Replication archive — "When Dictionaries Hit a Polysemy Ceiling: Validating a Measure of Visibility Bias with LLMs and Behavioral Evidence"

Replication archive for the manuscript "Measuring Visibility Bias in Bureaucratic Text: A Validated Instrument with Evidence from Chinese Municipal Government Work Reports" by Hongyang Xi. Pre-registration: OSF DOI 10.17605/OSF.IO/ZMJY5 (archived 2026-04-14). Contents: analysis scripts (Phase B TWFE event study, Phase B2 Sun-Abraham with Rounds 1-9, Phase E internal construct validity, Phase E2 Wikipedia third-party null, Phase F CFPS micro regression, welfare calibration); processed city-year VAI outputs; VAI lexicons (V_ORIG with 42 terms, F_ORIG with 38 terms, plus expanded V_EXT and F_EXT); CCDI inspection rounds panel with month-level timing; Wikipedia-zh scrape script; all figures; five S7 audit reports; full deviation log; README. Not redistributed here: (a) the 6,294 primary government work reports (copyright concerns of provincial publications); (b) raw CFPS microdata (Peking University user agreement); (c) raw MOHURD Urban Construction Statistical Yearbook. These are reconstructable from public sources; see README. License: CC-BY 4.0 (data, manuscript text); MIT (code). Restricted access during peer review — reviewer token provided in manuscript cover letter; upon acceptance the record will be flipped to Public.