Replication package for "Measurement validity in political event databases is audience-specific: evidence from China and its 25 trading partners"

Complete replication package for the manuscript "Measurement validity in political event databases is audience-specific: evidence from China and its 25 trading partners". The package covers the full pipeline from raw data acquisition to final submission outputs: raw retrieval products of four political event databases (GDELT 2.36 GB, ICEWS 210 MB, Phoenix release package + processed bilateral events, Tsinghua index with dual-channel extraction validation), all acquisition/cleaning code, day-to-month aggregation scripts (five algorithms), monthly scores (full and fair-coverage calibers), the trade panel (n = 6,685) and Pew polling panel (17 countries x 13 waves, N = 170), the hand-curated gold-standard event library (712 source-verified events, inter-coder Cohen's kappa = 0.866), the complete analysis suites (PPML-HDFE, local projections, event studies, fair/full double rerun with comparison ledger), all manuscript figures (300 dpi), and all main-text and SI table CSVs (Table 1-3 + Table S4-S12 series, numbered by first-citation order). Contents: eventdb-validity-audit_raw_data.zip (raw layer) + eventdb-validity-audit_code_outputs.zip (code, derived data, figures, tables, manuscripts) + README.md / README_zh.md (structure and six-stage reproduction path) + MANIFEST.csv (md5 for all 1,064 files) + CHECKSUMS.txt. Every headline number was independently re-derived before release (end-to-end Python and R reproduction, byte-identical outputs; see README.md). Environment: Python >= 3.12, R >= 4.6 (fixest). Rights in GDELT/ICEWS/Phoenix/Tsinghua/IMF/Pew source data remain with the original publishers. Code MIT, data CC-BY 4.0.