Night 2026-10-18: register (175 -> 173; closed R-981 R-1142 R-1063 R-1069 R-1143 R-994; opened R-1144..R-1147), STATUS morning list, golden 0.329.0 record, evidence, report

Golden 0.329.0 baked at the floor's version, NOT vouched. Hub 0.158.0 built, NOT deployed.
Sunday rotation: no duplicate bake of an unvouched golden at the floor's version.

documentation/backlog/CLOSED-ITEMS.md

@@ -26,6 +26,17 @@
 
 ---
 
+## 2026-10-18 — each app's own folder, no allow-list exception, drive retirement order, plain refusals, `:latest`, Immich (controller v0.329.0, catalog `d928932` `fc93846` `8cf03e0`)
+
+| Row | What | Closed | Evidence |
+|---|---|---|---|
+| **R-981** | **The homepage app gets the Docker socket (that is root in the guest) and has no login, no setup gate and no family gate.** (P3) | CLOSED 2026-10-18 — **controller v0.329.0 + catalog `d928932`**: the socket mount is gone from the template; no Felhom feature used it (the rendered `docker.yaml` empty, no `homepage.*` labels on the box); the box's allow-list is now EMPTY, so the socket is refused for every app. Live on demo-hp 9201: homepage opens (200 in the container, through traefik and on its public name), its container has no socket mount, 0 allow-list refusals on the three test boxes. The throwaway homepage was removed after. | `audits/night-2026-10-18/P1/r981-homepage-before.txt`, `r981-homepage-after-9201.txt`, `redproof-privpolicy.txt`, `P2/box-walk.txt` |
+| **R-1142** | **The catalog allow-list (R-971, controller v0.328.0) lets one app bind ANOTHER app's folder under `${HDD_PATH}/appdata/`.** (P3) | CLOSED 2026-10-18 — **controller v0.329.0 + catalog `d928932`/`fc93846`**: the `${HDD_PATH}` root is `appdata/<the app's own folder>/`; the folder map is compiled in with the allow-list (`appdata_folders`: paperless-ngx → paperless — the only slug ≠ folder in every template, the drill catalog and the three test boxes). The security review of the pushed commit found two more doors, closed before the release: an app NAMED like another app's mapped folder gets no root, and `USERDATA_PATH` is always derived, never the app's env. The catalog gate mirrors it. 0 refusals on the three test boxes. | `audits/night-2026-10-18/P1/redproof-privpolicy.txt`, `redproof-userdata-env.txt`, `redproof-catalog-gate-r1142.txt`, `P2/box-walk.txt` |
+| **R-1063** | **„Végleges leszerelés" leaves a half-state when the agent refuses: the dashboard marks the drive decommissioned first and never undoes it.** (P3) | CLOSED 2026-10-18 — **controller v0.329.0**: the agent is asked FIRST; the mark and the default move only after its yes. Its 403/409 come back typed and reach the household as a plain 409 (HU + EN: nothing changed, what to do). Proven with a fake agent over TLS and the web flow (red-proofed: the old order marks the drive and moves the default). **Not pressed live:** demo-hp's only user-data drive carries its apps and is not a backup target, so the agent would say yes. Follow-up: R-1144 (the „anyway" mode still stops the drive's apps before asking). | `audits/night-2026-10-18/P1/redproof-r1063.txt` |
+| **R-1069** | **A disk passed into a Felhom VM on a virtio bus cannot be formatted, and the household reads the agent's raw English refusal.** (P3) | CLOSED 2026-10-18 — **controller v0.329.0**: the agent's „device has no durable id" refusal (409) is `agentapi.ErrNoDurableID`; the format endpoint, the drive init and the wipe say in HU + EN that the dashboard cannot format this disk, the disk did not change, and the operator or whoever set up the virtual machine can give it a serial number (or a SATA bus). The agent's safety rule is unchanged. Unit-proven (red-proofed: the raw agent text reached the household); not re-run on an OMV VM tonight. | `audits/night-2026-10-18/P1/redproof-r1069.txt` |
+| **R-1143** | **The controller `build.sh` pushes `:latest` with EVERY build, release candidates included** (P4) | CLOSED 2026-10-18 — **controller v0.329.0**: `build.sh` adds `:latest` only for an x.y.z tag (one `TAGS` array feeds build, push and multiarch; `--print-tags`). Live: 0.329.0-rc.1 and rc.2 were pushed without a `:latest` line; 0.329.0 moved it (same digest `aee51f3a…`). `test_build_push_once.sh` 16 cases, red-proofed. | `audits/night-2026-10-18/P1/redproof-r1143.txt` |
+| **R-994** | **Seven apps peak above 80 % of a memory limit, and 23 apps' typical use is above their catalog `mem_request`.** (P3) | CLOSED 2026-10-18 — the last part, Immich, **catalog `8cf03e0`**: a 4,000-photo synthetic import on 9202 killed `immich-server` twice at 2048M (kernel memcg kill, Docker `OOMKilled=false`); anon peak 1,596 MB (77 %); ML 1,175 of 1536M (76 %). Limits now server 3072M, ML 2048M (`mem_limit` 6016M); `mem_request` stays 2048M (typical 1,888 MB), so the install check's verdict on an 8 GB box is unchanged. The other six were raised 2026-10-17; the 23 requests are R-1125 (done). The public estimator's Immich figure: R-1147. | `audits/night-2026-10-18/P6/` (peaks.md, memory-phase1/2.tsv, oom-kill-dmesg.txt, teardown.txt) |
+
 ## 2026-10-17 — the catalog allow-list, break-glass rotation (controller v0.328.0, hub v0.157.0, catalog `07d1ef7`)
 
 | Row | What | Closed | Evidence |

documentation/backlog/OPEN-ITEMS.md

@@ -124,20 +124,17 @@ stopping line that lies.
