> ## Documentation Index
> Fetch the complete documentation index at: https://docs.kviria.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Narrowing

> Progressive narrowing from a vague reference to exactly one verified place.

A user's reference is rarely precise. "The cafe near the bookstore" is
not a lookup - it is a narrowing process: several cafes match, the user
picks one, the pick narrows the context for the next turn.

Kviria is built around that process:

```
"I want a coffee"            -> candidates (which one?)
"the first one"              -> confirm   (is it Stable Cafe?)
"yes"                        -> resolved  (verified coordinate + evidence)
```

Each response is one of five decisions - a resolved target, a
confirm gate, a candidate set, a not-found, or a
clarifying question. Your agent's job between turns is simple: ask the
user the question the decision implies, then send their answer back with
the state Kviria handed you.

The funnel ends when exactly one place or address remains. That is the
only moment a coordinate is released.

## What the funnel never does

* **It never guesses.** Two matches stay two matches until the user
  picks.
* **It never narrows on quality.** Kviria ranks by facts (distance,
  category), never by opinion - "which one is better?" is your agent's
  question to answer from the returned data, or to decline.
* **It never narrates.** Responses are decisions and facts. The phrasing
  belongs to your agent - in your brand voice, in your language.