Evaluations and datasets

Inspect a data slice

Inspect a data slice through a bounded evidence-first AI operations workflow in the training lab.

Evaluations and datasets
8 min Core Lesson 84 of 180
cat /opt/ai-lab/evals/golden-set.jsonaiops evals results inspect-a-data-slicejq . /opt/ai-lab/evals/golden-set.json
Lesson 84 of 180 0/180 lessons 0/18 missions 0/11 briefings Evaluations and datasets · 8 min
0%
Continue View profile
AI operations terminal Training lab
New learnerrank 0XP 0%complete
learner@aiops:/home/learner $ AI operations lab: type a command, press Enter
Instructions 8 min

Click any instruction for the command details, the why, and the common mistake to avoid.

Inspect the evaluations and datasets baseline

Type this exactly: cat /opt/ai-lab/evals/golden-set.json

cat /opt/ai-lab/evals/golden-set.json
Run the evals review

Type this exactly: aiops evals results inspect-a-data-slice

aiops evals results inspect-a-data-slice
Confirm the evidence

Type this exactly: jq . /opt/ai-lab/evals/golden-set.json

jq . /opt/ai-lab/evals/golden-set.json
Lesson support

What to notice while you play.

Objective

Use commands and observable output to explain inspect a data slice without changing a real model or service.

Hint

Start with cat /opt/ai-lab/evals/golden-set.json. Then run aiops evals results inspect-a-data-slice before collecting the final evidence.

Why it matters

Inspect a data slice is an operator skill because AI behavior must be connected to versioned configuration, runtime state, and inspectable evidence.

Common mistakes
  • Skipping the baseline fixture before reasoning about inspect a data slice.
  • Treating one simulated output as proof of root cause instead of one bounded piece of evidence.
Reference

Commands in this lesson.

aiops models

Inspect the simulated model catalog, manifests, and version comparisons.

aiops prompts

Review versioned prompt metadata and deterministic prompt diffs.

aiops evals

Run and inspect bounded evaluation fixtures without model execution.

aiops rag

Inspect retrieval documents, chunks, index status, and reviewed results.

aiops traces

Read deterministic request traces, summaries, and latency evidence.

aiops guardrails

Review simulated guardrail policies, checks, and audit summaries.

aiops cost

Estimate cost and capacity from fixed training metrics.

aiops incidents

Review incident timelines, evidence bundles, and operator notes.

jq

Query JSON fixtures with a small declarative path subset; no expression execution occurs.