Datasets:

smearle
/

puzzlescript-gists

pretty_name: PuzzleScript Human-Authored Games (Full Gist Corpus)
license: other
language:
  - en
tags:
  - puzzlescript
  - games
  - code
  - world-models
  - grid-puzzles
size_categories:
  - 10K<n<100K
task_categories:
  - text-generation
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/puzzlescript_games.jsonl

PuzzleScript Human-Authored Games (Full Gist Corpus)

34,715 human-authored PuzzleScript games — the complete source text of each — collected from public GitHub gists.

This is the full corpus: every distinct gist is kept, and each row is tagged with its deduplication cluster so you can reduce to a unique set with a one-line filter. The deduplication is reproducible from the shipped dedup_master.json + dedup_master.py; non-vanilla PuzzleScript-Plus files are excluded (listed in quarantine_ps_plus_manifest.txt, reproducible via detect_non_vanilla.py).

Each game defines rewrite rules (mechanics) and one or more levels (initial states) — useful for world models of grid-puzzle dynamics, code generation, program synthesis, and game design.

Fields

field	description
`id`	the raw GitHub gist id
`content`	full PuzzleScript source
`source_collections`	named archive(s) this game was reconciled from (e.g. `PuzzleScript Gallery`, `Pedro's PuzzleScript Archive`, `itch.io`), or `GitHub gist` if seen only via a raw gist scrape
`parse_status`	`ok`, `parse_error`, `preprocess_error`, `timeout`
`n_objects`, `n_levels`	counts (parsed games)
`mechanics_hash`, `levels_hash`	canonical, name/art-invariant fingerprints
`dedup_group`	deduplication cluster key
`is_dedup_representative`	`True` for one game per cluster
`n_in_dedup_group`	cluster size

Deduplicating

from datasets import load_dataset
ds = load_dataset("smearle/puzzlescript-gists", split="train")
unique = ds.filter(lambda r: r["is_dedup_representative"])   # 25,105 distinct games

A game is a duplicate only when both its mechanics and levels fingerprints match an already-kept game, so genuine small variants are retained.

Provenance & license

Collected from public GitHub gists (the canonical PuzzleScript share target) and several public PuzzleScript archives, credited per-row in source_collections:

the PuzzleScript Gallery,
Pedro's PuzzleScript Archive (the PuzzleScript Game Database),
and itch.io.

No GitHub handles are emitted; the gist id is retained as the identifier. These are third-party, human-authored works redistributed for research; treat each game as belonging to its original author. License: other.