# pathwaycom/arc-task-gen

Generates original ARC-AGI-1-style tasks distribution-matched to the public eval set.

Repository: https://github.com/pathwaycom/arc-task-gen
Canonical: https://ross.abutalabs.com/products/arc-task-gen
Language: Python
License: MIT
License Family: permissive
Last push: 2026-08-11T09:52:10+00:00

## Health v2 (maintenance only)
Score: 56/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 97, release rhythm 35, longevity 2
- inputs: {"age_days": 29, "days_push": 22, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 6665, forks 45 (observed 2026-08-28T04:09:43.896235+00:00)

## What it is
A Python tool that generates original ARC-AGI-1-style puzzle tasks distribution-matched to the public evaluation set. It produces tasks in the standard ARC JSON format, enabling private evaluation of frontier models on problems they have not seen before.

## Use cases
- generate fresh ARC-AGI-1-style tasks for model evaluation
- create a private benchmark set to test for benchmark contamination
- evaluate reasoning models on unseen few-shot rule induction problems
- compare model performance on public vs newly generated ARC tasks
- produce ARC-format tasks.json files compatible with existing evaluation harnesses

## When to choose
- you need uncontaminated ARC-AGI-1-style evaluation data for LLM or reasoning model benchmarks
- you want to measure whether model performance on ARC reflects memorization vs genuine rule induction
- you need tasks in standard ARC JSON format for an existing evaluation harness

## When to avoid
- you need general-purpose synthetic data generation unrelated to ARC-style grid puzzles
- you want to train models on ARC tasks rather than evaluate them
- you need a full ARC evaluation harness rather than just task generation

## Facets
- artifact type: library
- maturity: active
- function: data-generation, machine-learning, benchmarking
- domain: artificial-intelligence, machine-learning, developer-tools
- platform: python, cli
- tags: arc-agi, benchmark-evaluation, synthetic-tasks, llm-evaluation, reasoning-models, few-shot-learning, algorithms

## Member repositories
- pathwaycom/arc-task-gen (main) score 56

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:43.896235+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:43:02.441020+00:00, confidence not recorded.
  - readme: https://github.com/pathwaycom/arc-task-gen (fetched 2026-08-28T04:09:43.896235+00:00, sha 5e8f854a5398)
- Data as of 2026-08-30T08:39:29.467469+00:00.