Despite the remarkable progress demonstrated by state-of-the-art AI systems, they nevertheless struggle when presented with new problems beyond those upon which they were trained.  This limitation has been brought into focus by François Chollet's [ARC-AGI](https://arcprize.org/arc-agi) benchmark suite (and subsequent [ARC Prize 2024](https://www.kaggle.com/competitions/arc-prize-2024) and [ARC Prize 2025](https://www.kaggle.com/competitions/arc-prize-2025) competitions), in which each task is presented as a series of <*input*, *output*> grids illustrating some specific transformation.  All tasks can be played at [arcprize.org/play](http://arcprize.org/play) &mdash; for one such example, visit the link below:

> https://arcprize.org/play?task=543a7ed5

In this competition, you will be presented with all 400 tasks from the public training set (v1) and challenged to produce a Python 3 program for each that exhibits the desired behavior.  Not only must these programs be functionally correct, but (as an added twist) should also be *as minimal as possible*.  A set of concise source codes emphasizing robustness and simplicity could potentially serve as canonical reference solutions for this seminal dataset, and &mdash; once open-sourced to the broader research community &mdash; might contribute toward the development of more versatile AI systems.

For example, consider the following (hypothetical) task #000:
![Task 000](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F25050929%2F03306cf1d4467d95bbb119bb1a7d5fc0%2Fshadows.png?generation=1745287496419783&alt=media)

Your .zip submission might include a file `task000.py` containing the following 150-character program:
```
def p(g):
 for r, row in enumerate(g):
  for c, color in enumerate(row):
   if r and c and color==5 and g[r-1][c-1] not in [0,5]: g[r][c]=0
 return g
```