AzURA Demonstrator FormaliSE2027 Version

This artifact accompanies the paper "When Assumptions Fail: Updating Models for Online Monitoring". It contains a commandline version of AzURA, a tool that repairs timed-automata models whose assumptions are violated at runtime: given a model and an observation that the model cannot reproduce, AzURA uses an SMT solver (Z3) to compute all minimal sets of changes to the model's assumptions (for example guards, invariants and clock resets) that make the observation possible again. The artifact delivers the tool, the 15 benchmark models of the evaluation, and the raw results behind Tables I-III of the paper (349 conflicting mutants over 14 models, since one of the original 15 did not produce a conflicting observation). It re-solves every recorded conflicting mutant, shows each mutation, conflicting observation and solution while it runs, checks the results against the shipped ones, and regenerates the three tables of the paper. Everything is packaged as Docker images, so no installation besides Docker is needed. As AzURA is still in development, this prototype version is wrapped in the evaluation workflow and not meant for general use.

SHA-256 of submission.zip: 5fe8cefc3660a8bbe0ba1bbcd47fdf5ca90e2338a6ab9c2a88e0a8590b3e1808

Special requirements: Docker only (images for linux/amd64 and linux/arm64); no GPU, no internet connection. The standard run takes 17-27 minutes with up to 8 processes (measured on a 14-core laptop with 15 GB for Docker); fischer-4 runs separately (estimated 25-45 minutes). All commands and expected runtimes are in README.txt in the zip.

Early light review (about 2 minutes): unpack the zip, open a terminal in the folder and run the following (use the arm64 file on ARM).

    docker load < docker/azura-artifact-linux-amd64.tar.gz
    docker run --rm azura-artifact smoke

The second command (about 20 s) runs the unit tests and replays three small models, then prints the paper's tables.

Reusable: the artifact is MIT-licensed, the sources are inside the images (/artifact), the only dependency (Z3) is pinned, all modules have docstrings, 862 unit tests run in about 10 s, and the scripts have documented options (model sets, processes, time limit, output folder). More details can be found in the accompanying readme.