---
license: cc-by-4.0
language:
- en
pretty_name: JevForge Mind2Web Gold Decisions
tags:
- web-agents
- decision-model
- candidate-ranking
- mind2web
configs:
- config_name: gold
  data_files:
  - split: train
    path: data/train.jsonl
  - split: dev
    path: data/dev.jsonl
  - split: calibration
    path: data/calibration.jsonl
  - split: test
    path: data/test.jsonl
  - split: ood
    path: data/ood.jsonl
---

# JevForge Mind2Web Gold Decisions

Private research snapshot of gold candidate-decision records used by
[JevForge](https://github.com/zwliJay/jev-forge).

## What is included

Each JSONL record contains a page state, a `choice` question over candidate
elements, a `noul` question about one candidate, complete gold target
distributions, a website group, and source annotation metadata.

| Split | Records | Websites |
|---|---:|---:|
| train | 4,642 | 49 |
| dev | 786 | 6 |
| calibration | 400 | 6 |
| test | 800 | 8 |
| ood | 386 | 4 |

The 7,014 record IDs and website groups are disjoint across splits. The
`records/all.jsonl` file contains the same records combined for conversion or
augmentation workflows.

## Important split semantics

All records are derived from the public `train` split of
`LangAGI-Lab/Mind2Web-axtree-cleaned-lite`. JevForge deterministically assigns
websites to its own internal splits.

Therefore:

- `test` is a JevForge held-out-site split, not an official Mind2Web test set;
- `ood` is a second held-out-site challenge split, not the official Mind2Web
  test-domain benchmark;
- no restricted, separately downloaded Mind2Web test archive is included.

## Gold-only scope

This repository contains the direct gold conversion only:

- positive Mind2Web candidates share the action target probability mass;
- sampled negative candidates receive zero action target mass;
- the `noul` target is derived deterministically from candidate membership;
- no generated score labels, raw generation replies, or model-service
  credentials are included.

## Reproduce the conversion

Source dataset revision:

```text
LangAGI-Lab/Mind2Web-axtree-cleaned-lite
080a4b67d0746270166a57f2a4096a6b4129523c
```

Converter defaults:

```text
seed=17
max_candidates=12
state_chars=2600
desc_chars=88
caps=6000,800,400,800,800
split unit=website
```

See `conversion_manifest.json`, `SOURCE.json`, and `checksums.sha256` for the
snapshot metadata.

## License and attribution

Mind2Web is licensed under CC BY 4.0. This transformed snapshot retains that
license and must be used with attribution to the original Mind2Web authors.
The JevForge conversion code is separately licensed under MIT.

Please cite the original Mind2Web work:

```bibtex
@misc{deng2023mind2web,
  title={Mind2Web: Towards a Generalist Agent for the Web},
  author={Xiang Deng and Yu Gu and Boyuan Zheng and Shijie Chen and Samuel Stevens and Huan Sun and Yu Su},
  year={2023},
  eprint={2306.06070},
  archivePrefix={arXiv},
  primaryClass={cs.CL}
}
```

The upstream project asks users not to redistribute its separately downloaded
official test archive. This dataset does not contain that archive.