PVBRF-ID: Private, Verifiable and Byzantine-Robust Federated Intrusion Detection for Oilfield Enterprise Networks
Aljumaily et al. · arXiv 2610.12704
The outcome is the verdict on the one claim below, under the protocol we fixed in advance. It is not a verdict on the paper as a whole.
Claim tested
PVBRF-ID beats FedAvg and is close to multi-Krum.
What we ran and what came back
The authors' simulation notebook, unmodified, on 10 seeds they did not use, with 3 of 10 sites mounting the adaptive attack. PVBRF-ID with clipping reached a macro-F1 of 0.652 (paper 0.641). FedAvg collapsed to 0.041, so the attack works as described. The difference from FedAvg was +0.611 (interval +0.583 to +0.640; bound +0.40). Against plaintext multi-Krum the difference was -0.051 (interval -0.069 to -0.033; bound -0.10): multi-Krum is better, as the paper reports. On the authors' own seeds, 12 of 15 result rows were bit-identical to their published file; the three FedAvg rows differed by up to 0.034, which does not change the comparison. The cryptographic parts (secure comparison, clipping and range proofs) are modelled in the simulation as ideal and are not tested here. One dataset and one attack.
What it does not support
Other attackers or data.
Record
Protocol, deposited before the run: 10.5281/zenodo.23442738
The authors reviewed this result before it was listed. The authors replied on 15 November 2026. They had read the pre-registration, had no objection to the outcome being listed as described, and confirmed the scope notes. They said they will look for the source of the difference in the three FedAvg rows.
How the protocol works is on the method page. If you are an author of this paper and something here is wrong, write to customers@aegismind.app and we will correct it on this page.