---
title: State
description: Supply one state for the case; match the model directly, or let Jev questions supply the missing facts.
icon: file-input
---

`state` is the data for the current case. Supply it once, alongside the model. In a combined Jev + Decision request, both use that same top-level state.

## Matching facts run the Decision directly

The model declares its inputs and their JSON shape. Supply dates as ISO strings, numbers as numbers, booleans as booleans, and choices as declared values.

The recorded cancellation model expects:

```json
{
  "customer_intent": "cancel_refund",
  "plan_amount": 60,
  "period_days": 30,
  "unused_days": 10,
  "accepted_on": "2026-10-21",
  "refundable": true
}
```

Submit `{decision, state}` to `POST /v1/systemone`. The receipt labels supplied inputs as `state` in `provenance`; their values stay in your original request. No provider runs. Missing or incompatible native inputs return `422 invalid_state` with their paths — there is no AI fallback that fills them in.

Include dates explicitly when rules depend on time so later runs use the same facts.

## Jev questions can supply facts

When a case supplies evidence rather than all the required facts, add Jev `questions` to the same request. With this state:

```json
{
  "message": "Please cancel my subscription and refund the days I have not used.",
  "plan_amount": 60,
  "period_days": 30,
  "unused_days": 10,
  "accepted_on": "2026-10-21",
  "refundable": true
}
```

…a `choice` question named `customer_intent` reads the message and supplies the missing fact. Jev runs first; its named answers feed the native model: Noul becomes a number in `[0,1]`, Choice a string, and Score a number. Put any confidence threshold in your model’s expressions; the API does not convert a probability into a boolean for you.

Jev answers **override state fields with the same names**. Use distinct names such as `damage_probability` for semantic observations, and keep trusted amounts and dates under their own input names. Do not put a second `state` inside `jev`.

The receipt records Jev facts in `read` with `source: jev` and labels input sources in `provenance`: `state` for supplied facts, `jev` for Jev answers. It does not echo matching state values. Facts that remain missing after Jev runs fail the Decision step: because Jev already completed, the call returns HTTP `200` with Jev’s answers and a `decision_error` instead of native answers — see [Errors](/docs/resources/errors).

The [quickstart](/docs/quickstart) runs both shapes with the same model. See [Using Jev with aityx](/docs/coming-from-jev) for a runnable combined payload and partial-failure handling.

## Keep the original case

Save the original state alongside the submitted model and receipt when you need decision evidence. The execution response does not echo the original state, model or questions. Keep your submitted data locally; `provenance` identifies input sources without duplicating their values.
