INFRASTRUCTURE LEARNINGINFRASTRUCTURE LEARNING
インフラ大全インフラ大全IT INFRASTRUCTURE / LEARNING
Infrastructure learning guide
Find a lesson or lab/

Esc to close · Search terms stay on your device

LEARNING日本語

Topic lessonTopic: Bedrock inference, retrieval and operating decisions

Test Amazon Bedrock Guardrails coverage and false decisions

Guardrails evaluate policies on supported paths with appropriate configuration. Verify expected denials and preservation of legitimate use.

What you will learn

Use harmless synthetic inputs to verify the policy version and both kinds of decision.

Prerequisite

The cases use fictional inputs.

What to predict

Guardrails evaluate policies on supported paths with appropriate configuration. Verify expected denials and preservation of legitimate use.

From conditions to the decision
From conditions to the decision1: Inputs — Align conditions — Coverage. 2: Coverage — Compare outcomes — Decision. 3: Decision — Verify the decision — Correction123InputsCoverageDecisionCorrectionFrom conditions to the decision1: Inputs — Align conditions — Coverage. 2: Coverage — Compare outcomes — Decision. 3: Decision — Verify the decision — Correction123InputsCoverageDecisionCorrection
  1. Inputs → CoverageAlign conditions
  2. Coverage → DecisionCompare outcomes
  3. Decision → CorrectionVerify the decision
Inputs
Synthetic glossary request
Coverage
Fictional sensitive classification label
Decision
Policy coverage on a new path is unknown
Correction
A legitimate term was rejected

Compare the cases

Synthetic observations
CaseConditionsDecision
an ordinary requestSynthetic glossary requestVerify expected allowance
a covered policy caseFictional sensitive classification labelVerify expected rejection and its scope
a new invocation pathPolicy coverage on a new path is unknownInspect policy version and input-output coverage
a false rejectionA legitimate term was rejectedPreserve conditions for the policy owner

Acceptance and limits

Evidence applies to the tested configuration. Quality, permissions, approval and data classification require separate checks.

Fix the invocation path and version.

Test allowed and rejected fixtures.

Include false decisions in acceptance.

Check your understanding

Choose an option to see the result and explanation.

Which decision applies to an ordinary request?
Read the answer and reason

Answer: Verify expected allowance · an ordinary request: Synthetic glossary request → Verify expected allowance

Which decision applies to a covered policy case?
Read the answer and reason

Answer: Verify expected rejection and its scope · a covered policy case: Fictional sensitive classification label → Verify expected rejection and its scope

Which decision applies to a new invocation path?
Read the answer and reason

Answer: Inspect policy version and input-output coverage · a new invocation path: Policy coverage on a new path is unknown → Inspect policy version and input-output coverage

Check, then explain in your own words

  1. Fix the invocation path and version.
  2. Test allowed and rejected fixtures.
  3. Include false decisions in acceptance.
Check your understanding: Use harmless synthetic inputs to verify the policy version and both kinds of decision.

Evidence applies to the tested configuration. Quality, permissions, approval and data classification require separate checks.

Include false decisions in acceptance.

Sources and conditions checked

The official references used in this lesson are collected here with their scope, conditions and corresponding sections. Independent expert review and live AWS comparison have not been performed.