 | **R-494** | Install & onboarding | P4 | **NARROWED 2026-09-14 by operator ruling → [P3-LOW] the hub COULD create the tunnel at customer creation, for a domain already on Cloudflare. Not blocking: every customer has their own domain and the operator creates the tunnel per day-0 A.1 (`architecture/01-topology-and-trust.md`).** *Original finding, kept:* **[P1-HIGH] A new customer's dashboard has NO reachable address unless the operator hand-makes a Cloudflare tunnel — the link in the setup-code mail is dead.** MEASURED 2026-09-14 on a fresh install from the public ISO (drill intervention **I1**): the claim mail points at `https://felhom.drill0242.felhom.eu`; that name has **no A and no AAAA** record (`dig @1.1.1.1`, control `felhom.enkisfelhom.hu` resolves); the hub has **no tunnel- or DNS-creation code** (`hub/internal/cloudflare/` holds only geo-rule removal; `cf_tunnel_token` is a pasted, optional form field, `configs.go:1478`) — day-0 runbook A.1 makes it a manual Cloudflare-dashboard step that nothing on the customer-create page asks for; the box's own split-horizon resolver on the appliance LAN IP answered `google.com` but not the dashboard name at 13:27:39Z; the agent applied the record at **13:27:44Z** (`lanresolver: applied split-horizon record … ip=192.168.0.158`, 3 m 46 s after the controller started), so the box CAN answer the name — **but only to a device that uses the box as its DNS server, and no document, screen or mail tells a household to do that**; the router and the installer-offered DNS answer nothing. The page was reachable only at the guest's LAN address with the name forced (`curl --resolve …:443:192.168.0.158`). **A volunteer could not have done that.** **What it needs:** an operator ruling — the hub creates the tunnel and DNS at customer creation, or the product gives a household a LAN address that works with no DNS change. | **READY — rank P3-LOW; owner: CC** **Re-ranked 2026-10-03: P3→P4: operator-side automation; the operator creates the tunnel by hand per the day-0 runbook.** | — | — | CC |
 | **R-504** | Install & onboarding | P4 | **[P3-LOW] `iso.felhom.eu` cannot show an index page on its own — its root returns 404, and the download page lives on the website instead.** MEASURED 2026-09-14: `https://iso.felhom.eu/` and `/index.html` → 404; only named objects answer. The host is an R2 bucket behind a custom domain; whether R2 would serve an uploaded `index.html` at `/` was **not measured** (uploading anything to the public bucket is a publication). The ISO v1.27.0 task puts the Hungarian download page at `felhom.eu/letoltes` (published with the ISO, after the operator's yes). **Remaining:** a redirect from `iso.felhom.eu/` to that page needs a Cloudflare rule the session has no credential for. | **WAITING-ON-OPERATOR — rank P3-LOW; owner: operator (Cloudflare rule)** **Re-ranked 2026-10-03: P3→P4: households are sent to the website's download page; the bare address is cosmetic.** | — | — | operator |
 
-## Apps & catalog — 20 rows (P3 11, P4 9)
+## Apps & catalog — 18 rows (P3 9, P4 9)
 
 | ID | Category | Sev | What | State | Blocked on | Next action | Owner |
 |---|---|---|---|---|---|---|---|
 | **R-1091** | Apps & catalog | P3 | **Catalog copy that is wrong or stale at the pinned tag, found writing the guides — each needs an operator word or a live check, so none was changed tonight.** From source at each app's tag (evidence per app in `audits/app-guides-2026-10-15/B/`): (1) and (2) **DONE 2026-10-16** (operator decision D12: reword): calcom's use case now names Apple Calendar or a CalDAV calendar instead of Google/Outlook, and bentopdf's tagline and use case say the browser does the work and the file stays on the device (catalog CHANGELOG 2026-10-16); (3) **sabnzbd**, **nzbget** use_cases say Radarr/Sonarr/Jellyseerr hand them jobs — the box wires only qBittorrent and Transmission (`mediawire.go`), and a hand link needs a path mapping; (4) **onlyoffice** first_steps 3–4 tell the household to wire Nextcloud by hand — the integration toggle does it (`internal/integrations/onlyoffice_nextcloud.go`); (5) **glance**, **homepage** first_steps send the household into config files in a Docker volume no household tool reaches; (6) **lubelogger** starts with only the „Dashboard" tab and in miles (shipped `appsettings.json`), so first step 3 cannot be done as written; (7) **home-assistant** first_steps promise auto-discovered devices — the app runs on a Docker bridge, where mDNS/SSDP were measured deaf (spike R-6); (8) **homebox** and **gitea** `add_people` name menus that moved or read differently at the tag; **bookstack** `default_creds` still says `admin@admin.com / password`, which after_install replaces; **dawarich** names the API key's page „Settings" (source suggests the account page); **code-server** says „install your tools" with no sudo; **komga** names the discontinued Tachiyomi; **seerr** `docs_url` points at Overseerr; **wanderer**'s comment still says it cannot close sign-up (R-714 is fixed in the same file); (9) **nzbget** and **grafana** deploy-field descriptions say the password can be changed later — nzbget re-applies its env password at every start and grafana ignores the field after first start (R-1086 is the qBittorrent/Transmission side). | OPEN — filed 2026-10-15 | — | Live-check (5)–(7) and (9) on 9202; fix (3), (4), (8) as copy with the freeze re-admitted; re-run the guide label check for any app whose copy moves | CC |
-| **R-1142** | Apps & catalog | P3 | **The catalog allow-list (R-971, controller v0.328.0) lets one app bind ANOTHER app's folder under `${HDD_PATH}/appdata/`.** The data root is `appdata/` as a whole: a per-app rule needs a name map, because paperless-ngx keeps `appdata/paperless`. A catalog push could give app A app B's files (not the box: no privilege, the household's own data). | OPEN — filed 2026-10-17 | — | A per-app appdata name in `.felhom.yml` (or a map in the controller), then `appdata/<own>/` as the root; catalog gate the same | CC |
 | **R-1092** | Apps & catalog | P4 | **Three apps show no guide: plant-it and home-assistant carry one HIDDEN (`checked_version: ""`), emby has none.** plant-it: `lifecycle: abandoned`, and its pinned image does not exist upstream, so there is no screen to check labels against. home-assistant: its frontend follows the browser language and the Hungarian strings are fetched at build time, not kept in git, so the Hungarian labels cannot be checked against source (the draft is written; the shipped frontend in the image could check them). emby: its server's web UI is in no public source at 4.11.0.4. Evidence: `audits/app-guides-2026-10-15/B/`. The other 62 templates show a checked guide (controller ≥ 0.315.0). | OPEN — filed 2026-10-15 | — | home-assistant: check the Hungarian labels against the frontend files in the pinned image (read-only, on a box running it), then set checked_version; emby: label-check against the image's own web files the same way; plant-it: nothing while abandoned | CC |
-| **R-981** | Apps & catalog | P3 | **The homepage app gets the Docker socket (that is root in the guest) and has no login, no setup gate and no family gate.** Fix: remove the socket mount, or put homepage behind the family gate. Security review 2026-10-12 (`audits/security-review-2026-10-12/`); detail held off-git for the operator (`/mnt/5_hdd/felhom.eu/evidence/security-review-2026-10-12-PRIVATE/`). | OPEN | — | Change the catalog template; prove on 9202 | CC |
