markov-ai/sample-100-hours

---
viewer: false
---

# Markov Sample Dataset — 100 Hours

A sample of 100 hours of screen recordings with structured annotations, spanning coding, design, productivity, and general computer use workflows.

Each session contains:
- `clip.mp4` — screen recording
- `events.json` — raw mouse/keyboard/app events (JSONL)
- `metadata.json` — session metadata (category, title, duration, tags)
- `narration.json` — structured annotations with action, intent, and app per segment

## Dataset Stats

**184 sessions · 100.2 hours total**

### Session Duration
| Percentile | Duration |
|---|---|
| P25 | 17.2 min |
| Median (P50) | 25.9 min |
| P75 | 42.8 min |
| P90 | 54.6 min |

Min: 10.0 min · Max: 140.1 min

### Category Distribution
| Category | Sessions | Share |
|---|---|---|
| Productivity | 60 | 32.6% |
| Design | 53 | 28.8% |
| General | 41 | 22.3% |
| Coding | 30 | 16.3% |

### App Distribution
| App | Share |
|---|---|
| Google Chrome | 24.4% |
| Figma | 15.6% |
| Adobe Illustrator | 9.1% |
| Adobe After Effects | 6.4% |
| Visual Studio Code | 5.7% |
| Microsoft Word | 5.2% |
| AI Apps | 4.3% |
| Brave Browser | 4.2% |
| Adobe Premiere Pro | 2.4% |
| Pages | 2.3% |
| Make | 2.1% |
| Opera GX | 1.9% |
| Comet | 1.7% |
| Antigravity | 1.5% |
| Preview | 1.5% |