docs(card): simplify dataset presentation and link Bara

README.md

@@ -38,92 +38,58 @@ dataset_info:
     num_examples: 160
 ---
 
-# GAIA2-mini
-
-A community distribution of the **official 160-scenario GAIA2 `mini/validation` subset** and its **294 published filesystem files**. Scenario and resource bytes are unchanged. The bundle is approximately **421.6 MB** plus documentation and manifests.
-
-| Capability | Scenarios |
-| --- | ---: |
-| Adaptability | 32 |
-| Ambiguity | 32 |
-| Execution | 32 |
-| Search | 32 |
-| Time | 32 |
-| Total | 160 |
-
-Counts are measured from the pinned Parquet file. The original `validation` split and row order are preserved. Each scenario contains simulated app state, user/environment events and oracle events used by ARE to establish reference behavior for scoring.
-
-```text
-GAIA2-mini/
-├── data/validation-00000-of-00001.parquet
-├── filesystem/demo_filesystem/    # Documents/, Downloads/, and other original paths
-├── tasks.jsonl                    # Row index, IDs, categories and event counts
-├── data-manifest.json              # File sizes and SHA-256
-├── provenance.json                 # Source revisions, compatibility and validation
-├── known-missing-files.json        # File-tree references absent from the official filesystem
-├── upstream/                      # Original dataset cards, preserved verbatim
-├── README.md
-└── NOTICE.md
-```
-
-## Download
-
-With uv installed:
-
-```sh
-uvx hf download zhangdw/GAIA2-mini --type dataset --local-dir GAIA2-mini
-```
+<div align="center">
 
-The files are ready in their final directory layout; there are no archives to reconstruct or Docker images to import. Keep the filesystem directory structure intact.
+<img src="assets/overview.svg" width="100%" alt="GAIA2-mini — 160 scenarios, 5 capabilities, 294 filesystem files">
 
-Programmatic access to the scenario table is also available:
+<p><a href="#download"><b>Download</b></a> &nbsp; · &nbsp; <a href="https://github.com/facebookresearch/meta-agents-research-environments"><b>ARE ↗</b></a> &nbsp; · &nbsp; <a href="https://arxiv.org/abs/2509.17158"><b>Paper ↗</b></a></p>
 
-```python
-from datasets import load_dataset
-
-tasks = load_dataset("zhangdw/GAIA2-mini", "mini", split="validation")
-```
+</div>
 
-The `data` column is a JSON string in ARE's native export format. The scenario identity inside that JSON is `metadata.definition.scenario_id`. The `id`, `scenario_id`, `split` and `category` columns retain their upstream meanings. `tasks.jsonl` is an index into the original zero-based Parquet row order.
+The official GAIA2 **mini / validation** subset, packaged with its published
+filesystem resources for native ARE evaluation. This community distribution keeps
+all scenario and resource bytes unchanged. **32 scenarios per capability**:
+Execution · Search · Adaptability · Time · Ambiguity.
 
-## Use with Glitter / ARE
+<a id="download"></a>
 
-This bundle targets **native Meta Agents Research Environments (ARE)** and records validation against **ARE 1.1.0**. ARE and the chosen agent harness are installed separately; actor and judge model access are configured separately as well.
+## 📦 Get the data
 
-For sibling directories `Glitter/` and `GAIA2-mini/`, set these values in `Glitter/.env`:
+With uv installed, download the complete **~422 MB** bundle:
 
