✅ Actions taken (Sam-approved 2026-06-01)

Sam approved 1A + empty trash, 2A, 3A, 4A, and Q5 "delete empty folders always." Executed same session:

• Removed 2,094 verified non-keeper duplicates from the five safe-tier folders = 4.36 GB freed. Each re-hashed (MD5) before deletion; 8 changed files and 4 with missing keepers auto-skipped; 0 errors. Rollback manifest at ~/Documents/Codex/monthly-dedupe/rollback-manifest-2026-06-01.csv (restore = copy the surviving keeper path back). Keepers never deleted, so an identical copy of every file survives.
• Removed 535 empty directories (rmdir-only, no data loss), including the empty Step-0 staging folder _archive_design-system-upload-2026-05-28.
• 2A: monthly_dedupe_run.py is now the canonical monthly script; dedupe_scan.py kept for the weekly job. 3A: Stray-Videos + Images to Check Out tagged as staging, not excluded. 4A: scan stays local. Q5: empty-folder removal is now a standing apply-phase rule. All recorded in ~/Documents/Codex/scripts/dedupe/CANONICAL.md.
• Disk: 126 GiB → 148 GiB free.

One open follow-up (not blocking): the scheduler-stored task prompt still has the original read-only/Step-0 wording; behavior decisions now live in CANONICAL.md. Offered to rewrite the stored prompt to match if Sam wants.

Importance: Medium. Task state: PENDING (read-only run complete; staging + 5 QA questions await Sam's approval). Run mode: READ-ONLY. No files moved or deleted.

What was done

Ran the monthly home-folder duplicate scan on the live Mac filesystem via Desktop Commander host shell. Walked ~/Downloads, ~/Documents, ~/Desktop, and home-root loose files with a 100 KB floor and MD5-after-size-collision hashing. Excluded Library, .git, node_modules, Parallels, Repos, Codex, SAIL-Logs, go, sam-wp, printing-press, and iCloud Drive (Archive)*. Wrote the full JSON report to ~/Documents/dedupe-report-2026-06-01.json and a markdown report to ~/Documents/Codex/monthly-dedupe/monthly-dedupe-2026-06-01-report.md.

Why the session ended

Completed normally. The scan is read-only by design, so it stops at a proposal. It found > 5 GB reclaimable, which per the task requires a Sam decision before any staging.

Results

Disk before run: 126 GiB free, 16% used. Reclaimable by type: PDF 2,616 MB, PNG 2,206 MB, JPG 1,671 MB, MOV 468 MB, DOCX 431 MB.

Step 0 — stale staging folders

Three candidates in ~/Downloads, all under 14 days, so no iMessage/Teams alert fired: _archive_design-system-upload-2026-05-28 (3d, empty), _index_scratch_2026-05-20b (11d, 768 K), _recovery-not-design-2026-05-28-1851 (3d, 210 M). Note: _index_scratch_2026-05-20b trips the 14-day alert around 2026-06-03.

Key patterns (systemic)

1. Review/staging mirror folders are the dominant source: Design Elements - Current/Duplicates, a- Review/from-Downloads, Images-Other/from-Downloads, Images-Other/from-SAIL-Images, and Downloads/from-Downloads hold ~4.4 GB of byte-identical copies of originals. Looks like past cleanup passes copied instead of moved.