 | **R-676** | Apps & catalog | P3 | **[P3-LOW] Watch: immich's first start restarted 12 times — decision 28's crash-loop stop (6 in 10 min) would stop it.** From the 2026-09-17 chaos night (DB connection dropped during the first-start geocoding import on a 6 GB guest; it did not recover that night). No healthy app in any drill evidence restarts on a first start (1831 samples, 40 live containers), so the threshold stands; this row exists so the first immich install under v0.269.x is watched. `audits/night-2026-09-24/A3/40-first-start-restarts.txt` **2026-09-25 night (read from source, v0.271.0): a DEPLOY's first start is NOT covered by decision 28's suppression** — `Deploying` clears when `compose up -d` returns (`deploy.go` "Clear deploying flag"), and `ObserveUnhealthy` then samples the app; an automatic update's step, verify and undo ARE covered (`Updating`, pinned by `TestD28_NoCrashLoopStopDuringAnAutomaticStep`). So a first start that restarts ≥ 6 times in 10 min is stopped — which R-676 already accepts for a broken first start; a healthy slow first start would be stopped too. **-- 2026-09-30: the first-start restarts are explained.** immich's first-start geodata import OOM-kills its database at 512M on a guest with no swap (R-732, measured: 61–104 kills); the 2026-09-17 chaos-night case (DB connection dropped during the import on a 6 GB guest) fits it. Fixed in the catalog (`56c4888`, 768M). The watch itself (decision 28 on a DEPLOY's first start) is unchanged. | **OPEN — P3; owner: CC (watch)** | — | — | CC |
-| **R-994** | Apps & catalog | P3 | **Seven apps peak above 80 % of a memory limit, and 23 apps' typical use is above their catalog `mem_request`.** MEASURED 2026-10-13 (`documentation/audits/sizing-2026-10-13/SIZING.md` §3, `evidence/catalog-check.json`). Near a limit — risk: the container is stopped for using too much memory: docmost (container 475 of 512 MB), romm (703 of 768), termix (117 of 128), uptime-kuma (227 of 256, tester-1 telemetry), grafana (435 of 512), immich (`immich-server` 1,665 of 2,048 during a 200-photo import), tandoor (827 of 1,024). No container was killed during the bench. Under-stated request (typical ÷ request): calcom ×4.0, adventurelog ×3.6, termix ×3.4, claper ×2.5, grafana ×2.2, komga ×2.2, docmost ×2.1, recipe-importer ×2.1, uptime-kuma ×2.1, zipline ×2.0, and 13 more ≤ ×1.8. Templates NOT changed (read-only run). | **NARROWED 2026-10-17** — the six other limits raised (catalog `bff82f9`: docmost 768M, romm 1024M, termix 256M, uptime-kuma 384M, grafana 768M, tandoor 1280M; peaks now 46–69 %; live on 9201: docmost 805306368 bytes); the 23 requests → R-1125 (operator: wait). **Left: Immich** (`immich-server` 1,665 of 2,048 MB on a 200-photo import; tester-1's real library peaked at 2.9 GB) — raise after a measurement with a larger library. | — | Measure Immich with a larger library (bench), then its limit | CC |
 | **R-1018** | Apps & catalog | P3 | **The OnlyOffice links (FileBrowser, Nextcloud) have no recorded live proof since the move to Proxmox.** `controller/internal/integrations/` holds two handlers; no audit after the Proxmox move shows a document opened from FileBrowser or Nextcloud in OnlyOffice (searched 2026-10-14). The media-app links of `15` §6 will copy this code, including its revoke-on-stop rule (`lifecycle.go:10-13`). | **OPEN** — filed 2026-10-14 | — | One live walk on a Tier-0 box: install OnlyOffice + FileBrowser, open a document, save, read it back; then stop one app and record what the link does | CC |
 | **R-1028** | Apps & catalog | P3 | **Plex's claim code can run out during the install, and the box cannot recover it.** plex.tv's code lives 4 minutes from the moment the household copies it; the clock covers the wizard, the image pull (`plexinc/pms-docker` is 351 MB, measured) and the first start. Plex reads `PLEX_CLAIM` only on its first run, and the field is `locked_after_deploy` — so a late code leaves an unclaimed server and the only way back is remove + install again. Reported by the operator during the 2026-10-14 test; on Tester 1's line the pull+start took 8.5 s and the claim worked (`audits/media-all-drives-2026-10-14/``G-auth/G3-plex-claim.txt`), so it bites on a slow line or a slow hand. **K6, 2026-10-14 night (`audits/media-drive-events-2026-10-14/``K6-1-unclaimed-plex.txt`):** an unclaimed Plex starts with no code at all, and `POST /myplex/claim?token=` on the RUNNING server reaches plex.tv (a wrong code → 500 after plex.tv's 401). So the box can pull + start first and ask for the code after: no race. A real claim needs the household's code — the morning step on the throwaway `plex-k6` on Tester 1. Narrowed 2026-10-14 night. **PROVEN 2026-10-15 morning (`K6-2-claim-after-start.txt`):** the operator's real code, given to the RUNNING unclaimed `plex-k6` → 200, `claimed` 0→1, `PlexOnlineToken` written, and the open door closed (no token → 401). `ADVERTISE_IP` was unset and the claim did not need it; whether clients find the server without it was not measurable (the throwaway had no published port). **BUILT 2026-10-15 (v0.313.0):** `after_start` field — not asked at install; the Plex page asks for the code once Plex runs and claims the running server; the gate opens only when Plex says claimed. Proven up to the code on 9201 (G5). | **NARROWED** — the dashboard claim with a real code is the last step | a household Plex code | Claim through the Plex page with a real code (G5c); then close | CC |
 | **R-1029** | Apps & catalog | P3 | **Plex's folder picker shows the drives as identical „media” folders.** With one bind per drive at the same path (`/mnt/felhom-drives/<drive>/userdata/media`), the leaf of every library folder is `media`; the drive's name sits two levels up, so in Plex's picker the household sees several folders all called „media” (reported by the operator during the 2026-10-14 test on Tester 1). It conflicts with `15` §2 reason 2 (the same path everywhere). | **READY** — filed 2026-10-14 | the layout decision (STATUS D9) | Decide in `15` §2: the players (`:ro`, never linked) may get a readable path per drive (e.g. `/media/<drive label>`) without breaking the hardlink rule, which binds only the download/library apps | CC |
-| **R-1036** | Apps & catalog | P3 | **The apps page lags the catalog by two apps.** `app-catalog-felhom.eu/templates/` holds **64** apps; `/alkalmazasok` and `/en/apps` render **63** cards, one of which (`filebrowser`) is built in and not a catalog template — so **62 catalog apps are listed**. Missing: **`plant-it`** and **`wger`**, both in the catalog since 2026-10-06/08. Their website assets are ALREADY present (`plant-it-logo.svg` + 3 screenshots, `wger-logo.png` + 3 screenshots, renamed to the catalog slug by `a3507761`, 2026-10-08) — only the cards were never written, so this is not a new-asset job. Found 2026-10-14 while reconciling the app count for the „felhő tárhely” page, where four places said **60** and the rest said **62**. **Done in that session:** every stated count on the site set to **62** (`index.html`, `szolgaltatasok.html`, `en/services.html` were 60), which is what the apps page actually shows — so no visitor can count the cards and find the number wrong. **62 is still not the catalog's number.** **To close:** write the two cards on both apps pages (gate 15 keeps the two pages' app sets equal, so they move together), then raise every stated count 62 → 64 in the same commit. **The operator chose this split at the 2026-10-14 STOP** over doing it inside the SEO change. **2026-10-15 (R-1048):** Prowlarr's card added on both pages and every stated count 62 → 63 (the media group 19 → 20), so the site still states what it shows; the two owed cards would make it 65. **Note before writing them:** `wger` is `lifecycle: hidden` and `plant-it` `lifecycle: abandoned` — neither is offered for a new install, so a card would advertise an app the wizard refuses. | **OPEN** | — | write the `plant-it` + `wger` cards on both apps pages, then 62 → 64 everywhere, one commit | CC |
