Zipf's law is deceptively simple: the second most frequent word appears roughly half as often as the most frequent, the third one-third as often, and so on. This inverse-rank pattern (S ∝ r^{-1}) and its generalizations appear across domains—word frequencies in novels, city sizes, internet links, protein interactions, and species abundances. For decades, Herbert Simon's 1955 model seemed to explain it all with a straightforward rich-get-richer rule: new elements either introduce a novel type (innovation) or attach to existing ones proportional to their current size.

A new preprint (arXiv:2604.13184, submitted April 2026, not peer-reviewed) by Pablo Rosillo-Rodes dismantles this orthodoxy. The author shows Simon's original analysis contains a critical error: in the zero-innovation limit, the model produces a winner-takes-all monopoly (exponent α → ∞), not the Zipfian α = 1 long assumed. This flaw went largely unnoticed despite thousands of citations in linguistics, urban science, network theory, and biology.

Through mathematical derivation from growth master equations, Rosillo-Rodes solves for the exact time-dependent innovation rate ρ_t required to produce any desired power-law exponent α ≥ 0 in pure rich-get-richer systems. For true Zipf behavior, ρ_t must decay specifically as 1/ln(N), where N is the cumulative number of unique types at time t. This slow logarithmic decay, not a constant probability, is what maintains the balance between popularity and novelty.

The paper further demonstrates this decaying innovation rate is not model-specific but emerges universally in any system displaying power-law size rankings, independent of underlying mechanisms. Simulations of the corrected model match word-rank distributions in famous novels, while classic Simon implementations diverge. However, the work is primarily theoretical; empirical validation uses unspecified novel collections without detailed sample sizes or statistical controls—a clear limitation. As a preprint, it awaits peer review and replication.

Previous literature missed this entirely. Mark Newman's influential 2005 review ("Power laws, Pareto distributions and Zipf's law," Contemporary Physics) catalogs these distributions' ubiquity yet accepts Simon-like preferential attachment without probing innovation dynamics. Likewise, Barabási and Albert's 1999 preferential attachment model for scale-free networks (Science) assumes fixed nodes and focuses on connectivity, overlooking the open-ended, time-varying novelty injection required for true Zipfian rankings in growing systems like languages or urban populations.

This work connects patterns others have siloed. In linguistics, the 1/ln(N) decay implies authors introduce new vocabulary at a precisely tuned slowing pace as texts lengthen—matching empirical Zipf but contradicting constant-innovation assumptions in many NLP models. In biology, it suggests ecosystem speciation or gene-family expansion follows identical innovation slowdowns, explaining why power laws appear despite vastly different selection pressures. In cities and economies, new entrants (startups, suburbs) must appear logarithmically less frequently to prevent monopolies or ghost towns.

The deeper implication challenges the entire "rich-get-richer" framing: the universal governor is not mere preferential attachment but a dynamic exploration-exploitation balance encoded in how novelty rates evolve. This preprint positions the corrected mechanism as a fundamental "Drosophila" model for all such systems, enabling more accurate forecasts of inequality, language evolution, and ecological resilience. Future empirical tests across domains will determine if this mathematical necessity truly reflects real-world growth processes. What seemed settled for seventy years was built on sand; the corrected view reveals an elegant, universal clock governing diversity across science.