discover_sorting

GitHub

hardcomparatorsPython

results

Measured by number of comparators used — lower is better.

80 comparators60 comparators

sota

Claude-Opus-4.6

reward 1.000

Usage

Run the reference answer to verify your environment is set up correctly

$ harbor run -p tasks/discover_sorting

Test model

$ harbor run -p tasks/discover_sorting \
  -a claude-code -m claude-opus-4-6

Description

Generate a correct 16-input sorting network using as few comparators as possible. A sorting network is a fixed sequence of compare-exchange operations, and correctness is verified exhaustively on all 65,536 binary inputs using the zero-one principle.

Files

path

permission

/app/solve.py✎ Edit

/app/main.pyRead-only

Rules

01Edit /app/solve.py only. Python standard library only.
02generate_network() must return a list of 2-tuples with 0 <= i, j < 16 and i != j.
03Incorrect networks score 0.