The first filtered audit (2026-10-22, against the saved global index) still counted thousands of repos that aren't real Tangled activity. Each was reviewed against the owner's rule: exclude mirrors, forks, backups or copies, and timer filler; keep agent-written work that is real change. - Forks Tangled doesn't flag, found by matching their commits on GitHub (Linux and Android kernels, Zed, Rockbox, coreutils, mpv and others). - Mirrors whose newest commits are also on GitHub, and 589 repos whose own description says mirror, fork or backup (most "[READ-ONLY] Mirror of ..."). - Rewritten copies that deduplication can't merge: lichen.page.review.backup, core.archive, overby.me's split repos, monorepo aggregators. - Jeffrey Ollie's repos, whose READMEs name his Forgejo as canonical. - Two timer repos (stats refreshes, hourly status snapshots). Each DID is tagged fork, mirror, copy, timer or synthetic, under a dated section; the evidence per group is in the pull request. With these, the 30-day count is 78,888 instead of 131,089 unfiltered. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Change-Id: I3aea4d5546b88908bf4b762cec28fc0698675d25