This record contains the article “Compression Physics via Metallic Odds Maps: A Domain-Agnostic Geometry for Frontier-Bounded Efficiency” and an accompanying reproducibility kit (ZIP). The paper introduces a frontier-normalized efficiency geometry built on the Generalized Efficiency Coefficient (GEC) framework and the Compression Scaling Kernel (CSK). Each observation is represented as (Y,X,Cmax⁡)(Y, X, C_{\max})(Y,X,Cmax), normalized to an empirical efficiency frontier, mapped into odds space, and passed through a compression pairing that merges disjoint observations. Across slices, the work fits a two-parameter metallic odds map tn+1=k+β/tnt_{n+1} = k + \beta / t_ntn+1=k+β/tn and compares it to constant and proportional nulls using AIC, permutation tests, bootstrap intervals, and robustness checks for pairing order and frontier uncertainty. The study applies this pipeline to three domains: (i) frontier LLM workflows across task ecologies (QA, code generation, summarization); (ii) Escherichia coli MC4100 single-cell microcolony growth under 25°C, 27°C, and 37°C; and (iii) macro-economic and infrastructure series derived from FDIC Quarterly Banking Profile data and EIA-930 U.S. bulk electricity demand. Metallic odds structure appears most strongly in intermediate constraint regimes (code-generation ecologies, 27°C growth, and hourly demand), while slack or saturated regimes collapse back to linear nulls. The accompanying reproducibility kit (v1.0.0) includes: derived GEC slices (Y, X, Cmax) for AI, biology, and economics, a metallic-odds proof harness (phi_proof_harness_demo.py, φ-version v5_soft_clamps) that audits slices, constructs compression pairs, fits metallic and null models, runs permutation tests and bootstrap CIs, and emits JSON proof reports, selected summary JSONs, and helper scripts to (re)build slices from public Tanouchi et al., FDIC, and EIA-930 data, and to regenerate exemplar figures using matplotlib. The kit does not redistribute raw Tanouchi growth tables, FDIC/EIA source files, or full LLM prompt/response logs. All such upstream data remain under their original licenses and terms. The derived slices and scripts are sufficient to reconstruct the compression transitions, refit the metallic and null models, recompute AIC gaps and φ-slice status, and regenerate the main cross-domain results reported in the paper. See README.md inside the ZIP for detailed instructions.