/ THE SHORT ANSWER
Keep authoritative systems and documents as the source of truth; treat the retrieval index as a derived, rebuildable layer. Every chunk should retain source identity, version, owner, permissions, effective date, and deletion path. Retrieval should respect user access before generation, answers should cite the underlying evidence, and updates must propagate predictably. The team needs an owner for sources, retrieval quality, and answer policy.
- 01Keep source systems authoritative and the index rebuildable.
- 02Carry provenance and permissions through ingestion and retrieval.
- 03Design update and deletion verification before launch.
/ dotSuper point of view
The index is not the knowledge. A trustworthy architecture preserves provenance and control from the answer back to the maintained source.
What the evidence says
NIST’s Generative AI Profile highlights information integrity, privacy, security, and confabulation risks relevant to knowledge-grounded systems.
OWASP’s guidance identifies sensitive-information disclosure and vector or embedding weaknesses among important LLM-application risks.
A practical decision framework
The following framework is dotSuper’s operating synthesis of the cited guidance. It is designed to make the decision inspectable, not to imitate a platform ranking formula, certification checklist, or legal test.
- Source layer: authoritative record, owner, version, policy, and lifecycle.
- Ingestion layer: parsing, chunking, metadata, permission, validation, and index status.
- Retrieval layer: identity-aware filtering, relevance, diversity, conflict, and freshness.
- Answer layer: evidence use, citation, refusal, escalation, feedback, and audit.
| Step | Decision to record |
|---|---|
| 01 | Source layer: authoritative record, owner, version, policy, and lifecycle. |
| 02 | Ingestion layer: parsing, chunking, metadata, permission, validation, and index status. |
| 03 | Retrieval layer: identity-aware filtering, relevance, diversity, conflict, and freshness. |
| 04 | Answer layer: evidence use, citation, refusal, escalation, feedback, and audit. |
How to put it into practice
Choose one document family and trace a source update, permission change, and deletion through every layer. Require an observable completion state and a reconciliation report.
Store generated summaries as derived artefacts with their source versions. Do not allow a summary to silently become the authority for future answers.
- Name the accountable owner and the decision this work must enable.
- Record the current evidence, assumptions, exclusions, and next review trigger.
- Measure a useful outcome rather than treating publication or deployment as success.
What this page cannot conclude
- 01Architecture requirements vary with scale, data types, permissions, latency, and regulatory obligations.
- 02Citations improve traceability but do not prove that an answer correctly interprets the source.
- 03Publication, technical eligibility, or good practice cannot guarantee ranking, referral traffic, citation, adoption, or a business outcome.
Sources
Make improvement a maintained operating rhythm.
The Optimisation Subscription keeps evaluation, governance, content, workflows, and product improvements moving as small accountable projects.
Explore the subscription