Compression-Interleaved Parenthesization Defects in LLM Replay Artifact Graphs: A Predeclared Null Test

DOI: https://doi.org/10.5281/zenodo.21355572 This archive contains a companion research paper and compact evidence capsule for a predeclared test of whether low-rank-compression-interleaved parenthesization defects in LLM replay-artifact cycles exceed an edge-spectrum-matched, graph-unconstrained orientation null. Across twelve frozen source runs and two nontrivial compression ranks, all 24 predeclared endpoints were informative and coverage-complete, and none met the predeclared criterion for positive directional excess. Descriptively, all 24 endpoint-level hierarchical median log-ratios were negative. That complete reverse-direction pattern was not predeclared and is not treated as confirmatory evidence. The capsule includes the paper and TeX source; the frozen execution plan; the first empirical result memo; a post-hoc descriptive reporting addendum; exact code snapshots from the recorded runner and summarizer commits; the canonical case manifest; per-case runner manifests and checksums; the complete endpoint CSV; block-level scores; mechanical execution receipts; and a transparent post-execution software-environment reconstruction. The current null preserves marginal edge spectra while breaking shared-node realizability. The result therefore does not support the targeted positive excess claim, but it cannot determine whether the negative comparison reflects a genuinely smaller observed defect or inflation of the null caused by destroying shared-node coupling. No new scientific experiment was run to prepare this archival package. Immutable Git tag: gate12c1-parenthesization-defects-v1.0.0 Repository: https://github.com/Udonburo/pale-ale Licensing: Non-code publication materials are released under CC BY 4.0; code under code/ is released under MPL 2.0. See LICENSES.txt in the capsule for the file-scope mapping.