Practical notes: RAG: Mistakes that Quietly Kill Retrieval Quality

Use this as an operator-facing rebuild of the ideas in “RAG: Mistakes that Quietly Kill Retrieval Quality”: clear stages, ordered code slots, and recovery notes that survive a handoff. The Overview stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments.

Why retrieval, not generation, is where RAG breaks

For the Why retrieval not generation stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap.

Core concepts, briefly

For the Core concepts briefly stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap.

Mistakes 1 to 4: chunking errors

For the Mistakes 1 to 4 stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap.

Mistake 1: defaulting to naive fixed-size splitting without testing alternatives

For the Mistake 1 defaulting to stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap.

from langchain_text_splitters import RecursiveCharacterTextSplitter

# Recursive splitting respects structure first, falls back to
# character boundaries only when a section is still too large.
splitter = RecursiveCharacterTextSplitter(
    chunk_size=512,       # tokens, not characters, if you swap in a token-aware splitter
    chunk_overlap=75,     # ~15% overlap, a reasonable starting point
    separators=["\n## ", "\n### ", "\n\n", "\n", ". ", " "],
)
chunks = splitter.split_text(document_text)

Mistake 2: picking a chunk size without regard to document type

For the Mistake 2 picking a stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap.

Mistake 3: ignoring or mishandling chunk overlap

For the Mistake 3 ignoring or stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap.

Mistake 4: ignoring the document’s own structure

For the Mistake 4 ignoring the stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap.

Mistakes 5 to 7: retrieval architecture errors

For the Mistakes 5 to 7 stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap.

Mistake 5: shipping vector-only search and never adding BM25

For the Mistake 5 shipping vector-only stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap.

from langchain.retrievers import EnsembleRetriever
from langchain_community.retrievers import BM25Retriever

bm25_retriever = BM25Retriever.from_texts(chunk_texts)
bm25_retriever.k = 20

vector_retriever = vectorstore.as_retriever(search_kwargs={"k": 20})

# weights are a starting point for RRF-style blending; tune against your eval set
hybrid_retriever = EnsembleRetriever(
    retrievers=[bm25_retriever, vector_retriever],
    weights=[0.4, 0.6],
)
results = hybrid_retriever.invoke(user_query)

Mistake 6: fusing results with a naive weighted average instead of RRF

For the Mistake 6 fusing results stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap.

Mistake 7: retrieving too many, or too few, candidates for the stage you’re in

For the Mistake 7 retrieving too stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap.

Mistakes 8 and 9: reranking and embedding model mistakes

For the Mistakes 8 and 9 stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Prefer structured outputs with schema validation over free-form prose when the next step is code or a tool call.

Mistake 8: skipping reranking entirely

For the Mistake 8 skipping reranking stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap. For the Mistake 8 skipping reranking stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments.

from FlagEmbedding import FlagReranker

reranker = FlagReranker("BAAI/bge-reranker-v2-m3", use_fp16=True)

def rerank(query: str, candidates: list[str], top_k: int = 5) -> list[str]:
    pairs = [[query, c] for c in candidates]
    scores = reranker.compute_score(pairs, normalize=True)
    ranked = sorted(zip(candidates, scores), key=lambda x: x[1], reverse=True)
    return [text for text, _ in ranked[:top_k]]

final_context = rerank(user_query, fused_candidates, top_k=5)

Mistake 9: using a general-purpose embedding model for a specialized domain

When working through the Mistake 9 using a stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph. Cache stable system instructions and tool schemas. Re-sending identical preamble is a common source of burn.

Mistake 10: skipping evaluation and shipping on vibes

When working through the Mistake 10 skipping evaluation stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Measure recall on a fixed question set before tuning prompts. Prompt churn rarely fixes a weak retrieval surface.

from deepeval import assert_test
from deepeval.metrics import FaithfulnessMetric, ContextualPrecisionMetric
from deepeval.test_case import LLMTestCase

def test_refund_policy_query():
    test_case = LLMTestCase(
        input="What is the refund window for the Pro plan?",
        actual_output=rag_pipeline.run("What is the refund window for the Pro plan?"),
        retrieval_context=rag_pipeline.last_retrieved_chunks,
        expected_output="Pro plan purchases can be refunded within 14 days of purchase.",
    )
    faithfulness = FaithfulnessMetric(threshold=0.8)
    precision = ContextualPrecisionMetric(threshold=0.7)
    assert_test(test_case, [faithfulness, precision])

Real-world applications

When working through the Real-world applications stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Measure recall on a fixed question set before tuning prompts. Prompt churn rarely fixes a weak retrieval surface. When working through the Real-world applications stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments.

Comparison: retrieval architecture choices

The Comparison retrieval architecture choices stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph. Separate chunking policy from retrieval policy. Changing one should not force a rewrite of the other when quality metrics move.

Which option should you choose?

The Which option should you stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Separate chunking policy from retrieval policy. Changing one should not force a rewrite of the other when quality metrics move.

Benefits and limitations of a well-built RAG pipeline

The Benefits and limitations of stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Separate chunking policy from retrieval policy. Changing one should not force a rewrite of the other when quality metrics move. The Benefits and limitations of stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments.

Production readiness checklist

For the Production readiness checklist stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap.

Conclusion

For the Conclusion stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap.

Useful resources

For the Useful resources stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap. For the Useful resources stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments.

Operational checklist

When working through the Operational checklist stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest.

Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion.

Measure recall on a fixed question set before tuning prompts. Prompt churn rarely fixes a weak retrieval surface.

Freeze a golden set before changing prompts or models. Moving both the system and the yardstick hides regressions.

Add a smoke test that exercises the critical path in CI with fixtures, not live paid APIs, whenever budgets allow.

Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish.

Before promoting the stack, freeze versions, capture a golden transcript for the critical path, and confirm rollback steps. Shared environments need rate limits, tenancy checks, and a clear owner for secret rotation. Prefer boring reliability over clever one-off demos.

Batch note for f9fa4aa0641e: keep provider keys out of the repo, set a per-session token ceiling, and store transcripts next to the eval fixtures so later model swaps stay comparable.

The hardening note 0 stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish.

Hardening detail 0/908: measure wall time, error class, and token spend for this note, then decide whether to keep the change based on a fixed question set rather than anecdote.

For the hardening note 1 stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion.

Hardening detail 1/908: measure wall time, error class, and token spend for this note, then decide whether to keep the change based on a fixed question set rather than anecdote.

When working through the hardening note 2 stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph.

Hardening detail 2/908: measure wall time, error class, and token spend for this note, then decide whether to keep the change based on a fixed question set rather than anecdote.

The hardening note 3 stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline.

Hardening detail 3/908: measure wall time, error class, and token spend for this note, then decide whether to keep the change based on a fixed question set rather than anecdote.

For the hardening note 4 stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments.

Hardening detail 4/908: measure wall time, error class, and token spend for this note, then decide whether to keep the change based on a fixed question set rather than anecdote.

When working through the hardening note 5 stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish.

Hardening detail 5/908: measure wall time, error class, and token spend for this note, then decide whether to keep the change based on a fixed question set rather than anecdote.