M2.11 module quiz
Ten questions. 80% to pass (8 of 10). Open book, unlimited retakes.
Answers are collapsed under each question. Answer first, then check.
Q1
The Bedrock rule family reads:
A. The control plane: what Bedrock objects exist and whether they are invoked
B. Your own prompts, their token counts, and the cost of each one of them
C. Application logs emitted by the calling service
D. The quality of the model’s output responses
Show answer
Correct: A. Which is why the family clusters into provisioned-throughput economics and orphaned artifacts, and why model selection and batching have no rule behind them.
Q2
RC-1601 (PT Idle) treats an absent Invocations series as the idle signal because:
A. CloudWatch always returns a series, even an entirely empty one, for any registered metric
B. Absence of data is never treated as a signal anywhere in the rule catalog
C. The discoverer deletes stored metric series for resources it considers idle
D. Invocations is a Sum-statistic count metric, so a series exists only when invocations are above zero
Show answer
Correct: D. The rule still hard-abstains when the org’s whole metrics map is empty, because a fetch outage would otherwise read as “every PT is idle” and mass-fire delete recommendations.
Q3
RC-1601’s savings are:
A. A fixed fraction of current cost
B. Half the monthly cost
C. The full recovered monthly cost
D. Unknown
Show answer
Correct: C. A no-commitment PT is freely deletable, so the whole bill is recoverable. It abstains entirely on any committed term, because a committed PT bills to term-end regardless.
Q4
RC-1607 (Agent Idle) reports zero findings on an org with nine agents because:
A. All nine of the agents are genuinely being invoked in production today
B. Bedrock agents are not discovered by the inventory pass
C. The rule was retired and no longer emits findings
D. A producer-contract gap means the metric query can never return a datapoint
Show answer
Correct: D. The query asks for AgentId in the AWS/Bedrock namespace; Bedrock Agent metrics live in AWS/Bedrock/Agents as InvocationCount dimensioned by alias ARN. The rule abstains rather than declaring every agent idle.
Q5
Four Bedrock rules carry the advisory category and $0 savings because:
A. They are unfinished features and are still sitting behind an internal feature flag today
B. The artifacts themselves are free; the spend lives in the foundation-model invocations underneath
C. They are low severity and filtered out
D. AWS publishes no pricing for those objects
Show answer
Correct: B. Filing a $0 finding under a cost-driven category would get it dropped by the low-savings suppressor, so it would never reach the customer at all.
Q6
RC-1610 (PT on Older Model Generation) prices savings as:
A. A flat 30% assumption applied across every single model family
B. It abstains always, since token mix is never observable
C. The full monthly cost of the deprecated deployment
D. The real rate delta between the deprecated model and its successor
Show answer
Correct: D. Both rates are looked up directly and the rule abstains when either is missing, when the successor is not cheaper, or when the PT is on a committed term.
Q7
RC-1603 (PT Below Break-Even) currently:
A. Fires with a concrete saving figure drawn from the SKU rate delta
B. Detects correctly but abstains, for lack of per-token on-demand rates
C. Is retired and no longer part of the catalog
D. Fires only on Anthropic-hosted model families
Show answer
Correct: B. The PT-side rate is ingested; the on-demand token side is not, so the break-even cannot be computed. It emits nothing rather than a $0 recommendation.
Q8
Model selection and prompt caching are optimized by:
A. No rule; they need your token mix, which never reaches the control plane
B. RC-1603 and RC-1606, which read the invocation mix
C. The Bedrock rule family generally, across all of its many levers today
D. The complexity router, which observes the token mix
Show answer
Correct: A. They are real levers and frequently the largest ones, and they must be sized from your own invocation logs. Assuming a rule covers them is the most common misreading of this module.
Q9
The RC-16xx range is split as:
A. All Bedrock rules in one block
B. Bedrock 1601..1620, SageMaker 1621..1634
C. Bedrock RC-1601..1610 plus RC-1615 and RC-1630..1634; SageMaker RC-1611..1614 and RC-1616..1628
D. Alternating by resource type across the whole of the reserved Bedrock rule-id range in the catalog
Show answer
Correct: C. Sixteen Bedrock and seventeen SageMaker, thirty-three in total. RC-1629 does not exist.
Q10
Batch inference suits:
A. All inference workloads, whatever the latency need
B. Non-real-time work such as scoring and classification
C. Real-time chat, where responses stream to a user
D. Only Anthropic models, which support the mode
Show answer
Correct: B. The trade is latency for rate. Applying it to a real-time path converts a cost win into a user-facing regression.
What’s next
Back to Bedrock + ML cost.