Tags

Tags give the ability to mark specific points in history as being important

  • v2.1.1

    066d30df · chore: record the archival DOI, so the release is citable rather than merely public · 
    Reproduction package, revision 1 submission

Archived at https://doi.org/10.5281/zenodo.21886912

This is the version the manuscript's Data Availability statement cites. It adds the
self-contained LaTeX source archive as a build target, removes the target journal from
the tracked tree, and fixes the closure checker on Windows, which was found by running
this package from a clean Windows 11 clone the way the reviewer described.

  • v2.1.0

    32b025b9 · build: the LaTeX upload only built inside this repository, and the tree named the venue · 
    Reproduction package, revision 1 upload

Adds the self-contained LaTeX source archive as a build target and removes the target
journal from the tracked tree; the venue now lives in the gitignored submission config.
Supersedes v2.0.0-revision1, which predates the Windows, CI and packaging fixes and
whose release pipeline never completed.

  • v2.0.0-revision1

    5ef55fe3 · docs: reviewer closure matrix, reproduction guide, and clone-aware validators · 
    IEEE Access Access-2026-31709, revision 1 artifact

Reviewer 1 raised seven substantive findings after auditing the code archive. All
seven were verified independently and all seven were correct. This tag is the state
of the reproduction package as submitted with the revision.

What the revision changed about the paper's central claim: the reported +0.10 AUC-ROC
gain from network-flow fusion is reproducible, but it is largely an artifact of
attributing each flow to every user who authenticated to its source host that day
(mean fan-out 4.09, with 16.8% of flows silently discarded by an undisclosed cap) and
of flow data covering only 29 of the corpus's 58 days, all 18 red-team days among
them. Under a temporally consistent single-user attribution rule and restricted to
flow-covered days, the flow modality contributes +0.006 with a confidence interval
spanning zero. Authentication-plus-process fusion survives both corrections at +0.031
(Holm-adjusted p = 0.012).

Verified at this tag, from a fresh clone with no corpus:
  384 unit tests, 41 end-to-end tests on the synthetic mini corpus, lint clean,
  every documented path resolves, all nineteen reviewer items close, and the
  generated LaTeX tables match the committed ones byte for byte.

Verified with the corpus: two independent rebuilds from scratch produce all 22 result
tables bit-identically.