The first version read all.json, which only contains included publications. That missed five already-excluded members of the original eight-account cluster. The report now also runs inside the indexer over every crawled publication via --cluster-out; the standalone CLI remains for saved included datasets. It still does not score, rank, filter, or write data.json.
The rule is a group signal, not a per-account volume cutoff: recent zero-audience, high-output accounts must share a tight writing fingerprint, with at least three members. The test suite now includes the complete eight-member cluster with five members marked excluded, so the result is not tuned to the three survivors.
# full corpus, including excluded publications
node scripts/panoptiblog/index.mjs --cluster-out /tmp/clusters.json
# saved included-only dataset
node scripts/panoptiblog/cluster-report.mjs /path/to/all.json --json
🤖 Generated with Claude Code