+| **R-1036** | Apps & catalog | P3 | **The apps page lags the catalog by two apps.** `app-catalog-felhom.eu/templates/` holds **64** apps; `/alkalmazasok` and `/en/apps` render **63** cards, one of which (`filebrowser`) is built in and not a catalog template — so **62 catalog apps are listed**. Missing: **`plant-it`** and **`wger`**, both in the catalog since 2026-10-06/08. Their website assets are ALREADY present (`plant-it-logo.svg` + 3 screenshots, `wger-logo.png` + 3 screenshots, renamed to the catalog slug by `a3507761`, 2026-10-08) — only the cards were never written, so this is not a new-asset job. Found 2026-10-14 while reconciling the app count for the „felhő tárhely” page, where four places said **60** and the rest said **62**. **Done in that session:** every stated count on the site set to **62** (`index.html`, `szolgaltatasok.html`, `en/services.html` were 60), which is what the apps page actually shows — so no visitor can count the cards and find the number wrong. **62 is still not the catalog's number.** **To close:** write the two cards on both apps pages (gate 15 keeps the two pages' app sets equal, so they move together), then raise every stated count 62 → 64 in the same commit. **The operator chose this split at the 2026-10-14 STOP** over doing it inside the SEO change. **2026-10-15 (R-1048):** Prowlarr's card added on both pages and every stated count 62 → 63 (the media group 19 → 20), so the site still states what it shows; the two owed cards would make it 65. **Note before writing them:** `wger` is `lifecycle: hidden` and `plant-it` `lifecycle: abandoned` — neither is offered for a new install, so a card would advertise an app the wizard refuses. | **WAITING-ON-OPERATOR 2026-10-18.** The night brief asked for the two cards; **not written**, because both apps are closed to new installs: `wger` is `lifecycle: hidden` and `plant-it` `lifecycle: abandoned`, so the install wizard refuses both and a card would offer an app a visitor cannot get. The site states what it shows (63 cards, 62 catalog apps listed). | operator decision | Pick: (A) leave both off the site and close this row (recommended: the site shows only what can be installed), or (B) re-open them for install first (wger's throttle R-752, plant-it's upstream), then write the cards | operator |
 | **R-1071** | Apps & catalog | P3 | **Jellyfin cannot use films that sit on a NAS share (the pilot's OMV), and the pilot's wish is exactly that.** MEASURED 2026-10-15: an OMV SMB share added as „Hálózati tárhely” mounts and verifies (`/mnt/felhom-drives/omv-filmek`) — but only with a user name AND a WRITABLE share (read-only → refused „az alkalmazások nem tudnak rá írni”) — and the household's „Újraindítás most” binds only local drives into Jellyfin (`per-drive binds 1 [/mnt/felhom-drives/filmlemez/userdata/media]`). By design: R-108 (a NAS is browse-only, never an app's namespace); `media_library.go` skips `Kind == "network"`. A design decision, not a defect (R-370). Evidence: `audits/omv-vm-2026-10-15/B6-omv-share.txt`. | WAITING-ON-OPERATOR | Operator: should a media app read a NAS share read-only (a new bind kind), or is the answer „put the films on a drive the box owns”? | Operator rules; if yes, a design note in `15-media-library.md` first. | Operator |
 | **R-76** | Apps & catalog | P4 | **FileBrowser-created folders break the setgid chain, and a drop-zone's mode is not stable** **MIGRATED FROM `ROADMAP.md` 2026-08-22 (R-369) — originally filed 2026-07-26, size S, roadmap state `idea (surfaced by the R-75 spike, 2026-07-26)`.** Moved verbatim; nothing added or reinterpreted. The roadmap keeps its copy as history, marked moved. | **OPEN — migrated from ROADMAP 2026-08-22, rank unchanged** **2026-10-06 night: not a catalog fix** — the setgid chain is set by the controller's FileBrowser setup; a controller row. | — | Two related findings from `audits/SPIKE-catalog-data-paths-2026-07-26.md` P3/P5, both **pre-existing** and deliberately left alone by that spike. **(a)** FileBrowser Quantum 1.3.3 creates files `0644` and folders `0755` and does **not** propagate the setgid bit — even though the entrypoint wrapper's `umask 002` really is in effect (`/proc/1/status` `Umask: 0002`). Group inheritance itself works (a file uploaded into a 2775 group-100 dir landed group 100, not the process gid 1000), so the convention's *group* half holds and only its *mode* half is lost. The consequence is proven with a control: inside a UI-created `0755` folder a gid-1000 process's file landed group **1000**, while the identical write into the 2775 parent landed group **100**. So **any folder a customer creates through FileBrowser breaks the shared-group chain one level down.** Latent today — every userdata-touching catalog app that declares an identity declares uid/gid **1000**, the same uid FileBrowser runs as, so owner permissions mask it; it bites the day a content app runs as a different non-root uid with gid 1000. The comment at `infra/infra.go:156` is right that the image ignores `-e UMASK` but does not say t | CC |
 | **R-707** | Apps & catalog | P4 | **[P2] 37 apps still start with a login a stranger can take (`09` §3 decision 45).** Audit of all 53 apps: `app-catalog-felhom.eu/FIRST-ADMIN.md` (class, fix route, status, measured or read). Open: **3 hard-coded defaults** — calibre-web (`admin / admin123`, measured working on demo-hp and 9202; its own `cps.py -s` route needs a generated password WITH a special character — our generator is letters+digits, a controller change), mealie (`changeme@example.com / MyPassword`), wger (`admin / adminadmin`); **34 open first-run screens** (the first visitor creates the admin: actualbudget, adventurelog, audiobookshelf, calcom, docmost, emby, ghost, gitea, gramps-web, home-assistant, homebox, immich, jellyfin, komga, n8n, navidrome, opengist, outline, papra, plant-it, radarr, rallly, recipe-importer, romm, seerr, sonarr, sparkyfitness, tandoor, termix, uptime-kuma, vikunja, wanderer, wishlist, zipline). **Stale notes:** romm's `default_creds` `admin / admin` answers 401 on demo-hp (like a wrong password) — the page now warns with a login that does not exist; zipline's looks stale too. **Measured on demo-hp 2026-09-28 (read-only):** bookstack's default still logs in on the INSTALLED app (the fix is for new installs; the page now warns). Each fix: route (a) env or (b) the app's own CLI/API via `after_install:`, proven on 9202 with the default failing and the generated password working; route (c) a page sentence. Several sessions (operator, 2026-09-28). **2026-09-29 (controller v0.280.0, catalog `d0e7e2e`):** every class-3 app fixed — mealie, wger, calibre-web by `after_install` (calibre-web with the new `password:24:special`), proven on 9202 fresh installs (`audits/login-gate-2026-09-29/D/`); the setup gate (decision 46, spike PASSED) built and live on immich, n8n, audiobookshelf (probes measured) and uptime-kuma (button) (`…/C/`); romm's and zipline's stale notes removed. **Left: 30 class-4 apps** — gate each (probe measured on 9202 where one exists — 11 upstream candidates listed in `…/B/B-VERDICT.md` §3; the button otherwise). **2026-09-29 afternoon (controller v0.281.0, catalog `6faf432`):** 28 more class-4 apps gated — 32 of 34 — each proven on 9202 (`audits/gate-rollout-2026-09-29/`B): stranger → gate page / 401, household reached the first-setup screen, the gate opened (9 by a measured probe, the rest by the press), the app answered after. seerr, outline, rallly: gated, their opening needs a media server / e-mail (not proven). **Left:** wanderer (R-714); plant-it is not installable. | **NARROWED** (2026-10-03 triage: the row's verdict was finished, but it names open work no other row carries — seerr, outline and rallly are gated, but the gate OPENING is not proven (needs a media server / e-mail)) — **CLOSED — 2026-09-29 (the rest → R-714)** | — | — | CC |
@@ -148,6 +145,7 @@ stopping line that lies.
