Driftless field notes
Practical guides for LLM workflows that move.
Model deprecations, changing labels, cheaper inference, RAG, and agents. Each guide shows how to keep the model, prompt, and eval data in sync with a reproducible example.
Your ticket classifier’s model got deprecated
Read guide →
Support changed the labeling policy
Read guide →
Prompt repair that doesn’t wait on memory
Read guide →
Finance wants cheaper inference, same bar
Read guide →
Offline F1 is lying — labels conflict
Read guide →
Free-form answers graded by an LLM judge
Read guide →
RAG QA: new answer model, same knowledge base
Read guide →
Tool-calling agent: new planner, same tools
Read guide →