● Home Knowledge Action Journal Homework Labs Study Try Case Make Training
Encode Cards Study History Free Recall Quizzes Digests Tasks Conjectures Belief Gym Audio Timeline Knowledge Items People Templates Outcomes DMXs Situation Analysis Council Debate Parable Branch Learning Story Engine Content Engine Explainer Rehearsal Bench Import Kindle Notes

📐 Encode

Every card for one document, both halves open. Nothing here is graded, scheduled or due — tick the ones that are still abstract and queue an example for each.

26 cards · 2 with examples
Observability
Q Why is total latency an insufficient metric for streaming generative interfaces?
A It hides regressions in time-to-first-token and inter-token latency, which are the user-visible quantities.
learning observability latency streaming generative-ai metrics
Q Why are token counts per invocation considered a quality signal, not just a cost/usage metric?
A Rising input tokens signals retrieval/history drift; rising output tokens signals prompt changes.
review llm-observability token-metrics generative-ai
Q Which CloudWatch metrics does Bedrock publish automatically?
A Invocations, latency, input/output token counts, client errors, server errors, throttles
review ⧉ same passage aws bedrock cloudwatch observability
Q What is the AWS mechanism to capture metrics (invocations, latency, input/output) from Bedrock?
A CloudWatch metrics, published automatically by Bedrock
review ⧉ same passage aws bedrock observability cloudwatch
Q What does CloudWatch Application Signals provide beyond raw metrics?
A SLO tracking — monitors whether services meet defined service-level objectives.
review aws cloudwatch observability slo
Q Why should high-cardinality dimensions (e.g., per-user) use logs rather than metrics?
A Each unique dimension value creates a separate metric stream, and metrics are priced per stream.
review observability cloudwatch cost-optimization metrics
Q Why are invocation logs typically sent to S3 rather than CloudWatch Logs for long-term retention?
A S3 is far cheaper, supports lifecycle tiering, and enables Athena for aggregate analysis.
review aws observability logging s3 cloudwatch
Q In a generative system's logging architecture, what role does each destination play when CloudWatch Logs and S3 are used together?
A CloudWatch for short-term operational search; S3 for cheap long-term retention and analysis.
learning observability aws logging cloudwatch s3
Q Why is application logging of retrieved chunks important?
A Without it, you can't distinguish retrieval failure from generation ignoring good context.
review observability rag logging retrieval
Q What does Bedrock Knowledge Bases logging capture that model invocation logging does not?
A Per-document ingestion outcomes — how you find which documents failed to process.
review ⧉ same passage aws bedrock observability knowledge-bases
Q What is the recommended way to observe ingestion pipelines within Bedrock Knowledge Bases?
A Enable Bedrock Knowledge Bases logging (off by default); it reports per-document ingestion outcomes.
review ⧉ same passage aws-bedrock observability knowledge-bases ingestion
Q What does AWS X-Ray produce from traced requests?
A A service map and per-request traces with timing breakdowns.
relearning ⧉ same passage aws observability x-ray distributed-tracing
Q What is AWS X-Ray?
A Distributed tracing tool producing service maps and per-request traces with timing breakdowns.
learning ⧉ same passage aws observability x-ray tracing
Q When is AWS Distro for OpenTelemetry (ADOT) preferred over X-Ray native instrumentation?
A When the system spans AWS and non-AWS components, or the org has standardized on OpenTelemetry.
review aws observability opentelemetry tracing
Q What does AgentCore Observability deliver to CloudWatch?
A Agent traces: reasoning steps, tool invocations, model interactions, and session context.
learning aws observability agents agentcore
Q What happens if CloudWatch Transaction Search is never enabled in an agent observability setup?
A Agent traces are absent even though observability appears configured — a one-time account-level setting is missing.
review aws observability cloudwatch agents
Q When logging agent tool calls, what should be captured beyond the tool name?
A The arguments passed and the result returned
review observability agents logging
Q Why is per-step token consumption the key metric for understanding agent cost?
A Agent cost variance is high; per-step data reveals the distribution behind that variance.
review llm-agents observability cost-management token-usage
Q What CloudWatch metric tracks guardrail interventions in Bedrock?
A InvocationsIntervened
learning aws bedrock guardrails cloudwatch observability
Q What does enabling guardrail trace reveal beyond whether a request was blocked?
A Which specific policy rule matched and with what confidence
learning aws bedrock guardrails observability
Q What output artifacts does a CloudWatch Synthetics canary produce?
A CloudWatch metrics, screenshots, and HAR files
learning ⧉ same passage aws cloudwatch observability synthetics
Q What is CloudWatch Synthetics?
A Scripted probes (Node.js or Python) run on a schedule, producing CloudWatch metrics, screenshots, and HAR files.
review ⧉ same passage aws cloudwatch observability synthetic-monitoring
Q How long does CloudTrail Event History retain management events?
A 90 days
review aws cloudtrail observability audit
Q What is the only way to audit Bedrock model invocations at the principal level in AWS?
A CloudTrail data events (opt-in, priced per event)
review aws cloudtrail observability audit
Q How do you make CloudTrail events trigger near-real-time alerts?
A Pair CloudTrail with EventBridge to react quickly to specific events.
learning aws cloudtrail eventbridge observability
Q What is the purpose of composite CloudWatch alarms?
A Combining multiple conditions into one alarm to reduce noise.
review aws cloudwatch observability alerting