 | **R-1100** | Apps & catalog | P3 | **One sign-in for the dashboard and FileBrowser (proxy auth) — spike first.** Since controller 0.320.0 every app with a folder has „Fájlok ↗", and the box can show FileBrowser's password — but FileBrowser still asks for its OWN login the first time, so the household signs in twice. FileBrowser Quantum has a proxy-auth method (`auth.methods.proxyAuth`, header-trusted) that could take the dashboard session through traefik; nobody has measured it. Design: `architecture/02-controller-module-map.md` „Logins and folders". | OPEN — filed 2026-10-16 (task logins-folders). | — | Spike on demo-hp: proxy auth at the pinned tag (source + measured), a forwardAuth in traefik that answers only for a dashboard session, what a stranger reaching `files.<domain>` with a forged header gets, and what happens to FileBrowser's shares. | CC |
 | **R-1099** | Apps & catalog | P4 | **The N100 may hold RomM games in the old upload place, now shown under a fake „roms” platform.** Until 2026-10-16 RomM's own upload wrote to the drive's `roms/roms/<platform>/`; after the mapping fix (catalog `fb18471`, R-1089) those files are at RomM's `library/roms/roms/…`. Measured on 9201: such a game showed as missing in its platform and again under a platform „roms” (`audits/romm-ghost-2026-10-16/B/b-after.txt`). Files put in with the file manager are not affected — they are now found. The N100 runs RomM (hub `/apps/romm`); CC does not log in to the lived-in box. | OPEN — filed 2026-10-16 | Operator: look at `roms/roms/` on the N100's drive in the file manager | If it holds platform folders: move them up one level into `roms/`, then run „Scan” in RomM; if it is empty, close | Operator |
 | **R-1117** | Apps & catalog | P4 | **Immich's and Nextcloud's Redis keep a volume they do not need** (upstream runs both without one; queues and caches only). Deferred from R-1096 on purpose: removing a volume from the compose leaves the old volume on installed boxes, and the nightly dump / an old backup's restore for a volume the compose no longer declares is unchecked. The RDB switch (catalog `5d9450b`) already removed the full-disk danger. | DEFERRED — filed 2026-10-17 | — | Measure the dump and an old-backup restore without the volume on 9202; then the template change + an orphan-volume note | CC |
+| **R-1147** | Apps & catalog | P3 | **The public machine estimator (`/gepigeny`) sizes Immich from a 200-photo bench; a 4,000-photo import needs more.** The estimator's Immich peak is 3,290 MB (`audits/sizing-2026-10-13/`); tonight's import on 9202 (`audits/night-2026-10-18/P6/`) peaked at 2,848 MB for the server alone with room (anon 1,596), plus 1,498 ML, 359 postgres, 43 redis — and killed the server twice at the old 2 GB limit. The limits were raised (catalog `8cf03e0`); the public figures were not, because they follow an operator ruling („raised only, 2026-10-14"). | OPEN — filed 2026-10-18 | operator: may the public figures move? | Re-run the estimator inputs with the 4,000-photo figures, show the operator the before/after verdicts, then publish | CC |
 
 
 ## App updates — 4 rows (P3 4)
@@ -167,7 +165,7 @@ stopping line that lies.
 | **R-973** | Backup & restore | P3 | **A compromised controller can make the whole-guest backup run again and again, and each run prunes older restore points: the local tier directly, and the off-site tier through the store's own nightly keep-last-2.** So `07` §8 row 10 (ransomware) is weaker than it reads. The second reader confirmed the local half; the off-site half is inferred. Fix: a cadence or daily cap on the local-API backup route, and off-site snapshots that a nightly prune cannot remove. Security review 2026-10-12 (`audits/security-review-2026-10-12/`); detail held off-git for the operator (`/mnt/5_hdd/felhom.eu/evidence/security-review-2026-10-12-PRIVATE/`). | OPEN | — | Add the cap with a test; decide the off-site retention shape | CC |
 | **R-232** | Backup & restore | P2 | **DooPlex's backup makes every copy inside the same box — and nothing tells anyone when it fails.** Surveyed read-only 2026-08-06 (`audits/RECON-dooplex-backup-2026-08-06.md`). **What works:** five sets, 14/14 successful runs in 14 days; a file was restored from the `data` repo and matched the live original **byte for byte**; every set except two is cross-disk; k3s is integrity-checked on every run. **What the matrix exposes, ranked:** (a) **`notify_failure` is a no-op** — `NOTIFY_ON_FAILURE=true` but `NOTIFY_WEBHOOK_URL` is commented out, so a failed backup notifies **nobody**; the project already has a working Resend path that CI uses. Cheapest item, and it makes every other failure visible. (b) **Nothing leaves the box** — no rclone, no remote repo, no off-site target anywhere; Longhorn's target is `nfs://192.168.0.180:` pointing at DooPlex itself, and the only outbound-looking cron pulls *inbound* from Hetzner for a different project. The machine that runs the hub managing the customers' off-site chain has no off-site copy of its own. (c) **The backup tree is a single writable path** and the restic repos are not append-only — one bad script or ransomware destroys every copy at once. (d) **Two same-disk sets**: `.claude-memory` and the PostgreSQL dumps, whose source directory sits *inside* the backup tree. (e) **Longhorn `retain=1`** — one generation per volume, so a corruption noticed a day late has no earlier copy. (f) **`/opt/backup/docs/BACKUP-RESTORE.md` does not exist** though the systemd unit advertises it. (g) **`secrets/restic-repo` has never held a snapshot** — `backup-secrets.sh` contains no `restic` call; the secrets are GPG files on `sda1` only. (h) **No restore has ever been run** beyond today's single-file probe — the matrix's "ever demonstrated?" column is otherwise entirely empty. **Not a finding:** the restic passphrase. The on-box copy is on `sdb1`, a different disk from the backups, and the **operator holds an offline copy out of band** — so a disk loss is recoverable. The narrow residual is that it is operator-held rather than system-held, unlike the customer case's hub-vaulted escrow, so it should be confirmed current and findable by someone else. **Nothing was changed by the recon.** | **NARROWED 2026-10-08 — owner Viktor.** (a) DONE with the operator's yes in chat: `notify_failure` now mails admin@felhom.eu through Resend; proven by one test mail that reached the inbox (`audits/day-2026-10-08/r232/`; no backup was started). (b) partly: the hub database leaves DooPlex nightly to ep0 (R-173); everything else stays on the box. (c)–(h) unchanged. **2026-10-09 (operator yes in chat): (b) and (h) DONE for Gitea and the secrets.** Nightly 00:20 `felhom-dooplex-offsite` pushes Gitea (614 MB of repositories, the `gitea` database dump taken first, `app.ini`) and the nightly GPG secrets export, encrypted with a NEW key, to ep0 `operator` (`host/dooplex-gitea`) on the hub-DB write-only token — no change on ep0; first push 57 s; restore test weekly (manifest, `git fsck` every repo, `pg_restore --list`), run once: 27 805 files, 10 repos OK; `DooplexGiteaOffsiteStale`/`DooplexGiteaRestoreTestStale` in force (`absent()` included, promtool + 2 red-proofs); every backup unit now mails on failure (`OnFailure=`, dry run reached the inbox). **First real Gitea restore:** into a throwaway on the bench, 10/10 repos, product `main` = live, a file byte for byte, a login (`runbooks/gitea-restore.md`). The hub came back too (R-173, closed). **The registry is left out on purpose:** its images rebuild from the code. **Correction to the recon §7:** the API mirror holds ONE repo (`homelab-manifests`), not all — Gitea had a single copy, on DooPlex. `audits/dooplex-survival-2026-10-09/` **LEFT:** (c) append-only for the on-box tree, (d)–(g) as written; the paper copy of the new key (operator). **READY** for the rest | — | — | operator |
