UniCode is an automated dataset of algorithm problems and test cases built on an evolutionary generation strategy. It aims to replace traditional static, manually generated problem sets and provide more diverse, challenging, and robust programming problem resources.

Dataset composition

• Problem generation (evolutionary strategy): includes three mechanisms: single problem expansion, same-type fusion, and cross-type fusion, to automatically construct novel and diverse algorithm problems.
• Test case generation (multi-source input): combines random input, adversarial input, and LLM synthetic test cases to cover a wide range of input spaces and edge scenarios.
• Results verification (multi-stage filtering): The test output is ensured to be reliable and error-free through brute-force verification, multi-solution stress filtering, model consistency voting, and LLM conflict resolution.

This dataset uses a systematic problem generation and validation pipeline to construct well-structured, challenging, and uncontaminated problems and test data, making it suitable for scenarios such as algorithm research, code generation model evaluation, and competition training.