Blitz Decision Fatigue Dataset
A specialized collection of 50,000 anonymized Blitz games annotated with precise clock times to analyze cognitive degradation in low-time situations.
Abstract
Current chess datasets often lack granular time usage data, making it difficult to study the "human" element of decision making under pressure. The Blitz Decision Fatigue dataset bridges this gap by extracting per-move clock times from 50,000 Lichess games (3+0 and 3+2 time controls).
It is specifically curated to test hypotheses related to "Decision Fatigue" — the deterioration of decision quality after a long session of making choices.
Data Structure (Schema)
| Field Name | Type | Description |
|---|---|---|
| game_id | string | Unique 8-char identifier from Lichess. |
| move_ply | int | Half-move number (1, 2, 3...). |
| clock_time | float | Seconds remaining on the clock. |
| time_spent | float | Seconds spent thinking on this specific move. |
| eval_delta | float | Change in Stockfish centipawn evaluation (Before - After). |
| is_blunder | boolean | True if eval drop > 150cp. |
Methodology & Source
Source: Lichess Open Database (Jan 2025 dump).
Filtering: Only games where both players rated 1500-2200. Games with < 10 moves removed.
Anonymization: Usernames have been hashed to `User_XXXX` to allow longitudinal analysis without revealing identity.