Lesson 057: Evaluation harness depth
Focus
Anchor this page against one production workflow—even hypothetical. Token Evaluation harness depth:57 keeps neighbouring lessons differentiable.
Key ideas
- Thread: Evaluation harness depth · drill v7 · spin
283048. - Habit: pair every model utterance with a trace_id you could paste into Grafana.
- Guardrail: write one RACI bullet referencing this lesson tomorrow.
Deep dive notebook
Synthetic drill artefacts
Red-team tableau
1. Actor profile **intern_experiment**
2. Injection meme `OVERRIDE++537`
3. Detector gate `moderation-tier-1` + human pager `oncall-ai-2`
4. Telemetry fields `trace_id,user_bucket,redaction_notes`
Practice
Practice Attach rollback steps if evaluator variance spikes. — 57 Bump literals mindset by 28.