← Curriculum track ← Learn hub
Quanta GenAI Curriculum · LLMOps · Advanced

LLMOps Advanced — 055: pair eval regressions on `Eval harness hardening` — memo `68681 [55]`

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

Deep dive notebook

Synthetic drill artefacts

Token CFO scratchpad

- prompt_budget: 2039
- completion_budget: 680
- cache_key: `6c04`

Hypothesis: halving completions moves P95 ~8% — record actuals.

Practice

Practice Draft three eval assertions QA must pass before prompt promotion. — 55 Bump 12.