 | **R-893** | Backup & restore | P3 | **After a failed OFF-SITE replay, the rollback pours the NEWER pre-restore copy over the OLDER volume just put back.** Read in source 2026-10-06 (R-638 option A, not measured): `internal/backup/offbox_reconstitute.go` writes the undo copy from the live (newer) database, replaces the volumes with the snapshot's older tars, then — when the replay fails — `rollbackSafetyDump` loads that newer dump over the older database volume. The loader only drops what the dump knows, so tables the newer migration removed stay; and when the snapshot's older definition was written, the rollback branch does not put the newer definition back, so the older app starts on rolled-back data; non-database volumes stay at the snapshot's state. An order change cannot fix it (the only undo is a logical dump, and its volume was replaced). Known limit in `07` §6.3. **-- 2026-10-08 design:** `audits/day-2026-10-08/design-R-893.md` — re-verified; and the screen's „your data is back as it was" (`err.backup.db_restore_failed_rolled_back`) is false for files, other volumes and the app version. Question D8 on STATUS's decision sheet. **-- 2026-10-08 14:16 operator ruling D8 (`09` §3 decision 192):** yes, both, in that order — first the app stays stopped for support (option C), next „put back exactly as it was" (option A). | **NARROWED — 2026-10-09: the first half DELIVERED in controller 0.304.0. Slice 0 (the live measurement) NOT run: 9202 has no off-site, and a dump cannot be cut by hand inside an encrypted off-site copy on Tester 1; it needs a scratch off-site repo — a plan, not a finding.** **NARROWED 2026-10-08 — the first half (D8 option C, the app held stopped) is built on controller main and ships tomorrow; what stays open is the second half, „put back exactly as it was” (option A: a pre-restore copy of the app's volumes and placed files, a fit check, disk for one copy; the slice-0 measurement on 9202 first).** **OPEN — filed 2026-10-06** **2026-10-06 night: verified in source, no code** — `offbox_reconstitute.go:758` (undo dump from the live DB), `:778-860` (files and volumes from the snapshot), `:804` (the snapshot's definition is written when its version differs), `:898` (the rollback loads the newer dump over the older volume; nothing writes the live definition back). Not a reorder fix: it needs R-638 option B (a rebuilding loader) or a pre-restore volume copy (disk cost; R-685's class). Which state a household gets after a failed off-site replay is the operator's call. Next: the 9202 measurement, then the design. | a design: R-638 option B (a loader that rebuilds instead of overlays) or a pre-restore volume copy | Measure it once on 9202 (a forced replay failure after an off-site restore over a migrated app); then a design for the operator | CC |
-| **R-895** | Backup & restore | P2 | **The hub's clean-up-window check trusts the snapshot counts the box sends, so a broken-into box (or past-dated fakes added through the add-only key) can shrink the real off-site history without an alarm.** READ 2026-10-06 night in source (R-822's design): the before/after comparison uses counts the box itself reports (`hub/internal/offsitekeys/service.go:284`, `:343`); new fakes keep the count level. Decision 68 already accepts a box-trusted count. | **OPEN — filed 2026-10-06 night** **2026-10-07 07:58: kept open for later (`09` §3 decision 166).** **2026-10-17: READY — design updated with the measurement** (`audits/agent-allowlist-rotation-2026-10-17/P4/design-R-895.md`): the hub needs NO new credential (the sealed sub-account password it already uses daily; write-capable today); the provider shows upload times (2026-09-01 evidence on demo-hp's sub-account). Slice 1 in hub v0.157.0 (deployed 2026-10-17 on the operator's yes): the daily key audit also lists `snapshots/` read-only and LOGS count/newest/oldest — the first 07:10 run is the live measurement. | the first logged listings (hub 07:10 audit), compared with the box's own count | Option B of `audits/night-burndown-2026-10-06/design-R-822.md`: the hub lists the repo's `snapshots/` files over its own login before and after a window and alarms on snapshots no box run explains | CC |
+| **R-895** | Backup & restore | P2 | **The hub's clean-up-window check trusts the snapshot counts the box sends, so a broken-into box (or past-dated fakes added through the add-only key) can shrink the real off-site history without an alarm.** READ 2026-10-06 night in source (R-822's design): the before/after comparison uses counts the box itself reports (`hub/internal/offsitekeys/service.go:284`, `:343`); new fakes keep the count level. Decision 68 already accepts a box-trusted count. | **NARROWED 2026-10-18 — slice 2 built, not deployed** (hub v0.158.0): the 07:10 audit compares the `snapshots/` listing with yesterday's by name — files gone that no clean-up window explains → `offsite_snapshots_vanished` (error), files no reported box run explains → `offsite_snapshots_unexplained` (warning); operator-only, hub-origin (not in `allowedEventTypes`). First listing = baseline; a move-aside or RESET drops it; vanishing judged on a pinned tier only; files younger than 1 h judged the next day. Red-proofed twice (`snapshot_watch_test.go`). **Not built:** the window's before/after on the hub's own count (design §3.1) and refusing a window while unexplained files exist (§3.2). Unmeasured: the provider shell's `ls -l` time zone (± 3 h slack until then). | the operator's yes to deploy hub v0.158.0 | Deploy; read the first 07:10 lines (newest upload vs the box's last run = the time-zone calibration), shrink the slack; then decide §3.1/§3.2 | CC |
 | **R-923** | Backup & restore | P2 | **The password manager (Vaultwarden) runs on DooPlex and holds the keys that open DooPlex's own off-site copies — a loss of DooPlex loses the keys with the copies.** Operator finding, 2026-10-09 11:19: "the password manager also runs on DooPlex (Vaultwarden), so I probably need to back up its DB somewhere too." Measured: `vaultwarden-system/vaultwarden` (1.37.3), PVC `vaultwarden-data` 7.4 MB (`db.sqlite3` + WAL, `rsa_key.pem`, no attachments); its only backups were Longhorn on DooPlex. **A key in a password manager that runs on DooPlex is not off DooPlex** — R-173's "both keys are off DooPlex" (2026-10-05) and the hub-DB runbook Step 0 were wrong in that sense. Also seen: Vaultwarden warns its `ADMIN_TOKEN` is plain text (an Argon2 hash is its guidance). | **NARROWED 2026-10-09 (operator yes in chat):** Vaultwarden's own `backup` copy + `rsa_key.pem` ride the nightly encrypted DooPlex→ep0 job (`scripts/dooplex-offsite/`), checked on every push and every Sunday (integrity, users, items — rows only). That copy opens only with the DooPlex off-site key, and the vault only with the master password — so the circle is broken only once both are on paper outside DooPlex: `runbooks/break-glass-sheet.md`, `runbooks/total-loss-of-dooplex.md`. **LEFT:** the operator prints the sheet and stores it away from home; the `ADMIN_TOKEN` hash (a DooPlex change, operator's word). | operator: print the sheet | Operator: print + store the sheet; then close | operator |
 | **R-49** | Backup & restore | P3 | **[P2] The offsite capture set is ~90% cache and duplication — 1.1 GB of a 1.2 GB immich "photo backup".** Measured 2026-07-19: `immich_ml_cache.tar` **823 660 032 B (~60%)** — re-downloadable ML model weights; `immich_postgres_data.tar` **308 251 136 B (~23%)** — a raw tar of the postgres data dir that DUPLICATES the logical `.sql` dump captured beside it; `upload/backups/` **18 MB** — immich's own nightly dump, a backup inside the backup, growing daily; plus the stranded pre-v3 `dccc13fe…` tree (~36 MB) no DB has ever referenced. Actual irreplaceable content: **72 MB of originals**. **MIGRATED FROM `ROADMAP.md` 2026-08-22 (R-369) — originally filed 2026-07-19, size S–M, roadmap state `idea`.** Moved verbatim; nothing added or reinterpreted. The roadmap keeps its copy as history, marked moved. | **OPEN — migrated from ROADMAP 2026-08-22, rank unchanged** **Re-ranked 2026-10-03: P2→P3: wasted space and transfer, no data risk; needs a capture-set ruling, not a sale blocker.** | — | **Evidence: `audits/DIAG-immich-restore-round2-2026-07-19.md` §4 (full byte breakdown).** This is the customer's offsite quota and transfer cost, and it lands on the Hetzner sub-account they are billed for. **Recorded, deliberately not changed** — a capture-set exclusion is a data-loss-shaped decision and gets its own ruling, not a drive-by edit. Candidates in priority order: (a) `immich_ml_cache` — pure cache, strongest case; (b) the `postgres_data` volume tar where a logical dump of the same DB is already captured (the dump is what the restore path actually replays); (c) `upload/backups/`. Likely generalises past immich into a template-classification rule about cache volumes and self-backup directories, so it should be specified against the catalog, not one app | CC |
