Cross-linguistic strength of Zipf's law of abbreviation, and its phoneme-vs-orthography sensitivity, on open corpora

Zipf's law of abbreviation predicts that more frequent words are shorter. For a language, let each word type have corpus frequency $f$ and length $\ell$ (in orthographic characters, or in phonemes); the law predicts a negative monotone relation, quantified by the Spearman correlation $\rho(\ell, \log f)$ (expected $<0$). QUESTION: over a broad open-corpus language sample (Wikipedia dumps and/or UD word lists), (a) is $\rho(\ell,\log f)$ significantly negative in every language, (b) what is the cross-linguistic distribution of the effect size, (c) which languages/scripts show the weakest effect, and (d) does the effect strengthen when length is measured in phonemes (via an explicit grapheme-to-phoneme mapping) rather than orthographic characters?

Acceptance. FULLY RESOLVES: for $\geq 100$ languages from an open corpus (Wikipedia dumps or UD), compute per-language Spearman $\rho(\ell,\log f)$ with significance, report the fraction significantly negative and the full effect-size distribution, and — for the subset with an available grapheme-to-phoneme resource (e.g. epitran / PHOIBLE-informed rules) — a paired character-vs-phoneme comparison of $|\rho|$; ship the reproducible pipeline from public dumps. PARTIAL: character-based analysis over $\geq 20$ languages, or a focused phoneme-vs-orthography study on $\geq 5$ languages, or an outlier characterization.