> For the complete documentation index, see [llms.txt](https://tekpedal.gitbook.io/turkish-ev-charging-intents/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://tekpedal.gitbook.io/turkish-ev-charging-intents/dataset-design.md).

# Dataset design

## Intent taxonomy

The dataset contains 24 records in each of eight intent classes.

| Intent               | Turkish label       | Suggested capability               |
| -------------------- | ------------------- | ---------------------------------- |
| `FIND_STATION`       | İstasyon bulma      | Station map or location index      |
| `COMPARE_PRICE`      | Fiyat karşılaştırma | Current tariff comparison          |
| `ROUTE_PLANNING`     | Rota planlama       | Route and charging-stop planner    |
| `CHARGING_SPEED`     | Şarj hızı           | Charging-speed guidance            |
| `VEHICLE_COMPARISON` | Araç karşılaştırma  | Structured vehicle comparison      |
| `HOME_CHARGING`      | Evde şarj           | Home-charging guidance             |
| `BATTERY_HEALTH`     | Batarya sağlığı     | Battery-care content               |
| `OWNERSHIP_COST`     | Sahip olma maliyeti | Cost calculator or ownership guide |

## Record schema

```json
{
  "id": "tp-01-001",
  "text": "İstanbul içinde yakınımdaki elektrikli araç şarj istasyonu nerede?",
  "intent": "FIND_STATION",
  "intent_label_tr": "İstasyon bulma",
  "target_path": "/harita",
  "split": "train",
  "language": "tr",
  "provenance": "synthetic-editorial"
}
```

| Field             | Description                         |
| ----------------- | ----------------------------------- |
| `id`              | Stable example identifier           |
| `text`            | Turkish-language query              |
| `intent`          | One of the eight intent identifiers |
| `intent_label_tr` | Human-readable Turkish label        |
| `target_path`     | Suggested TekPedal content route    |
| `split`           | `train`, `validation`, or `test`    |
| `language`        | `tr` in version 1                   |
| `provenance`      | `synthetic-editorial` in version 1  |

## Fixed splits

Split assignment is deterministic inside every class:

* 16 training examples;
* four validation examples;
* four test examples.

No tokenization or text normalization is applied. Consumers receive the UTF-8 Turkish text as maintained in the release.

## Provenance

All version 1 examples were generated editorially from documented subject lists and templates. They were not copied from search engines, customer conversations, support tickets, private analytics, or third-party datasets.

The `synthetic-editorial` marker is included in every record so that downstream users retain this limitation and do not present the corpus as observational data.