I’ve been working on an empirical study of attention dynamics on Hacker News; analyzing decay curves, preferential attachment, survival, and early-engagement prediction using ~300k items and 72k temporal snapshots collected in December 2025.
Preprint available on SSRN
Today I saw this Show HN: 22GB of Hacker News in SQLite, served via WASM shards. Downloaded the HackerBook export and ran a subset of my paper’s analytics on it.
Caveat: HackerBook is a single static snapshot (no time-series data). Therefore I could not analyze lifecycle analysis, early-velocity prediction, or decay fitting. What can be computed: distributional statistics, inequality metrics, circadian patterns.
Score distribution (CCDF + power-law fit)
Attention inequality (Lorenz curve + Gini)
Circadian patterns (volume vs mean score, UTC)
Score vs direct comments (proxy)
Direct comments distribution (CCDF, proxy)
Mean score vs direct comments (binned, proxy)
Sources: Paper (SSRN) · hn-archiver repo · HackerBook repo