Lesson 050: Eval slices that compound
Focus
Anchor this drill to one production LLM workflow—even hypothetical. Token Eval slices that compound:50 keeps neighbouring lessons differentiable.
Key ideas
- Thread: Eval slices that compound · drill v10 · spin
157933. - Habit: attach a trace_id to every completion you would paste into an ops dashboard.
- Guardrail: add one RACI bullet for prompt or index changes before tomorrow's standup.
Deep dive notebook
Synthetic drill artefacts
Prompt scaffold
ROLE: LLMOps analyst cohort 50
INPUTS:
- excerpts tagged [chunk_id ...]
- policy_bundle_0
TASK:
1) Summarize deltas with citations
2) Confidence LOW|MED|HIGH + evidence
3) If facts missing emit MISSING_FACTS
USER_SEED >>> What changed between rollout 29 and 23?
Practice
Practice Simulate degraded retrieval once; capture user-facing fallback copy. — 50 Bump 34.