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

# Testing Prompts

> See exactly how any prompt routes before it hits production.

# Testing Prompts in the Playground

The Playground lets you see how any prompt will be routed before it hits production. Use it to verify tier classification, check cost estimates, and debug unexpected routing decisions.

***

## How to Test a Prompt

1. Open **Playground** from the dashboard sidebar
2. Type or paste your prompt into the input box
3. Click **Route** or press `Ctrl+Enter`

> 📷 **\[Screenshot: Playground with a prompt in the input box and the Route button visible]**

***

## Reading the Result

After routing, you see a result panel with:

> 📷 **\[Screenshot: Routing result panel showing tier badge, model name, confidence bar, cost comparison, and fallback chain]**

### Tier Badge

Shows which tier was assigned, for example `STANDARD`. Color-coded:

| Color  | Tier     |
| ------ | -------- |
| Gray   | NANO     |
| Green  | SIMPLE   |
| Teal   | LIGHT    |
| Blue   | STANDARD |
| Purple | COMPLEX  |

***

### Selected Model

The model Routor would use for this request. Shown as `provider/model-name`, for example `deepseek/deepseek-v4-pro` or `google/gemini-3.5-flash`.

***

### Confidence Score

How certain the classifier is about the tier assignment, from 0 to 1.

| Range        | Meaning                               |
| ------------ | ------------------------------------- |
| 0.75 to 1.0  | High confidence, clear classification |
| 0.55 to 0.74 | Medium confidence, reasonable match   |
| Below 0.55   | Borderline, sat near a tier boundary  |

***

### Cost Estimate

Shows the estimated cost for this request on the selected model vs what it would have cost on a baseline model like Claude Opus 4.8.

Example:

```
Selected:  $0.0008   (Kimi K2.6)
Baseline:  $0.0250   (Opus 4.8)
Savings:   96.8%
```

***

### Fallback Chain

The ordered list of models Routor would try if the primary fails:

```
1. zai/glm-5.2                  primary
2. moonshot/kimi-k2.6           first fallback
3. google/gemini-3.5-flash      second fallback
```

***

## Testing Against a Profile

To see how a prompt routes under a specific profile:

1. Select the profile from the **Profile** dropdown above the input box
2. Type your prompt and click **Route**

The result will reflect the profile's tier floor, ceiling, and capability caps.

> 📷 **\[Screenshot: Profile dropdown above the prompt input with a profile selected]**

***

## Example Prompts to Try

These cover the full tier range and help you understand how the classifier works:

| Prompt                                     | Expected Tier | Why                                                                              |
| ------------------------------------------ | ------------- | -------------------------------------------------------------------------------- |
| "hi"                                       | NANO          | Single greeting word                                                             |
| "what's 2+2"                               | SIMPLE        | Trivial fact                                                                     |
| "explain what a closure is"                | LIGHT         | General explanation                                                              |
| "refactor this auth middleware"            | STANDARD      | Detailed code task                                                               |
| "design a microservice architecture"       | COMPLEX       | Architecture level                                                               |
| "prove the square root of 2 is irrational" | COMPLEX       | Formal mathematical proof, routed to a reasoning-specialist model within COMPLEX |

Run each one and compare the confidence scores to build intuition for how the classifier works.

***

## What the Playground Does Not Do

* It does not make a real API call to the model. It only runs the routing decision.
* It does not charge credits for testing prompts.
* It does not show the actual model response, only the routing decision.

To see a real response, use the actual API with your app or a tool like Postman.