Question 380
A triage model yields different confidence scores for nearly identical customer complaints, but staff treat each score as a deterministic fact. An oversight committee wants the cause and mitigation recorded. Which observation is the best evidence that the correction works?
Explanation
Correct Answer
The key evidence is the observed result: comparable complaints have calibrated uncertainty ranges and appropriately reviewed decisions. It is relevant because probabilistic predictions were mistaken for exact rules despite input sensitivity.
Incorrect Answers
“Cross-account access tests do not reveal records outside the authorized audience” could occur while the stated failure remains unresolved. It tests: Cross-account access tests do not reveal records outside the authorized audience. The required validation instead is “Comparable complaints have calibrated uncertainty ranges and appropriately reviewed decisions”, because probabilistic predictions were mistaken for exact rules despite input sensitivity.
This checks a different technical dependency: Testing demonstrates acceptable outcomes for relevant accent cohorts. The case requires proof of “Comparable complaints have calibrated uncertainty ranges and appropriately reviewed decisions”, which establishes the relevant condition: Probabilistic predictions were mistaken for exact rules despite input sensitivity.
That observation would verify a control associated with: Adversarial cases trigger safe handling rather than unreviewed harmful instructions. It does not establish the needed acceptance outcome. The direct evidence is “Comparable complaints have calibrated uncertainty ranges and appropriately reviewed decisions”, testing: Probabilistic predictions were mistaken for exact rules despite input sensitivity.