Knowledge base

Compass

The organised body of material a system answers from, with an owner and a review cycle rather than a folder.

What is a knowledge base?

A knowledge base is the organised body of material a system is permitted to answer from. Each source in it has a named owner and a stated review cycle. That ownership is what separates a knowledge base from a shared drive that happens to be indexed.

Also called: Knowledge corpus

Why a knowledge base needs an owner before it needs an index

The phrase is routinely used for whatever happens to be indexed, which is where most retrieval deployments begin to fail. A file share holds superseded drafts alongside the current policy, and nothing in the material itself says which is which. A knowledge base is the decision an organisation has taken about which of those documents it will stand behind. That decision has to be recorded per source, with a person answerable for keeping it current.

  • Ownership makes it possible to ask someone why a document is still current.
  • Scope decides which questions the system should decline on principle.
  • A review cycle catches a superseded policy before an employee acts on it.
  • Material nobody owns accumulates until retrieval quality degrades for every question.

How a knowledge base works

  • Candidate sources are inventoried and each is assigned a named owner.
  • Scope is agreed, naming the subjects the corpus is authoritative for.
  • Each source is classified by sensitivity and by the entitlements governing access.
  • Documents are ingested with their version and their effective date retained.
  • A review cadence is set per source, proportionate to how fast it changes.
  • Retired documents are withdrawn from the index rather than left to compete with current ones.

How a knowledge base is evaluated

A knowledge base is assessed on whether its material is current and accounted for. Examine the corpus itself rather than a sample prepared for a demonstration:

  • share of sources with a named and contactable owner
  • proportion of documents past their stated review date
  • count of superseded versions still reachable by retrieval
  • coverage of the question topics users actually submit
  • age distribution of material by source

Limitations, risks and common failure modes

  • A knowledge base is inventoried once at launch and never reconciled again.
  • Ownership is assigned to a team mailbox that nobody reads.
  • Superseded drafts remain indexed alongside the approved version of the same policy.
  • Scope is defined so broadly that the system answers questions it has no material for.

Related terms

Subject matter expert review loopThe named people who check answers in a defined cycle, and the record of what they corrected.TaxonomyThe agreed structure of categories a body of material is organised by, which retrieval leans on more than most teams expect.Enterprise searchFinding material across an organisation's systems, the category an answering layer is measured against.Refusal and abstentionThe system declining to answer when the evidence is not there, which is a feature rather than a shortfall.Retrieval augmented generation (RAG)Answering from documents retrieved at the moment of the question, so the answer reflects your material rather than the model's memory.Semantic searchFinding material by what it means rather than by the words it happens to contain.

All 315 terms