latex2docx benchmark pack (template-preserving AST-based LaTeX-to-DOCX compiler)

Benchmark pack accompanying the journal submission 'A template-preserving AST-based LaTeX-to-DOCX compiler with native Office Math ML equation output' (Coruh, 2026). Every numerical figure in the manuscript is either reproducible from the deposit (single-tool corpus) or auditable from the canonical CSV / JSON snapshots it carries (four-pipeline matrix and the structural-presence regression check). Contents. A single 188 MB zip with 2907 entries: the production binary (latex2docx 1.0.4 Windows installer), the 40-class publisher template matrix, the 39 feature-coverage mockups, the 203 graded body documents, the per-stem pdflatex reference-PDF cache (~45 MB), benchmark runner scripts, metric modules and the canonical CSV / JSON snapshots from the run reported in the manuscript. A step-by-step reproduction guide (BENCHMARK_GUIDE.md) is included at the root of the unpacked archive. Reproducibility scope. The single-tool corpus (R_surv, M_edit, P_fix, T_build, V_sim) reproduces end-to-end from the bundled binary against the bundled corpus. The four-pipeline benchmark matrix additionally requires Pandoc 3.5, make4ht and plastex on the local PATH; the corresponding CSV / JSON snapshots are bundled so reviewers can audit every figure without re-running the matrix. Access. This record is published under Restricted access. The DOI and landing page are publicly resolvable; the files require a Zenodo secret access link issued by the corresponding author. During peer review the link is supplied to the journal's editorial office under separate cover for distribution to assigned reviewers; after publication, the corresponding author continues to issue secret links to qualified researchers on reasonable e-mail request ([e-mail]). See LICENSE-NOTES.md for the full access conditions. Licensing (subject to access conditions above). Depositor-authored documentation, scripts and measurement CSV / JSON snapshots: CC-BY-4.0 with attribution. Forty publisher LaTeX classes: upstream LPPL. Bundled bin/*.exe binaries: proprietary, commercially licensed under a 30-day evaluation licence anchored to the binary's build date (commercial licence and current evaluation builds via https://coruhtech.github.io/latex2docx-toolkit/). Re-distribution of the deposit, or of any component in isolation, requires the corresponding author's written authorisation. Authoritative per-component disclosure is in LICENSE-NOTES.md. Integrity. SHA-256 manifest in SHA256SUMS.txt at the deposit root. The zip itself hashes to 33651f0071c6ee1866e9ae10de317ce2349c228e7ff8e11df6218d965e8804a3.