---
license: cc-by-4.0
language:
- en
- sk
- el
- sv
- cs
- pl
- it
- hu
- ro
- nl
- pt
- bg
- fr
- de
- es
pretty_name: Total.js Forge
size_categories:
- n<1K
task_categories:
- text-generation
tags:
- code
- totaljs
- jcomponent
- javascript
---

# Total.js Forge dataset

Documentation, recipes and component pages for **Total.js 5** (`total5@0.0.19`) and **jComponent 20**, written so that a language model can learn the two frameworks. Source: https://forge.totaljs.es/

Execution evidence is recorded per page and content revision. `verified: true` requires a matching execution receipt for the declared framework version; the evidence includes the content hash, date, runtime and test kind. Documentation without an executable test is marked `documentation`. A test without a matching receipt is marked `unrecorded`, not a failed test. Legacy imports do not acquire execution evidence merely by being published. Translations reuse the English code evidence; their prose is not certified by that execution.

Total.js 4, jComponent 19 and older are **not** covered: their APIs differ and a model should not mix them with these.

## Files

| File | What it is |
|---|---|
| `dataset.jsonl` | One chat example (`messages`) per page, in English: 509 rows. |
| `{section}/dataset.jsonl` | The same, for one section: `docs`, `total5`, `jcomponent` or `components`. |
| `dataset-{lang}.jsonl` | The pages that have a real translation, in that language: sk 509, el 509, sv 509, cs 509, pl 509, it 509, hu 509, ro 509, nl 509, pt 509, bg 509, fr 509, de 509, es 509. The code is identical in every language. |
| `dataset-instruct.jsonl` | The English pages with three different prompts each (1527 rows). |
| `benchmark.jsonl` | 154 tasks with a test. The page is the reference solution; the test decides whether another solution works. |
| `preferences.jsonl` | 9 pairs of wrong and right code for the same task, with the reason. For preference tuning. |
| `index.json` | The catalogue of all the pages with their addresses. |
| `llms.txt`, `llms-full.txt`, `llms-{section}.txt` | The same content in plain text for assistants that read the web. |

## Fields

`id`, `section`, `kind` (`recipe`, `guide`, `reference`, `comparison` or `component`), `framework`, `framework_version`, `language`, `url`, `markdown_url`, `tags`, `license`, `verified`, `verification`, `verification_status` (`executed`, `failed`, `unrecorded`, `documentation`), `verification_evidence` (receipt or null), `title`, `description`, `messages` (user prompt and assistant answer in Markdown), `files` (the code blocks with their file names), `test` and `test_kind` (`http`, `browser` or `component-browser`).

## License

The text and the code of the pages: **CC-BY-4.0**. Attribute "Total.js Forge" with the address of the page (`url` in every row). The component pages are generated from the documentation of the Componentator components (MIT) and name their authors.

## How the data is checked

The repository has a command that runs every page: it starts a temporary Total.js application with the files of the page and sends the requests of its test, or loads the example in Chromium. The editor requires matching test results before publishing code examples; file imports carry no implicit execution proof. Text-only docs are checked without execution. A reference test can fail when the framework changes (for example, when `total5` adds a global name that the page does not list).