Detection Is Not Identification: simulation benchmark of bibliometric measures of epistemic change — code, data and figures

Reproducibility package for a simulation benchmark that evaluates whether bibliometric indicators can detect epistemic change in a research field, identify which mechanism produced it, and track its intensity, in a setting where the generating mechanism is known by construction. A stochastic citation-network model with preferential attachment, citation aging, topical homophily and multiplicative noise was run across an eighteen-cell Latin-hypercube design. The package contains 1,728 simulated fields measured at three post-event horizons (5,184 indicator observations) under a stationary control and six change mechanisms (incremental, recombination, new field, paradigm replacement, convergence, fad), plus 864 further fields (2,592 observations) for a non-stationary control in which publication volume and reference-list length grow without any epistemic change. Contents: the simulation notebook and all analysis scripts; the complete observation files and every result table; the seven manuscript figures as vector PDF and EPS and as high-resolution PNG; a traceability table mapping each reported number to its source file and variable; and a reference verification table with DOIs. Code is released under the MIT License; data and figures under CC BY 4.0. The OpenAlex calibration sample is redistributed under OpenAlex's CC0 waiver. The manuscript itself is not included in this record.