 | **R-231** | Backup & restore | P3 | **`/opt/backup/scripts/` on DooPlex is unversioned host state** — found 2026-08-06 while adding the auto-memory store to the backup set. No repository tracks the scripts that protect the recovery chain, so the edit made that day (`CLAUDE_MEMORY_DIR` in `backup-config.sh`, multi-path restic call in `backup-data.sh`) exists only on the box. This is the same class the part-2 session was closing, found inside the fix for it; the change is transcribed in `felhom.eu/workspace/README.md` so it is at least *recorded*. **Two related facts, both understating current safety:** the backup destination (`/mnt/5_hdd/backup`) is on the **same physical disk** as the workspace it protects, and the DooPlex backup set has **no off-site leg** (`sync-hetzner-backups.sh` is jarrs.eu and pulls *from* Hetzner *to* DooPlex). Bringing a root-owned production backup script under version control, and deciding what installs it, is its own scoped change. | **VERIFY** — **MEASURED 2026-10-13 — the due check is GREEN** (read by the sizing session, two channels): `systemctl show dooplex-backup.service` → `Result=success`, `ExecMainStatus=0`, ran 03:27:39–03:33:29 CEST; its journal ends „All backup components completed successfully!”, and the manifests leg (the header-token change) logged „Successfully backed up homelab-manifests” with the plain URL `https://gitea.dooplex.hu/admin/homelab-manifests` (no token in it). Left open, not closed: the row also carries the same-disk and no-off-site facts, which closing would drop — the owner decides. **VERIFY** — **2026-10-12 evening: the token left the clone URL** (operator yes in chat; homelab-manifests `1943a70`, installed by the role): `backup-manifests.sh` sends it as an `http.extraHeader` through `GIT_CONFIG_*` env; read-only proof: `git ls-remote` of the private repo with the header → HEAD, without → refused. The 2026-10-13 03:10 run is the test of both changes. **VERIFY** — **2026-10-12: in git and installed from it** (operator yes in chat, operator in the session): homelab-manifests `fdd324b`, `homelab-ansible/roles/dooplex_backup/` (11 scripts + the timer units, byte-identical to the live ones; no secret in any file; `install.sh` dry run / `--apply`). The install changed only the units' mode (744 → 644). Baseline: last night's run `Result=success`, exit 0 (`audits/answers-2026-10-12/s20-backup-scripts.txt`). **Seen by the commit's security review, NOT changed (a DooPlex behaviour change needs the operator's word):** `backup-manifests.sh` puts the Gitea token into the clone URL for the mirror run (visible in the process list for those seconds) and strips it from the stored remote afterwards; no token VALUE is in any file, and the repo is private (anonymous API read = 404). A header-based fetch (`git -c http.extraHeader=…` from the token file) would remove it — one line in the decision list. **READY** — operator ruling 2026-10-12 (S20, `09` §3 decision 215): yes — the scripts into a repo + an install script; the operator present for the install. **READY** — owner Viktor | — | Owner: close the row (the install from git is proven) and file the same-disk / no-off-site facts as their own row if they still hold | CC |
@@ -192,15 +190,13 @@ stopping line that lies.
 | **R-32** | Backup & restore | P4 | **[P2-HIGH] RESET must purge the customer base dir; the orphan card must stay honest; unattributed bytes must be visible.** The rehearsal's S7 said in advance that an orphan card would BE a finding — and one appeared (16:58:14). Cause: RESET's `"hetzner":"ok"` leg destroys the sub-account, but **a Hetzner sub-account is an access-control object, not a data object** — its directory survives, so re-enabling offsite recreated an account over the previous lifecycle's ciphertext, encrypted under a key that same RESET had destroyed. **MIGRATED FROM `ROADMAP.md` 2026-08-22 (R-369) — originally filed 2026-07-21, size M, roadmap state `idea`.** Moved verbatim; nothing added or reinterpreted. The roadmap keeps its copy as history, marked moved. | **OPEN — migrated from ROADMAP 2026-08-22, rank unchanged** **2026-10-06 night: design written** (`audits/night-burndown-2026-10-06/design-R-32.md`) — pick A: purge the repo + set-aside copies through the sub-account's own password login BEFORE deleting it (no main-account credential); the „repo data destroyed" comment at `hub/internal/offsite/offsite.go:276-280` is false for the shared tier; part (3) waits on a read-only measurement (`du` on port 23). Operator: agree to route A. **2026-10-07 07:58: `09` §3 decision 167 — option A yes, and delete the old RESET leftovers once.** **2026-10-07 (Part D step 2, read only):** the hub shows bx11 holding 3.3 GB of data while the four live customers account for ~0.6 GB → **~2.7 GB belongs to no live customer**; no live customer shows a set-aside folder, so the leftovers are in homes whose sub-account is gone, reachable only by the pool box's MAIN account. **The one-time clean-up is STOPPED:** no main-account login for bx11 is held on DooPlex (the API token in the credentials file sees no box). Needs: the main-account login, or the operator lists and deletes in the Hetzner console (`audits/day-2026-10-07/D/D2-listing-readonly.txt`). **2026-10-07: option A DELIVERED (hub 0.142.0):** RESET purges the repo and every `.orphaned-*` through the sub-account's own login before deleting it, and keeps the sub-account if the purge fails (red-proved, `audits/day-2026-10-07/D/`). Not yet seen live (no scratch RESET today). **The one-time clean-up of the ~2.7 GB of leftovers is STOPPED:** it needs the pool box's main-account login (see above). **2026-10-07 14:43 — `09` §3 decision 173: leave the ~2.7 GB of old leftovers for now; re-ranked P2 → P4.** Option A itself is delivered (hub 0.142.0). What stays open is ONLY the leftovers: they were computed (box total minus live customers), never listed, and deleting them needs the pool box's MAIN login, which reaches every customer's data. **Blocker: a safe main-account path.** | — | **Ruling from the run (three parts, deliberately separate):** (1) because RESET destroys custody, the ciphertext it leaves behind is unrecoverable **BY DESIGN** → RESET gains a **main-account purge of the customer base dir** (the existing operator ack already covers it); (2) the **move-aside guard STAYS** for reinstall-*without*-RESET — there custody survives and the card's "history recoverable" promise is true (R-26 depends on exactly that); (3) the operator **Restic tab shows per-customer directory bytes vs attributed snapshot bytes**, so dead data cannot hide. Measured on the pool box that night: **49 M attributed** (2 snapshots, 48.717 MiB) against **1.4 G + 3.0 M unattributed** across TWO `.orphaned-*` dirs. Evidence `restic-and-pool.txt` | CC |
 | **R-1087** | Backup & restore | P4 | **Whole-box copies left on the OLD local target are never pruned after the target moves.** MEASURED 2026-10-15 (storage-identity build B space check): the N100's host root (`local`) holds three copies from 2026-07-28 (5.5–5.7 GB each, ≈ 16.8 GB); its agent's `local_backup_target` is now `felhom-backup` (the USB disk, retention 3), and retention prunes only the current target. On a small-root box such leftovers would starve the local copies (R-548). Evidence: `audits/storage-identity-buildB-2026-10-15/space/felhom-pve-9201.txt`. | OPEN | — | Decide whether the agent prunes its own `vzdump-lxc-<vmid>-*` files on a former target (after a newer copy exists on the new one), or the target move asks; until then, the N100's three July files can be removed by hand with the operator's word. | CC |
 
