Reset the canonical tree to six composites — reach, output, consistency, style, pulse, and connectivity — with initial weights that are retuned in the weight PR at the top of the stack. Zero is now the 0th percentile, arbitrary scoring caps are removed, link density is per kiloword, and word rarity uses mean Zipf from bundled FrequencyWords lists. The ranked cohort is English-only; non-English publications are indexed and shown in the filter counts with the caveat that the current tools are English-specific.
Author profiles are fetched for attribution, but author scores, ranks, labels, and the accounts leaderboard are removed. The snapshot is regenerated, the score tree is version 7, and the audit history records the change. The scheduled job no longer writes labeler data; the author-stats labeler remains a later release.
Scores and ranks change by design. The filter dialog now explains the language exclusion and the opt-out through the publication's standard preferences.
🤖 Generated with Claude Code