RAG Formation · Reference

Prompts that answer only from evidence

A standalone cheat sheet for learning and testing RAG quality on this site.

$ reference --status
▸ course RAG Formation
▸ mode standalone reference
▸ status Complete
● learn → test → verify

Reference sheet

The guided RAG Formation course content is complete; this reference is available for self-directed practice.

One-line rule

A prompt cannot bring back a missing source; check retrieval and context before blaming generation.

Term	Meaning
Grounded generation	Generating an answer from the supplied evidence instead of the model’s general knowledge.
Instruction hierarchy	System and developer rules define what the model may do when the evidence is incomplete or conflicting.
Prompt contract	A testable agreement about answer format, refusal behavior, citations, and conflict handling.

Need	Record	Don't claim
Baseline	Command, configuration, observed result	The baseline is a target.
Change	One hypothesis or decision	You can tell what caused the result when several things changed.
Metric	For one failure, name the available evidence, unsafe behavior, proposed instruction, and evaluation category to rerun.	One metric tells you everything about quality.
Safety	A failure and a regression case	A fluent answer proves the system is reliable.

cd demo
python3 eval.py run --mode answer --workspace meridian_demo --limit 10 --tag lesson-10-prompt-baseline

Artifact: A versioned prompt rule covering evidence use, missing facts, citations, and refusal behavior.

Continue learning: Open the matching standalone lesson · Course roadmap