Lesson 047: Eval slices that compound
Focus
Bias toward observable metrics: latency, cost, escalation rate. Token Eval slices that compound:47 keeps neighbouring lessons differentiable.
Key ideas
- Thread: Eval slices that compound · drill v7 · spin
63562. - 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
Logging contract
| Field | Retention |
|-------|-----------|
| trace_id | 12d |
| prompt_hash | policy-defined |
| completion_excerpt | 24h hot |
Mask tier `pii-v2` before persistence.
Practice
Practice Pair with security on one prompt-injection tabletop scenario. — 47 Bump 9.