RAG Anti-Patterns — Part 1: Ingest, Chunking, Embedding, Index & Retrieval Sins — 42 Templates & Checklists for Auditing Ingest, Chunking, Embedding, Index & Retrieval Failures in Production RAG

## It worked in the demo. By Wednesday it was quoting a discontinued tent pole as orderable.

Nothing had crashed. The logs were green. The model had not changed. The corpus had simply become real — ten times the documents, a hundred times the questions, revised policies with the exceptions at the end, and near-identical pages for similar products. **RAG Anti-Patterns — Part 1** is the working audit for the half of RAG failure that happens before generation is even asked a question: 42 copy-paste audit templates and checklists distilled from Chapters 1–6 of the dedicated anti-patterns book — learn the catalogue method (symptom, root cause, detection, lasting fix) and the two meta-sins that protect every other sin, fix the door at ingest, measure the cut at chunking, calibrate the lens at embedding, hold the index to a freshness contract, and measure the retrieval call itself.

Every card carries the book's tested positions built in: an anti-pattern is a practice that looks like engineering, demos well, and reliably produces a failure class; detection is a named signal with a threshold, never a hero reading answers; a prompt patch on a structural defect is containment with an expiry date, never a resolution; five copies of one source are one witness, so corroboration is counted by distinct sources; metadata is the control plane, and no contract means no index; a vector is meaningless without the model that produced it, so a model change is a new index and a full re-embedding; freshness is a written contract per source with a lag budget and a breach behaviour; top-k is a recall decision derived from recall@k curves per query type, never a tutorial default; and every chunk in the prompt must answer the question "what are you here to prove?" Where the book gives a number — its Jaccard near-duplicate threshold, its illustrative audit readings, its worked freshness SLA shape — the card carries the book's number and labels it as the book's or as illustrative, exactly as the book does. Bracketed fields are yours to fill; nothing here invents a threshold for your system.

### What's inside — 42 templates & checklists in 7 sections

**1. Why RAG Fails in Production — The Catalogue Method & the Two Meta-Sins (6)** — Catalogue Entry Card — Symptom, Root Cause, Detection, Lasting Fix · Demo-vs-Production Slice Audit · Broken-Promise Chain Worksheet · Anti-Pattern Triage Checklist — Bug, Limitation, Mistake, or Anti-Pattern · Meta-Sin Audit — the Cosmetic Patch & the Blame Relay · Configuration Smell Test & Silent-Failure Economics Sheet

**2. Garbage In, Gospel Out — The Ingest Sins (ING-1 to ING-4) (6)** — ING-1 The PDF Shredder — Parse-Quality Audit Card · ING-2 The Silent Truncation — Source-vs-Indexed Length Audit Card · ING-3 The Duplicate Flood — Near-Duplicate & Retrieved-Set Echo Audit Card · ING-4 The Metadata Bonfire — Per-Field Completeness Scan Card · Ingest Build Report & Release Gate · Ingest Scanner Run Sheet

**3. Confetti and Bricks — The Chunking Sins (CHK-1 to CHK-4) (6)** — CHK-1 The Confetti Cannon — Length Distribution & Completeness Audit Card · CHK-2 The Brick — Upper-Tail & Answer-Density Audit Card · CHK-3 The Orphaned Heading — Heading-Attachment & Table-Header Audit Card · CHK-4 The Overlap Echo — Intra-Index Duplicate & Unique-Content Audit Card · Chunking Audit Routine · Chunking Release Checklist

**4. The Wrong Lens — The Embedding Sins (EMB-1 to EMB-4) (6)** — EMB-1 The Brochure Model — Domain Probe Gate Card · EMB-2 The Silent Swap — Provenance & Canary-Pair Consistency Card · EMB-3 The Frozen Map — Drift, Re-Probing & Gold-Set Refresh Card · EMB-4 The Mirage of Coverage — Per-Language & Per-Register Slice Card · Mixed-Space Detector Run Sheet · Embedding Lifecycle Gate

**5. The Museum Index — The Index Sins (IDX-1 to IDX-4) (6)** — IDX-1 The Museum — Freshness Lag & Freshness Contract Card · IDX-2 The Sieve Backwards — Filter-Yield Telemetry Card · IDX-3 The One Big Bucket — Cross-Corpus Hit Rate & Namespace Audit Card · IDX-4 The Unversioned Overwrite — Versioning Audit & Time-to-Rollback Drill Card · Freshness Calculator & Weekly Four-Number Review Sheet · Index Operations Checklist

