Enterprise knowledge is not only in database rows
Orders tell you what happened.
Policies tell you what should happen. Contracts explain obligations. Manuals explain how equipment should be used. Investigation reports explain why something failed.
Metrolane therefore models documents and knowledge as part of the enterprise — not as files sitting outside it.
A document is an enterprise object
A Metrolane document can have:
That means a policy is not only text.
It can also answer:
- Which version is current?
- Who may see it?
- What enterprise object does it apply to?
- Is it ready for semantic retrieval?
- Where did it come from?
Warranty Policy
Extended 36-Month Warranty Policy
This document has:
- a canonical Document identity
- a source representation
- a published version
- an enterprise relationship
- retrieval/index state
It is therefore useful for both normal document browsing and context-aware retrieval.
Documents can connect to the enterprise
Examples in Metrolane include relationships such as:
These links make retrieval more useful than searching a disconnected pile of text.
From document to retrieval
A simplified retrieval pipeline is:
The important idea is not the vector itself.
The important idea is:
Retrieve the right enterprise knowledge for the right reader and the right context.
Day-0 Semantic Search is not yet RAG
Day-0 Semantic Document Search returns:
It does not generate an LLM answer.
RAG appears later when an LLM uses retrieved enterprise knowledge as grounding.
Access comes before retrieval results
Suppose two people search for the same HR document.
The question is not only: Which chunk is most similar?
It is also: Is this reader allowed to know this document exists?
Same estate, different reader
Access changes the visible enterprise information universe.
It is not merely a cosmetic filter applied after retrieval.
Published does not mean searchable
The document can be:
- Exists
- ✓
- Published
- ✓
- Browseable
- ✓
- Semantic-search ready
- ✗
because its grounding/index state is still pending.
This gives the visitor an important trust rule:
Search must not silently use content that has not reached an approved retrieval-ready state.
Three independent states
A Document page needs to keep these separate.
Those states can legitimately differ.
Example:
is not contradictory.
It means:
The document exists and may be readable, but semantic retrieval is not currently allowed to use it.
What retrieval should know about context
A mature enterprise retrieval request may need more than text similarity.
Conceptually:
- Question
- ↓ Reader identity / persona
- ↓ Eligible documents
- ↓ Applicable enterprise context
- ↓ Semantic / lexical retrieval
- ↓ Relevant passages
If the question is about a warranty claim, a policy governing warranty behavior should rank differently from an unrelated HR procedure — even if some words overlap.
Bridge to EKA
At this point Metrolane has two major knowledge foundations:
Later, EKA can use both.
That is where retrieval becomes part of an LLM-grounding architecture.
Chapter takeaway
- Documents are enterprise objects, not loose files.
- Enterprise relationships make document context explicit.
- Retrieval is governed by access and applicability.
- Published does not automatically mean searchable.
- Semantic Search retrieves. EKA later reasons over retrieved knowledge.