Data and code for: Genomic context rather than counting mode or genotype sets the sign of repeat-layer non-additivity in citrus

This deposit accompanies a manuscript asking whether the transposable-element (TE) layer of a citrus genome responds non-additively to combined abiotic stress, and what determines the sign of that response. The answer the manuscript reports is that the frequency of departure is reproducible across counting modes and genotypes, whereas its sign is not: restricting the analysis to intergenic insertions removes it, while unique-only counting and the second genotype reduce its magnitude without changing its sign. The deposit contains: (1) the TE annotation produced by this study, the filtered gene annotation actually used, and the per-family genomic footprint (annotation/); (2) the multi-mapping and unique-only count matrices for the 24-library single-genotype and 48-library two-genotype designs, together with the six-library feasibility probe (counts/); (3) the per-term tables behind every reported repeat-layer, gene-layer, genotype, identifiability, intergenic, H3 and permutation result (analysis/); (4) Supplementary Tables S1-S9 and the source values behind each figure (tables/, source/); (5) the analysis code, the figure-rendering code and a 134-quantity check of every number reported in the manuscript (code/); and (6) the pre-registration, the analysis pipeline and the provenance of every reported number (docs/). Reproduction is by a single command. The scripts are classified in SCRIPT_CLASSES.tsv as reproduce, render or typeset; a reproduce script must exit 0, whereas rendering a figure may be skipped if matplotlib is unavailable and is not part of the numerical reproduction contract. Two derived tables are shipped frozen rather than regenerated: the Section 3.3 unique-only comparison and the Figure 8b sign cross-tab. The Section 3.5 analysis set of 2,702 families is reproduced exactly from the deposited matrices; the narrower 1,956-family comparison set is not, because the filter that produces it is not encoded here. Both frozen tables are checked, not recomputed, by the verification script, and the README states this explicitly. The raw sequencing reads are public under BioProject PRJNA839431 (GEO GSE203331). The Dfam 4.0 consensus repeat library and the Citrus sinensis Valencia reference assembly are third-party resources and are not redistributed here; their exact releases and download locations are recorded in README.md. A MANIFEST.md5 covers every file.