Grounded answers

When does external evidence improve the answer enough to justify a retrieval system?

Decide when retrieval is justified, approve its source versions, test eligible passages and updates, then route insufficient evidence and verify answer claims.

AI VISTA / DECISION01 / 03

WORKING DECISIONRetrieval earns its cost when knowledge changes, access matters, and important claims must be traceable.

01INPUT

02CHECK

03BOUNDARY

RETURN TO THE REAL TASK✓

DECISION BRANCH

Answer the decision this topic actually owns.

ASK FIRST↓
Does the answer depend on changing facts, versions, access rights, or traceable support for important claims?
IF↓
Those conditions apply: approve sources before testing retrieval hits, citation support, and no-evidence handling.
OTHERWISE✓
Use reviewed fixed content or consult the authoritative source directly instead of maintaining RAG for a few stable facts.

WORKED CASE

Which policy version should a multi-region leave assistant cite?

Employees encounter a group handbook, regional supplements, and a newly issued leave notice. Keywords overlap while region, effective date, and access rights differ.

WORKING ARTIFACTA retrieval trial and citation audit checking whether eligible passages are found and whether citations support the claims.

01WEAK SHORTCUT

Put every PDF in a vector database and call the answer grounded whenever it includes a link.

02BETTER JUDGMENT

Establish authoritative sources and coverage before choosing retrieval. Results carry region, version, effective date, and permission, while important claims map to passages that actually support them.

03ACCEPTANCE

Test known answers, expired policy, cross-region conflicts, and unauthorized questions together. Claims are supported, superseded material does not win, and restricted content does not leak.

ENTRY & EXIT SIGNALS

Know when to enter—and when the decision is good enough to leave.

A topic is not an endless knowledge directory. Entry signals tell you whether the problem belongs at this layer. Exit signals decide whether to continue instead of substituting time spent reading for work completed.

AI VISTA / DECISION STATUSGrounded answers

01ENTER HERE

Consequential answers need citations or traceable support
Sources conflict by region, version, or permission
A RAG build is proposed without maintenance ownership

02LEAVE WHEN

Authoritative sources, coverage, and owners are registered
Important claims are actually supported by current passages
Expired, conflicting, and unauthorized questions stop safely

11 PRACTICES

Build the judgment, make the artifact, then review it with real cases.

01 Prompting, RAG, or fine-tuning? Choose by failure modeDiagnose the failure first, then compare instructions, request-time retrieval, and fine-tuning on the same holdout cases and operating gates. Grounded answers 02 Decide whether RAG is worth buildingCompare knowledge volatility, source volume, citation needs, access rules, and maintenance cost before adding retrieval. Grounded answers 03 How do you approve sources before adding them to a RAG knowledge base?Decide which exact documents and revisions may ground a RAG answer by checking authority, ownership, use rights, audience, effective dates, and conflicts before indexing. Grounded answers 04 How do you validate PDF, table, and image extraction before RAG indexing?Compare parsed content with original pages and cells before chunking. Check OCR, reading order, table relationships, figures, and source locations. Grounded answers 05 How should you chunk RAG documents? Keep the full evidence intactCompare fixed windows, overlap, and document structure on clauses and tables, then choose chunk boundaries using complete evidence, citations, and cost. Grounded answers 06 Keyword, vector, hybrid, or reranking for RAG? Choose with a matched trialCompare BM25, dense vectors, hybrid fusion, and reranking on exact IDs, paraphrases, and mixed questions while holding evidence, access, context, and latency gates fixed. Grounded answers 07 Evaluate RAG retrieval before judging the answerBuild a small retrieval trial that checks gold passages, no-answer questions, document versions, ranking, and access boundaries before generation. Grounded answers 08 Document-level access control for enterprise RAG, from search to citationCarry verified identity, tenant scope, and document ACLs into every chunk, then test denial across retrieval, model context, citations, cache, and revocation. Grounded answers 09 How do you verify that a RAG document update or withdrawal reached the index?Trace a source revision or withdrawal through ingestion, chunk and vector cleanup, search visibility, caches, and answer probes before calling a RAG knowledge base current. Grounded answers 10 When RAG lacks evidence, should it answer, clarify, or hand off?Route a RAG request to a supported answer, a specific clarifying question, an honest unable-to-verify response, or human review when retrieved evidence is incomplete. Grounded answers 11 Test citations for support and freshnessAudit whether each important claim is supported by the cited passage, uses the right version, and survives a freshness check. Grounded answers

BOUNDARY

What this topic page will not do

A grounded answer can still be stale, conflicted, or misread. Source governance, citation checks, and human escalation remain separate controls.

01Do not treat vector retrieval itself as an authoritative source02Do not use one link to hide a claim that the passage does not support03Do not connect sources to generation without enforceable access filtering

When source, version, permission, and no-answer behavior are registered, decide whether retrieval is worth maintaining.

Courses and topics open directly from the main navigation.

Courses