RAG Formation · Reference
Refusal, uncertainty, and hallucination control
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 → verifyReference sheet
Refusal, uncertainty, and hallucination control
The guided RAG Formation course content is complete; this reference is available for self-directed practice.
One-line rule
A fluent partial answer can still hallucinate if it adds an important claim the evidence does not support.
Core terms
| Term | Meaning |
|---|---|
| Refusal | A clear statement that the indexed evidence does not establish the requested fact. |
| Uncertainty | A calibrated explanation of what the evidence supports and what remains unknown. |
| Hallucination | An important claim presented without enough support from the allowed evidence. |
Evidence card
| 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 | Track refusal accuracy and hallucination rate separately, plus one answer that crossed the boundary. | One metric tells you everything about quality. |
| Safety | A failure and a regression case | A fluent answer proves the system is reliable. |
Local command
cd demo
python3 eval.py run --mode answer --workspace meridian_demo --category unanswerable --tag lesson-12-refusalArtifact: A refusal policy with test cases for no evidence, partial evidence, conflict, and ambiguity.
Continue learning: Open the matching standalone lesson · Course roadmap