Kopterix: A Low-Footprint Instrument for Observing Agent-Native Populations, with a Two-Month Validation on Moltbook

Kopterix is a passive longitudinal instrument for observing agent-native populations on social platforms where autonomous agents produce the public posts. The instrument samples posts from Moltbook every six hours across three temporal layers, a surface layer of 0 to 2 hours, a mid-stream layer of 3 to 8 hours, and a residue layer of 12 to 24 hours, and computes scalar measures of lexical entropy, semantic homogeneity, and drift between runs over the resulting series. This deposit is the complete record of a two-month validation run. It contains the raw sampled inputs, the canonical analysis scripts, the final tables and figures, the software environment record, and a README with full provenance notes. The analyses cover temporal structure, embedding geometry diagnostics, memory timescales, and changepoint detection. They are evaluated against a shuffle baseline and an exact permutation baseline for layer labels. MANIFEST_SHA256.txt at the archive root lists every archived file except itself with its SHA-256 checksum. Archive integrity was verified through extraction and exact hash comparison before upload. The accompanying PDF is the full validation report compiled from the canonical source. The archive contains all materials needed to reproduce the analyses.