Case Notes
Draft · 12 min read
The 5 failure modes we find in almost every RAG audit
Retridge Engineering
RAG & evaluation practice
Draft · not yet published
This note is on the editorial calendar. The outline below shows the intended structure; the article layout, code styling, and diagram treatment are the same as published notes.
Planned outline
Stale duplicates outranking the current document
Chunk boundaries cutting through tables and lists
No abstention path when context is insufficient
Permission-blind retrieval over mixed-sensitivity corpora
Reranking configured once and never evaluated again
How each one shows up in a failure distribution
In the meantime, the published note on retrieval-versus-correctness covers the measurement approach these all build on.
RAG Engineering Notes
One note a month on retrieval quality, evaluation, and AI security. No marketing.
Related notes
Why your chatbot retrieves the right document but still answers wrong
AI Engineering
Prompt injection: the security hole most AI teams ship with
AI Security
Evaluation-first AI development: a practical guide
Evaluation
Seeing this pattern in your own system? The audit measures all five gates in two weeks.
Book a free consultation