File size: 6,155 Bytes
16c4485
1008dfe
16c4485
6f1a92b
16c4485
6f1a92b
16c4485
6f1a92b
 
 
 
 
16c4485
6f1a92b
16c4485
 
 
 
6fe24bc
e29c271
 
7632b87
16c4485
 
 
 
7632b87
16c4485
 
 
 
 
c6a3c2f
 
59888b7
 
c6a3c2f
 
 
 
 
 
 
 
 
 
 
 
 
3c97d3c
c6a3c2f
 
 
 
 
16c4485
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3c97d3c
16c4485
3c97d3c
 
 
 
16c4485
3c97d3c
16c4485
 
 
1008dfe
16c4485
 
 
 
 
 
 
 
 
 
 
 
4679c8a
 
 
 
 
 
 
6f1a92b
4679c8a
 
 
16c4485
 
6f1a92b
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
---
license: apache-2.0
task_categories:
  - other
language:
  - en
tags:
  - mouse-movement
  - captcha
  - biometrics
  - human-behavior
  - gaming
size_categories:
  - 10K<n<100K
---

# CaptchaSolve30k - Human Mouse Movement Dataset

**The largest open-source dataset of human task-specific mouse trajectories by session count and unique participants, with 30,000 discrete sessions from thousands of users. The first and only open dataset of complete human captcha-solving interactions with full behavioral replays.**

Each session captures mouse/touch trajectories, timing data, and puzzle state at physics-tick resolution. Suitable for bot detection research, human-computer interaction studies, and ML model training.

## Dataset Overview

| Metric | Value |
|--------|-------|
| Total Sessions | 30,000 |
| Game Types | 3 |
| Physics Sample Rate | 240 Hz |
| Raw Input Sample Rate | 1000 Hz |
| Average Session Duration | 8-15 seconds |

### Privacy & Anonymization

Data was collected by Capycap, Inc. under a privacy policy disclosing collection practices, anonymization methods, and intended use for AI research. Users had the option to opt out of data storage.

- All coordinates are normalized to a 200×200 logical grid—raw screen coordinates were never stored
- Render size was randomized per session, removing device-specific signatures
- Sessions cannot be linked to individuals or correlated with each other
- No IP addresses, cookies, browser fingerprints, or persistent identifiers were collected

### Terms of Use

By downloading this dataset, you agree to the following:

1. **No re-identification**: Do not attempt to identify individuals, link sessions to specific users, or combine this data with other sources for identification purposes.

2. **No biometric profiling**: Do not use this data to build biometric identification systems, behavioral authentication, surveillance tools, or any system designed to identify or track individuals based on interaction patterns.

3. **Permitted uses**: This dataset is intended for AI research, bot detection, human-computer interaction studies, and training machine learning models (including generative models and automation agents).

4. **Redistribution**: If you redistribute this data or derivatives, you must include these same restrictions.

Violation of these terms may constitute a breach of contract and could result in legal liability.

## Game Types

### Sheep Herding
Guide sheep into a pen by drawing paths with your mouse. Tests continuous cursor control and path planning.

### Thread the Needle
Navigate through gaps without touching walls. Tests precision movement and spatial awareness.

### Polygon Stacking
Drag and stack polygon shapes on a platform. Tests click-drag interactions and placement accuracy.

## Data Format

Each session is a JSON object with the following fields:

| Field | Type | Description |
|-------|------|-------------|
| `index` | int | Session index in the dataset |
| `gameType` | string | One of: `sheep-herding`, `thread-the-needle`, `polygon-stacking` |
| `puzzleSeed` | int | Seed used to generate the puzzle (for reproducibility) |
| `duration` | int | Session duration in milliseconds |
| `physicsTickCount` | int | Number of physics ticks (at 240 Hz) |
| `tickInputs` | array | Input state at each physics tick |
| `inputStream` | string | Base64-encoded raw 1000 Hz mouse samples |
| `inputSampleCount` | int | Number of raw input samples |
| `touchscreen` | bool | Whether input was from a touchscreen |

### tickInputs Format

Array of objects representing input state at each 240 Hz physics tick:

```json
{
  "x": 150.5,      // X position in grid coordinates (0-200)
  "y": 100.2,      // Y position in grid coordinates (0-200)
  "isDown": true,  // Mouse button / touch pressed
  "sampleIndex": 42 // Index into inputStream for this tick
}
```

### inputStream Format

Base64-encoded binary stream of raw 1000 Hz mouse samples. Each sample is 9 bytes:
- 4 bytes: X position (float32, little-endian)
- 4 bytes: Y position (float32, little-endian)
- 1 byte: Button state (0 = up, 1 = down)

## Usage

### Loading with Datasets Library

```python
from datasets import load_dataset

dataset = load_dataset("Capycap-AI/CaptchaSolve30k")

# Access a sample
sample = dataset['train'][0]
print(f"Game: {sample['gameType']}, Duration: {sample['duration']}ms")
```

### Decoding inputStream

```python
import base64
import struct

def decode_input_stream(b64_stream, sample_count):
    data = base64.b64decode(b64_stream)
    samples = []
    for i in range(sample_count):
        offset = i * 9
        x, y = struct.unpack('<ff', data[offset:offset+8])
        is_down = data[offset+8] == 1
        samples.append({'x': x, 'y': y, 'isDown': is_down})
    return samples
```

## Demo & Replay

Explore the dataset interactively with our demo space:

- **Visualize sessions**: Watch recorded mouse movements replay in real-time over the actual game
- **Inspect trajectories**: See exactly how humans navigate each puzzle type
- **Load samples**: Paste JSON from the dataset or load random samples directly
- **Play the games**: Try the puzzles yourself and export your own session data in the same format

[Open Demo Space](https://huggingface.co/spaces/Capycap-AI/CaptchaSolve-Demo)

## License

Apache License 2.0 - free for research and commercial use.

## Citation

```bibtex
@dataset{captchasolve30k,
  title={CaptchaSolve30k: Human Mouse Movement Dataset},
  author={Capycap Inc.},
  year={2026},
  url={https://huggingface.co/datasets/Capycap-AI/CaptchaSolve30k}
}
```

## Authors

This dataset was created by:

- **Tony Chen** - [GitHub](https://github.com/TonyChen06) | [LinkedIn](https://www.linkedin.com/in/tonychen06/)
- **James Gu** - [GitHub](https://github.com/jamesgu888) | [LinkedIn](https://www.linkedin.com/in/jamesgu888/)
- **Christopher He** - [LinkedIn](https://www.linkedin.com/in/chrismhe/)
- **Kevin Zhou** - [GitHub](https://github.com/zhoism) | [LinkedIn](https://www.linkedin.com/in/kevinzhou3/)

Published by Capycap Inc.

## Links

- [Demo Space](https://huggingface.co/spaces/Capycap-AI/CaptchaSolve-Demo)