Lesson 055: Eval harness hardening
Focus
Bias toward observable metrics: latency, cost, escalation rate. Token Eval harness hardening:55 keeps neighbouring lessons differentiable.
Key ideas
- Thread: Eval harness hardening · drill v5 · spin
322231. - 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
Retrieval partition plan
| Slice | Tokens | Mode | Notes |
|-------|--------|------|-------|
| FAQs | 426 | hybrid@0.54 | preserve tables |
| Policies | 475 | dense@0.67 | ACL metadata required |
Drill: justify chunk boundaries for lesson 55.
Practice
Practice Simulate degraded retrieval once; capture user-facing fallback copy. — 55 Bump 9.