**6. Top-k Superstition — The Retrieval Sins (RET-1 to RET-4) (6)** — RET-1 The Magic Number — Recall@k Curve & Depth-Derivation Card · RET-2 No Second Look — Rerank-Delta & Rank-of-First-Relevant Card · RET-3 Context Stuffing — Context Utilisation & Quality-vs-Size Audit Card · RET-4 The Fusion Accident — Per-Signal Contribution & Known-Item Card · Retrieval Measurement Practice Sheet · Tuned-Pipeline Operating Checklist

**7. Part 1 Gates & Working Artefacts (6)** — Part 1 Anti-Pattern Register — All Twenty-Two Sins on One Board · Stage-Measurement Dashboard Specification · Bad-Answer Trace & Containment Protocol Card · Weekly Index + Retrieval Review Agenda · Rebuild & Cut-Over Runbook · Part 1 Production Gate — Chapters 1–6 Sign-Off — the book's six chapter checklists consolidated into one evidence-backed gate (35 lines)

### Who this is for

RAG platform and search engineers, ML engineers who own a retrieval pipeline, and technical leads asked to explain why a system that demoed beautifully misleads people in production. The source book assumes you already run — or are about to run — a retrieval-augmented system; no research background is required. If your team cannot say, from records, which model produced every vector in your index, start at Template 20.

### What this pack is not

- **Not Part 2.** Prompt sins, evaluation sins, security sins, cost sins, organizational sins, and the book's closing audit habit — Chapters 7–12, with Appendices A and B — live in the Part 2 companion pack; nothing here restates them, and this pack is complete and useful without them.

- **No runnable code or tooling.** The book's smell test, ingest scanner, chunking analyser, mixed-space detector, freshness calculator, and retrieval simulator are described as audit procedures and run sheets inside the cards — the pack ships the templates that specify, run, and record these audits, not a working scanner, analyser, or monitoring product.

- **No vendor or product guidance.** Parsers, embedding models, vector stores, and rerankers are treated as patterns and measurement problems, exactly as the book treats them. No product is selected, installed, or configured here.

- **No benchmark or results claims.** The book's deployment stories are composite teaching examples, labeled as the book labels them, and its audit figures are illustrative readings, labeled illustrative. This pack promises a method and a set of instruments — never a measured improvement in your system's accuracy.

- **No guarantee of correctness.** The book's own position: a catalogue converts endurance into engineering — it names, measures, and schedules failure classes; it does not eliminate them. These cards manage and expose risk; they never promise a failure-free system.

- **Not the architecture build pack.** The companion RAG Architecture Templates & Checklists pack (Product #2) covers how to build the pipeline — ingest, chunking, embedding, retrieval design decisions. This pack assumes that pipeline exists and audits how each of its reasonable decisions composes into a production failure. The two packs do not restate each other.

### Companion products

The **RAG Architecture Templates & Checklists** pack (Product #2) is the build-side companion — the architecture this pack deliberately audits rather than restates. **RAG Anti-Patterns — Part 2: Prompt, Evaluation, Security, Cost & Organizational Sins** is this pack's sibling, built from Chapters 7–12 and Appendices A and B of the same book: a companion for when you are ready for the second half — never a requirement for this pack to work. This pack stands alone.

*Note: the $99 Architecture Templates Bundle currently contains Products #1–#3 only. This pack launches standalone; no bundle inclusion is promised for this product.*

Built by the author of **RAG Anti-Patterns**, from the failure catalogue behind production retrieval systems.

### About the author

Naveen Kumar Josyula is the author of 35 AI books on Amazon KDP and the CEO and founder of AISoltrics Technologies Inc. He is the author of the source book, *RAG Anti-Patterns*, from which every template in this pack is distilled.

### Support promise

If anything in this file is wrong or unclear, email support@aisoltrics.com before you judge it — we fix verified issues within one business day and send you the corrected file free. (You can also reach the founder's office at aadhya@aisoltrics.com.) Updates to this product are free to past buyers via your library. Sold by AISoltrics Technologies Inc.