-```dotenv
-GLITTER_DATASET_PATH=../GAIA2-mini/data/validation-00000-of-00001.parquet
-GLITTER_FILESYSTEM_PATH=../GAIA2-mini/filesystem/demo_filesystem
+```sh
+uvx --from huggingface_hub hf download zhangdw/GAIA2-mini --type dataset --local-dir GAIA2-mini
 ```
 
-Glitter converts the Parquet scenarios automatically. For another ARE integration, `DEMO_FS_PATH` must point to the `filesystem/demo_filesystem` directory before importing ARE. ARE remains responsible for loading apps, advancing events and scoring with its native judge.
-
-## Validation and upstream resource gaps
-
-Validation details are recorded in `provenance.json`. **All 160 scenarios loaded and completed ARE 1.1.0 oracle preprocessing successfully; all 110 distinct MCP tool schemas validated; no model calls were made.** Source hashes, scenario identities, category counts and automatic Parquet materialization were also checked. These checks establish data compatibility; they are not model evaluation scores or Linux deployment tests.
-
-The scenario file trees contain **four paths absent from the pinned official filesystem repository**, listed in `known-missing-files.json`. Each appears in 159 scenario file trees. This package preserves the official resource set and records the missing references. The downloaded 294 resource files are complete and verified against the official source revision. Agents exploring a missing path may still encounter the upstream missing-resource behavior.
-
-## Sources and integrity
+The scenario table is in `data/`; the 294 resource files are in
+`filesystem/demo_filesystem/`. Keep the resource paths intact.
 
-| Component | Official repository | Pinned revision |
-| --- | --- | --- |
-| Scenarios | [meta-agents-research-environments/gaia2](https://huggingface.co/datasets/meta-agents-research-environments/gaia2) | `78ea3bdbdeec2bdcd6afa5420915d8a22f23ed99` |
-| Filesystem | [meta-agents-research-environments/gaia2_filesystem](https://huggingface.co/datasets/meta-agents-research-environments/gaia2_filesystem) | `132e26376f5e963bb59f64bcccdd02188cb08dee` |
+<details>
+<summary><b>Python access & native ARE</b></summary>
 
-The scenario file is the upstream `mini/validation-00000-of-00001.parquet` (162,426,171 bytes), with SHA-256:
+```python
+from datasets import load_dataset
 
-```text
-88430154e09bb728c666bbbf367bb805c8a84f8bdee427e7d469887022ab71fe
+tasks = load_dataset("zhangdw/GAIA2-mini", "mini", split="validation")
 ```
 
-The 294 filesystem resources occupy 259,216,295 bytes. `data-manifest.json` records SHA-256 and sizes for all delivered files except the manifest itself and Hub-managed `.gitattributes`. Download caches are not part of the distribution.
+Each row's `data` field is a native ARE scenario JSON string; `tasks.jsonl` indexes
+the original row order. To use the downloaded resources, set `DEMO_FS_PATH` to the
+absolute path of `GAIA2-mini/filesystem/demo_filesystem` before starting ARE.
+Install ARE and configure your agent and judge models separately.
 
-## Attribution and intended use
+</details>
+
+---
 
-The original benchmark and filesystem are published by Meta AI Research / Meta Platforms, Inc. Please see [NOTICE.md](NOTICE.md) and the preserved upstream dataset cards for the CC BY 4.0, Llama-output and third-party-content terms. Benchmark canaries and upstream notices are preserved. The upstream authors request that this evaluation data not be used for training.
+**Sources:** Meta's [GAIA2](https://huggingface.co/datasets/meta-agents-research-environments/gaia2)
+and [filesystem](https://huggingface.co/datasets/meta-agents-research-environments/gaia2_filesystem).
+[Revisions & validation](provenance.json) · [Checksums](data-manifest.json) · [Licenses & citation](NOTICE.md).
 
-This repository packages existing evaluation data and does not define a new benchmark or add train/test partitions.
+All 160 scenarios passed ARE 1.1.0 loading and oracle preprocessing; no model score
+is claimed. Four upstream file references are unavailable; see
+[known resource gaps](known-missing-files.json). The upstream authors request that
+this evaluation data not be used for training.
 
-[ARE code](https://github.com/facebookresearch/meta-agents-research-environments) · [Project](https://facebookresearch.github.io/meta-agents-research-environments/) · [Paper](https://arxiv.org/abs/2509.17158)
+Related project: [Bara](https://github.com/zhangdw156/Bara) — a collection of native
+ARE ReAct research implementations inspired by RSI papers.

data-manifest.json

@@ -1,8 +1,8 @@
 {
   "schema_version": 1,
   "source_snapshot_complete": true,
-  "files_count": 302,
-  "total_bytes": 421726451,
+  "files_count": 303,
+  "total_bytes": 421726186,
   "excluded": [
     "data-manifest.json",
     "Hub-managed .gitattributes",
@@ -16,8 +16,13 @@
     },
     {
       "path": "README.md",
-      "size": 6208,
-      "sha256": "cd32b743ec09c5e307cfcfeba1e73f45f84696e4948b83d9771326242412ea86"
+      "size": 3065,
+      "sha256": "813aaf1014a1bff086d1556f1f866276d8587665a833c1c858720b7d40a6b9ec"
+    },
+    {
+      "path": "assets/overview.svg",
+      "size": 2878,
+      "sha256": "e7ef939f521dd89de97538c9f5b72382b14e405dc53679bab9dc27d0d063ab51"
     },
     {
       "path": "data/validation-00000-of-00001.parquet",