# How requests will be routed (Lab)

> A separate typed model picks a fixed route for each request, or abstains and leaves the call to you. It runs in shadow mode today.

Status: Lab · Canonical: https://paracrux.com/docs/lab/typed-request-routing

Lab: shown in demos and case studies, not for sale. This page describes what we are building, not something you can buy.

## What we are building

- A model that chooses rather than writes: for each request it picks one of a fixed set of routes (inspect, create box, profile search, clarify or unsupported) with a confidence score.
- Below 80% confidence it abstains, and the decision comes back to you.
- A missing input routes to a question. Asked for a box with no height, it routes to clarify instead of inventing one.
- In our benchmark of 28 September 2026, 16 cases each run twice, 26 of 32 answers cleared the 80% floor and all 26 matched the expected route; 6 abstained. The median answer took 270 ms.

## See it

- [How the Agent decides, with the benchmark's real answers](https://paracrux.com/how-it-decides)

## What it is not, yet

- Shadow mode only. It records the route it would choose and how confident it is, but it does not steer the Agent yet.
- The benchmark is a small, deliberate test set, not a production accuracy figure.

## What you can do today

- [When the Agent stops or asks](https://paracrux.com/docs/when-the-agent-stops.md) (Available): Why a job stops, what each message means and what to do next. It stops rather than guess, and never retries a CAD step on its own.
- [Create a simple part from a brief](https://paracrux.com/docs/create-a-simple-part.md) (Available): Describe a block with all three sizes and get a checked native part and STEP file. Leave a size out and it asks.

## Covers

- Typed request routing (Lab): A separate model picks a fixed route for each request, or abstains below 80% confidence. Shadow mode only.