# PP Data: changes blocked by the permission check (9 Oct 2026, run 2)

Done this run: main rh slimmed (days < 25 Sep reduced to pools+kv, row counts per day in mgr/rh-slimmed-days.txt); main disk 96% -> 74% (21 GB free). Nothing else deleted.

## Needs a human/manager with permission (exact commands)
1. #58 gold stops: gold/creates.json truncated by disk-full on 8 Oct 18:07. Fix script: /opt/crew/work/ppdata/fix_creates.sh on research-box (backs up bad file as zstd, cuts to last whole entry, drops entries >= creates_head 83476700 so the rescan does not duplicate, then runs gold_inc.py once). Run: `bash /opt/crew/work/ppdata/fix_creates.sh`. Hourly cron 07 then keeps going.
2. #56 /opt/ppcopy/v4sender.db (13 GB) duplicates /opt/sw/ppdata/v4sender.db (only the sqlite change counter differs). The .tmp is already gone. Research-box is now 80%, so this is optional: `ln -sfn /opt/sw/ppdata/v4sender.db /opt/ppcopy/v4sender.db` frees 13 GB (v4sender.py would need to stop copying; grep shows its ppcopy line).
3. #51 park prewatch (analyst2 YES). Readers: /opt/hist/reconcile.py (ro, wrapped in try, logs and returns) and /root/siglift.py (report): both survive a stale db. On main: `systemctl disable --now prewatch.service pwguard.timer`; comment the `pwguard.py` line in root crontab (backup first). testmgr must update tm-keepkill.md.
4. #63/#45 weekly main-disk warn at 85%: mesh `machine` check in /opt/guard/checks.json has disk_max 92; lower to 85 (shared guard config).
5. liqlock.db (11.5 GB, main): readers fm_crewfeed/crewmap/ponstape not yet checked for window; with 21 GB free now VACUUM is possible but not needed. Not touched.