Artifact(updating)

Artifact Url

Code

Download the Aritfacts

Environment

Install the Python dependencies used by this pipeline, for example:

pip install openai anthropic word2number tqdm math-verify wolframclient timeout-decorator

For API-based steps, set keys through environment variables:

export OPENAI_API_KEY=...

export DEEPSEEK_API_KEY=...

export DASHSCOPE_API_KEY=...

export ANTHROPIC_API_KEY=...

For using Mathematica execution, also set:

export WOLFRAM_KERNEL_PATH=/path/to/WolframKernel

Alternatively, if WolframKernel is already available in your system PATH, you can omit this variable.

Before setting WOLFRAM_KERNEL_PATH, make sure Wolfram Engine is installed locally.

Download Wolfram Engine from:

https://www.wolfram.com/engine/

Workflow

1. Seed Formalization and Filtering 

2. Baseline Establishment 

3. Structured Mutation Generation 

4. Parameterized Template Extraction .

5. Controlled Calibration via Randomization 

6. Solver Execution and Semantic Verification 

Main Files

• pipeline.py: end-to-end orchestration

• front_half_pipeline.py: formalization, verification, filtering, and seed preparation

• executor.py: tool-aware execution

• prompts.py: generation, repair, reconstruction, and verification prompts

• parameterized_mutation.py: template extraction and parameterized variant generation

• seed_adapter.py: seed adaptation

Command

python pipeline.py \

  --raw-dataset-file /path/to/raw_dataset.json \

  --mutation-count 10 \

  --parameterized-count 3 \

  --timeout-seconds 20 \

  --intermediate-dir /path/to/intermediate_outputs \

  --output reports/batch_report.json