches the source document.
Incorrect Answers
A:Arbitrary prose can produce many synonyms and break consumers. The contract needs canonical values.
C:Opaque numbers are not self-explanatory and may not map consistently to the approved values. Use a documented enum.
D:Invented categories violate the allowed set. Uncertainty should be handled through an explicit error or review path.
Question 219
A structured extractor must return a list of line items with quantity and unit price for each item. Which schema design is appropriate?
- Define an array of objects with explicit quantity and price fields
- Use one top-level scalar for the entire invoice
- Allow the model to change key names for each array element
- Flatten all lines into one unstructured string
Explanation
Correct Answer
A:Nested typed elements represent repeated structured rows. Validation can check item-level presence and numeric consistency.
Incorrect Answers
B:A scalar cannot represent multiple line-item records faithfully. The structure must support repetition.
C:Inconsistent element schemas break deterministic parsers. Each item should follow the same contract.
D:Free text is harder to validate and aggregate. The business requirement is per-item machine-readable fields.
Question 220
A validator rejects an output because a field expected to be a number contains a currency symbol. What contract clarification is most relevant?
- Delete the amount field and assume downstream software will guess it
- Define numeric values separately from currency codes or display formatting
- Change the entire record to arbitrary text to accept the symbol
- Treat invalid JSON types as automatically harmless
Explanation
Correct Answer
B:Machine-readable amounts should use a consistent representation. Presentation symbols are different from the value's type.
Incorrect Answers
A:An omitted required amount cannot support totals or payment checks. The value must be explicit.
C:Relaxing all typing loses validation benefits. Fix the offending field representation instead.
D:Type mismatches can produce runtime errors or incorrect calculations. The schema should be honored.