Exam-style
An internal audit emails a team running a document assistant on Amazon Bedrock and a risk-scoring model trained on Amazon SageMaker AI. Which passage did one sentence of the assistant's answer come from? Which ingestion run loaded that passage, and what did it change? Which datasets is the corpus built from, and who owns each? How was the deployed model evaluated, and who approved it? Which four artefacts answer those, in order?
Reveal the answer
B. A source citation, a data lineage record, a data catalogue entry, and an Amazon SageMaker Model Card
Sort the four questions by the size of the thing asked about, and each lands on one artefact. One answer gets a source citation: what an Amazon Bedrock knowledge base returns alongside a generated response, carrying the cited text and its source location. One item’s journey gets data lineage, the record of where the data came from and what each step did to it. The inventory gets data cataloging, on AWS an AWS Glue Data Catalog entry holding schema and location, with owner and licence added by people and access granted through AWS Lake Formation. The model gets an Amazon SageMaker Model Card, versioned on every edit, recording intended use, risk rating, training details, evaluation results and an approval status. Swapping the middle two fails: a catalogue entry describes a dataset as a whole, not the path one record took through it. CloudTrail records which API was called and by whom, not which passage a sentence came from, and a citation names the passage but not which version was loaded. One card covers none of the first three.
Q. One audit email asks which passage a sentence came from, which run loaded that passage, which datasets the corpus is built from and who owns them, and how the deployed model was evaluated and approved. Which four artefacts?
A. A source citation, a data lineage record, a data catalogue entry, and an Amazon SageMaker Model Card, in that order.
Why? The discriminator is the scope of the thing being described. One answer gets a source citation, returned by an Amazon Bedrock knowledge base alongside the generated response, carrying the cited text and the location it was retrieved from. One data item’s journey gets data lineage, the record of where that item came from and what each step did to it. The inventory of datasets gets the catalogue. On AWS that is an AWS Glue Data Catalog entry with owners and licence terms filled in, access governed through AWS Lake Formation. The model gets an Amazon SageMaker Model Card, carrying intended use, risk rating, training details, evaluation results and an approval status. Four scopes, four artefacts, and each one stops where the next one starts. The trap is reaching for the Model Card every time, because it reads as the most formal document in the set. Three of the four questions are not about a model at all.