Playtrace

Human playthrough traces of PuzzleScript games.

This is the collection server's PostgreSQL database exported one Parquet file per table. Rows are unmodified. Each table is a separate config.

Tables

config	rows
games	882	the puzzle games themselves
labelers	3	contributors who played them
assignments	278	which contributor was given which game
episodes	6370	finished playthroughs — the traces
attempt_drafts	91	unfinished playthroughs
game_checkpoints	183	per-level progress
game_allocation_history	276	allocation audit trail

Schema

games — 882 rows

column	type	values
stem	string
title	string
author	string
source	string
source_kind	string	local_curated, pedro_puzzlescript_net
source_url	string
external_id	string
quality_tier	string	candidate, curated
updated_at	timestamp[ns]
total_levels	int64

labelers — 3 rows

column	type	values
id	string
username	string
created_at	timestamp[ns]

assignments — 278 rows

column	type	values
id	string
user_id	string
game_stem	string
task_tier	string	optional, required
status	string	assigned
created_at	timestamp[ns]
passed_levels	int64
total_levels	int64
earned_cents	int64
progress_updated_at	timestamp[ns]

episodes — 6370 rows

column	type	values
id	string
game_stem	string
labeler_id	string
steps	string
step_count	int64
active_duration_ms	int64
solved	bool
created_at	timestamp[ns]

attempt_drafts — 91 rows

column	type	values
labeler_id	string
game_stem	string
steps	string
step_count	int64
active_duration_ms	int64
updated_at	timestamp[ns]

game_checkpoints — 183 rows

column	type	values
labeler_id	string
game_stem	string
level_value	int64
checkpoint_value	string
updated_at	timestamp[ns]
completed	bool
passed_levels	int64
total_levels	int64
earned_cents	int64
progress_counted_at	timestamp[ns]

game_allocation_history — 276 rows

column	type	values
game_stem	string
first_user_id	string
allocation_kind	string	assignment, existing_assignment
allocated_at	timestamp[ns]

The traces

episodes.steps and attempt_drafts.steps are JSON arrays holding one entry per action taken in the game, each carrying the board state at that point. That makes the rows wide: episodes.parquet is the bulk of the dataset despite holding only 6,370 rows.

Games are keyed by stem, joining to game_stem in the other tables. Contributors are keyed by id, joining to labeler_id and user_id.

Loading

from datasets import load_dataset

ds = load_dataset("lkevincc0/playtrace", "episodes")

import pandas as pd

episodes = pd.read_parquet("hf://datasets/lkevincc0/playtrace/data/episodes.parquet")

License

Apache 2.0