docs: refresh README for v3.0 (single task.jsonl, 220 tasks, no scenario_tier)

README.md

@@ -16,37 +16,52 @@ task_categories:
 configs:
 - config_name: LexBench-Browser
   data_files:
-  - split: l1
-    path: LexBench-Browser/l1.jsonl
+  - split: All
+    path: LexBench-Browser/task.jsonl
 ---
 
 # LexBench-Browser
 
-A public subset of LexBench-Browser benchmark for evaluating AI agents on web browsing tasks.
+Public release of the LexBench-Browser benchmark for evaluating AI browser agents on real-world web tasks.
 
 ## Dataset Description
 
-LexBench-Browser is a benchmark designed to evaluate AI Agents on real-world web browsing tasks. This public subset contains **187 tasks** that do not require login credentials, making them suitable for open evaluation.
+LexBench-Browser evaluates AI agents on multi-step browsing tasks across 50+ Chinese and English websites — e-commerce, social, video, tools/education, finance, and more. This public release contains **220 tasks** that do not gate on user login, suitable for open evaluation. The full dataset (including login-gated tasks) is available in the private mirror.
 
-### Dataset Statistics
+### Dataset Statistics (v3.0, 2026-04-28)
 
 | Attribute | Value |
 |-----------|-------|
-| Total Tasks | 187 |
-| Version | 2.0 |
-| Scenario Tier | L1 (No login required) |
-| Languages | Chinese (115), English (72) |
+| Total tasks | 220 |
+| Version | 3.0 |
+| Languages | Chinese (142), English (78) |
+
+#### Task type
+
+| Type | Count | Description |
+|------|-------|-------------|
+| T1   | 175   | Information retrieval |
+| T2   | 45    | Website operations |
+
+#### Reasoning type
+
+| Reasoning | Count |
+|-----------|-------|
+| single_step  | 122 |
+| multi_step   | 74  |
+| deep_analysis | 24 |
 
 #### Domain Distribution
 
 | Domain | Count |
 |--------|-------|
-| ecommerce | 25 |
-| finance_gaming | 41 |
-| general | 6 |
-| social_lifestyle | 28 |
-| tools_education | 44 |
-| video_platform | 43 |
+| finance_gaming   | 45 |
+| video_platform   | 44 |
+| tools_education  | 41 |
+| general          | 34 |
+| social_lifestyle | 31 |
+| ecommerce        | 24 |
+| gaming           | 1  |
 
 ## Download
 
@@ -56,16 +71,12 @@ LexBench-Browser is a benchmark designed to evaluate AI Agents on real-world web
 from datasets import load_dataset
 
 dataset = load_dataset("Lexmount/LexBench-Browser")
-
-# Access a specific split
-l1_tasks = dataset["l1"]
-print(l1_tasks[0])
+all_tasks = dataset["All"]
+print(all_tasks[0])
 ```
 
 ### For Users in China (Mirror)
 
-If you have difficulty accessing Hugging Face, use the mirror:
-
 ```python
 import os
 os.environ['HF_ENDPOINT'] = 'https://hf-mirror.com'
@@ -74,7 +85,7 @@ from datasets import load_dataset
 dataset = load_dataset("Lexmount/LexBench-Browser")
 ```
 
-Or set environment variable before running:
+Or set the variable before running:
 
 ```bash
 export HF_ENDPOINT=https://hf-mirror.com
@@ -84,34 +95,29 @@ python your_script.py
 ### Manual Download
 
 ```bash
-# Install huggingface-cli if not installed
 pip install huggingface_hub
-
-# Download dataset
 huggingface-cli download Lexmount/LexBench-Browser --repo-type dataset --local-dir ./data
 ```
 
 For China users:
+
 ```bash
 export HF_ENDPOINT=https://hf-mirror.com
 huggingface-cli download Lexmount/LexBench-Browser --repo-type dataset --local-dir ./data
 ```
 
-
-### Task Format
-
-Each task contains the following fields:
+## Task Format
 
 ```json
 {
   "id": 1,
   "query": "Task description",
-  "scenario_tier": "L1",
   "task_type": "T1",
   "reasoning_type": "multi_step",
   "domain": "ecommerce",
   "difficulty": "medium",
   "login_required": false,
+  "login_type": "",
   "target_website": "www.example.com",
   "language": "zh",
   "website_region": "zh",
@@ -121,27 +127,30 @@ Each task contains the following fields:
     "common_mistakes": ["Common mistake 1"],
     "scoring": {
       "total": 100,
-      "items": [{"name": "...", "score": 30, "description": "..."}],
-      "deductions": [{"reason": "...", "penalty": 20}]
+      "items": [{"name": "...", "score": 30, "description": "..."}]
     }
-  }
+  },
+  "score_threshold": 60,
+  "robustness_tags": ["chinese_rendering", "data_extraction"]
 }
 ```
 
-### Field Descriptions
+### Field descriptions
+
+- **task_type**: `T1` (information retrieval) or `T2` (website operations).
+- **reasoning_type**: `single_step` | `multi_step` | `cross_platform` | `deep_analysis`.
+- **domain**: business domain (`ecommerce`, `social_lifestyle`, `video_platform`, `tools_education`, `finance_gaming`, `general`, `gaming`).
+- **difficulty**: `easy` | `medium` | `hard`.
+- **language**: task description language (`zh` for Chinese, `en` for English).
+- **website_region**: target website region (`zh` for Chinese sites, `en` for international sites).
+- **login_required**: whether the task is login-gated. Almost all public tasks are `false`; a small number of API-style tasks are `true`.
+- **robustness_tags**: ordered, deduplicated list across 6 categories / 16 tags (popup interference, sequence complexity, content dynamics, anti-crawl, localization, complex interaction).
 
-- **scenario_tier**: `L1` - Tasks that do not require login
-- **task_type**: `T1` (Information retrieval) or `T2` (Operation execution)
-- **reasoning_type**: `single_step`, `multi_step`, `cross_platform`, or `deep_analysis`
-- **domain**: Business domain (ecommerce, social_lifestyle, video_platform, etc.)
-- **difficulty**: `easy`, `medium`, or `hard`
-- **language**: Task description language (`zh` for Chinese, `en` for English)
-- **website_region**: Target website region (`zh` for Chinese sites, `en` for international sites)
+> **Migration note (v2.x → v3.0)**: the per-record `scenario_tier` field has been removed and the per-tier files (`l1.jsonl`, `l2.jsonl`, `l3-api.jsonl`, `l3-security.jsonl`, `tasks.jsonl`) have been replaced by a single `task.jsonl`. Slice the data with `login_required`, `domain`, or `risk_control` instead.
 
 ## Related Resources
 
-- [LexBench-Browser-Private](https://huggingface.co/datasets/Lexmount/LexBench-Browser-Private) - Private subset requiring login (L2/L3)
-- [LexBench-Online-Mind2Web](https://huggingface.co/datasets/Lexmount/LexBench-Online-Mind2Web) - Online Mind2Web benchmark
+- [Lexmount/LexBench-Browser-Private](https://huggingface.co/datasets/Lexmount/LexBench-Browser-Private) — full dataset including login-gated tasks (377 tasks).
 
 ## License
 
@@ -149,8 +158,6 @@ Apache 2.0
 
 ## Citation
 
-If you use this dataset, please cite:
-
 ```bibtex
 @misc{lexbench-browser-2026,
   title={LexBench-Browser: A Benchmark for Web Browsing AI Agents},