Analysis artefacts for "Information Fusion in Cloud Intrusion Detection: A Fusion-Carrier Taxonomy and Coded Corpus, with an Audit of Decision and Model-Update Fusion"

Scripts and cached per-observation score arrays behind the quantitative claims of the survey "Information Fusion in Cloud Intrusion Detection: A Fusion-Carrier Taxonomy and Coded Corpus, with an Audit of Decision and Model-Update Fusion" (manuscript INFFUS-D-26-04676, Elsevier Information Fusion). No new dataset was introduced in the survey. The public benchmarks referenced (NSL-KDD, UNSW-NB15, CICIDS2017) are available from their original distributors. The TestCloudIDS corpus is publicly available on IEEE DataPort. The released version of that corpus contains duplicated flows and conflicting labels; the repair applied in the article is released here with the analysis scripts, and users of the corpus should apply it until a corrected version is published. The bundle holds the fusion-carrier coding sheet and the prevalence tests computed from it, the citation-context evidence sheet behind every code, the cloud-telemetry subset, the sensitivity scripts that illustrate the coding error, the prospective diagnostics, the paired bootstraps, the attainable ceiling, the federated aggregation and Krum-variance runs, the corrected FedAdam runs, the ten-seed stacking repeats, the diverse-pool and NSL-KDD analyses, the pool-by-combiner sweep, and the no-fusion control. No second-coder sheet is released for the 232-row census, for the reason given in the Supplementary Material; the 50-row pool-screen second coding is released. The released outputs let a reader check the reported tables on a laptop CPU, and the tables differ in how much of the pipeline behind them can also be re-executed. Eleven tables cannot be re-executed from this bundle and ship cached output instead. README.txt states which case each table falls under and gives the command where one applies. Regenerating the underlying runs rather than the tables retrains a 1D-CNN, so it needs a GPU and the corpora, which are distributed by their own sources. One script in the bundle checks the rest. verify_claims.py recomputes a ledger of 145 named claims from the released files, all agreeing at submission. It is a ledger and not a sweep: the decision-level cells of the out-of-fold table and the fixed Kittler family are checked by hand and are not in it. It needs no corpus and no GPU. Where a quantity would require a corpus that is not redistributed, it says so in place of the value rather than passing the claim. The bundle also carries ENVIRONMENT.txt, which records the interpreter, the resolved versions of every library the scripts import, and the GPU and driver the federated and 1D-CNN runs used, so that a reader who reproduces a different number can tell a library change from a coding one.