Code, benchmarks and experiment data for the ICAPS 2026 GenPlan workshop paper "Verifinsta: Verifying If an Instance Belongs to a Domain"

Included Files benchmarks.zip contains the formalised domains (in folder formally-represented-domains and on the top level) and the instances (in folder learning-ipc-2023-benchmarks) that were verified against the formalised domains. Living versions of the [formalised domains](https://github.com/aibasel/formally-represented-domains) and [instances](https://github.com/ipc2023-learning/benchmarks) can be found on GitHub. The three formalised domains on the top level deviate from their living versions, modified-childsnack-domain.pddl corresponds better to the automatically generated instances of the IPC 2023 learning track and the two val-\* files replace '<' by 'le' to be compatible with the plan validator VAL. experiment-scripts.zip contains the scripts for conducting the experiments. logs.zip contains the log files from the three verification experiments. Folders \*-eval contain processed data for the experiments: html files with aggregated results, as well as 'properties' files, which contain a JSON directory with combined data of all runs of the corresponding experiment. In essence, the properties file is the union over all properties files generated for each individual run. verifinsta-downward.zip and verifinsta-val.zip contain the verifinsta code as a Python package each. The former was used for the experiments using Fast Downward, the latter for the experiment using VAL. The live version of [verifinsta](https://github.com/grucla/verifinsta) can be found on GitHub. Reproducing the Data Software versions used in the original experiments: Python 3.11.5 Apptainer 1.4.5 pip 26.1.1 Create a folder structure where the following folders and files are next to each other: verifinsta-downward verifinsta-val experiment-scripts learning-ipc-2023-benchmarks formally-represented domains modified-childsnack-domain.pddl val-modified-childsnack-domain.pddl val-spanner-domain.pddl Create a Python virtual environment and install the requirements from experiment-scripts/requirements.txt. Verification using Fast Downward For the experiments using Fast Downward (v7.py, v7-30min.py) install verifinsta-downward (cf. README in verifinsta-downward), download the Apptainer file of Fast Downward 24.06 (cf. README in verifinsta-downward), and place it next to experiment-scripts, verifinsta-downward, etc. For the main experiment run v7.py from experiment-scripts. For the experiment with a time limit of 30 min (instead of 5 min) run v7-30min.py from experiment-scripts. The detailed experiment data can be found in experiment-scripts/data/v7/, respectively in experiment-scripts/data/v7-30min/, (assuming that you started the experiments from the experiment-scripts folder). Processed and aggregated data is located in v7-eval, respectively v7-30min-eval, next to it. Verification using VAL For the experiment using VAL (v7-val.py) install verifinsta-val (cf. README in verifinsta-downward) and VAL. Note that for the original experiments an older version of VAL was used. To use this older version build VAL as follows (assuming Ubuntu) and add the resulting validate binary to your PATH: sudo apt install g++ make flex bisongit clone https://github.com/KCL-Planning/VAL.gitcd VALgit checkout a5565396007eee73ac36527fbf904142b3077c74make clean # Remove old binaries.sed -i 's/-Werror //g' Makefile # Ignore warnings.make Add a file called sas_plan containing the single line ; cost = 0 (unit cost) next to experiment-scripst, verifinsta-val, etc. Run the experiment with v7-val.py from experiment-scripts. The detailed experiment data can be found in experiment-scripts/data/v7-val/ (assuming that you started the experiments from the experiment-scripts directory). Processed and aggregated data is located in v7-val-eval, next to it.