Research Data and Code for "Committing Against a Stealthy Attacker: Detection Limits and the Value of Verification When Attacks Mimic Faults"

This package holds the data and code behind the article "Committing Against a Stealthy Attacker: Detection Limits and the Value of Verification When Attacks Mimic Faults". The article studies an operator who commits in advance to a rule for flagging anomalies on a trading gateway, facing an attacker who knows the rule and keeps the visible effect of the attack close to that of an ordinary fault. The results of the article are proved in the text, and two computations complement them. The first is a numerical study that computes the operator's best commitment on randomly drawn linear-Gaussian problems. The second compares the threshold rule tuned to the attack that does the most harm at a given level of stealth with the best threshold rule, outside the conditions under which the first is proved optimal. The package contains everything behind both: the result files with a record of how each was produced, the solver of the game, the scripts of the two computations, and the seed ranges from which every problem is drawn. The computations read no outside data. After installing the package, a single build command rebuilds every number of the article from the result files in a few seconds. The README explains how to rerun the two computations from their seeds and how to check the new outputs against the recorded checksums. Licences: the code is under the MIT License; the results, run records and documentation are under CC BY 4.0.