-## Storage & devices — 12 rows (P3 6, P4 6)
+## Storage & devices — 11 rows (P3 5, P4 6)
 
 | ID | Category | Sev | What | State | Blocked on | Next action | Owner |
 |---|---|---|---|---|---|---|---|
-| **R-1069** | Storage & devices | P3 | **A disk passed into a Felhom VM on a virtio bus cannot be formatted, and the household reads the agent's raw English refusal.** MEASURED 2026-10-15 (OMV KVM plugin, „Add host disk”, default Bus `virtio`): `POST /api/disks/format` → 500 „agentapi: format: HTTP 409: format refused: device has no durable id to bind the format to (path-only formats are not permitted)”. A virtio-blk disk without a serial has no `/dev/disk/by-id` link inside the guest, so the anti-retarget guard refuses — correct and safe. The same disk on Bus `sata` got `byid:ata-QEMU_HARDDISK_QM00003` and formatted. Affects every hypervisor that passes disks as virtio without a serial (OMV KVM, Proxmox without `serial=`). The pilot guide says `sata`. Evidence: `audits/omv-vm-2026-10-15/B4c-second-disk-format.txt`. | OPEN | — | Translate the refusal for the household („Ezt a lemezt a Felhom nem tudja biztonságosan azonosítani — a virtuális gépben állítsd a lemez buszát SATA-ra”); optionally fall back to a filesystem-UUID binding after a blank-format. | CC |
 | **R-1073** | Storage & devices | P3 | **Felhom in a virtual machine sees no disk health: SMART is UNKNOWN for its own system disk and any passed disk.** MEASURED 2026-10-15 on the OMV-VM copy: every storage target reported `smart.health = UNKNOWN`, all counters null (the guest sees QEMU disks). On the pilot's box the NVMe and the „old, not stable” Toshiba would be watched by OMV only, and Felhom's drive-health alarms (R-330/R-332's whole point) cannot fire. Evidence: the host report in `audits/omv-vm-2026-10-15/README.md` (memory/disk section) and `C-hub-events-Tester4.txt`. | WAITING-ON-OPERATOR | Operator: is a VM install a supported shape, and if so, is „OMV watches SMART” acceptable or must Felhom say on the dashboard that it cannot watch the disks? | Operator rules; then CC adds a household line on the Meghajtók page when SMART is unavailable. | Operator |
 | **R-1093** | Storage & devices | P3 | **An EMPTY partition asks for the typed „data will be erased" confirmation.** MEASURED 2026-10-16 (build B, scratch VM, agent 0.158.1): `POST /api/storage/init` for a blank part → `needs_confirmation`, reason „device has a partition table" — `lsblk` reports the PARENT disk's table (PTTYPE) for a partition, so the agent's probe reads an empty part as data-bearing. Safe direction (more friction, never less), but `16` §6 says an empty part is self-serve and the wizard shows „adatot tartalmaz". Evidence: `audits/storage-identity-buildB-2026-10-15/live/12-partitions-format.txt`. | OPEN | — | In the agent's device probe, ignore a PTTYPE that a partition inherits from its parent (judge only the part's own signature and children); a red-proved test with a blank part fixture. Next agent release. | CC |
 | **R-1062** | Storage & devices | P4 | **The N100's USB drive cannot be split in two without losing the box's whole-guest backup target.** The operator asked (2026-10-15 16:31) for two halves for multi-drive tests. Read first (`audits/n100-lived-in-2026-10-15/B/`): no family files — 2.7 GB nightly vzdumps of guest 9201 (`dump/`, PVE storage `felhom-backup` = the whole-guest backup target), 3.1 GB app backups + an off-site restore copy, 422 MB leftover app data of removed apps, empty media folders. The agent refuses a decommission of a backup target (`refuseIfBackupTarget`) and the N100 has no other drive to move that role to. Nothing changed. | **WAITING-ON-OPERATOR** — filed 2026-10-15 | operator decision | Keep one drive (CC's pick: the lived-in box keeps its local whole-guest backup), or give a word to drop that backup leg / add a second drive first | operator |
-| **R-1063** | Storage & devices | P3 | **„Végleges leszerelés" leaves a half-state when the agent refuses: the dashboard marks the drive decommissioned first and never undoes it.** `handleStorageDecommission` → `finalizeDecommissionWith` calls `settings.SetDecommissioned` (and may move the default) BEFORE `agent.Decommission`; on the agent's 403 (protected role) or 409 (whole-guest backup target) it returns the error and keeps the mark — the dashboard would show a retired drive the agent kept serving. Read from code 2026-10-15 (`felhom-controller` `internal/web/storage_handlers.go`); NOT pressed on the lived-in N100 on purpose. | **OPEN** — filed 2026-10-15 | — | Measure on demo-hp (a backup-target drive), then: ask the agent first (or undo the mark and the default on a refusal); a test with a refusing agent seam, red first | CC |
 | **R-1064** | Storage & devices | P4 | **Removing a storage path does not mention the cross-drive backup copies it holds.** demo-hp 2026-10-15: `/mnt/sizing-bench` removed through `/settings/storage/remove` (302, „Adattároló eltávolítva") while calibre-web's and paperless-ngx's `cross_drive.destination_path` pointed at it; the copies stay on disk, the records still name the removed path. The in-use check counts HDD_PATH only (`appsUsingPath`). | **OPEN** — filed 2026-10-15 | — | Read what tonight's tier-2 run (01:30) does with the two records; then: name the copies in the removal's confirmation (or count them as in use) | CC |
 | **R-330** | Storage & devices | P3 | **Disk health Phase 2 — the three SMART attributes the wire does not carry.** The failing drive's most telling counter was **187 `Reported_Uncorrect`**, sitting at normalized **1** against threshold **0** with a raw count of **1001** — one point from failing and structurally unable to get there. Also wanted: **199 `UDMA_CRC_Error_Count`** (cabling) and **188 `Command_Timeout`**. None are on the agent→controller wire today, so v0.215.0's ladder could not use them. Phase 2 also persists periodic SMART **samples** (the right home is `metrics.MetricsStore`, NOT the Phase-1 state file, which is one record per disk and must stay that way). **This is a declared WIRE change, so under the G-1 gate the hub must model the new fields in the SAME session** — that is precisely why it was kept out of Phase 1, where it would have turned a one-word severity fix into a three-repo change | **READY** — **2026-10-12: the wire half is CLOSED in this row** — released in agent v0.150.0 (boxes run 0.154.0), controller and hub decode 187/188/199. **Only open work:** the verdict slice (use the three counters in the disk verdict and the household's words) and persisting samples. **READY (M) — NEW 2026-08-14** **2026-10-06 night: the wire half BUILT on main (unreleased):** the agent sends SMART 187 `reported_uncorrect`, 188 `command_timeout` (raw, vendor-packed on some drives — carried as reported), 199 `udma_crc_errors` (pointer + omitempty: unknown is absent, never 0); the controller decodes them and the hub's host-report mirror models them (G-1 wire gate green; it convicted the agent half alone). **Carried only** — no verdict, banner, mail or alarm reads them (`TestR330_CountersChangeNoVerdictYet`); using them changes what a household is told and is the next slice, with persisting samples. Red-proofs `audits/night-burndown-2026-10-06/r330/`. Ships with agent v0.150.0 and the next controller and hub releases. | R-328 (closed) | The verdict slice: 187/188/199 into the disk verdict with tests; then persist samples | CC |
 | **R-332** | Storage & devices | P3 | **The new Hiba-from-counters path has never fired on real hardware.** v0.215.0's whole point is a verdict the product could not previously reach, and it is proven only against the committed fixture's values in unit tests (12 scenario groups, 11 of 12 red-proofs failing as required). The live validation on demo-hp proved the **negative** — three healthy disks still read Rendben across the deploy, no false alert — and the **severity wire** end to end, but no live disk has actually reached Hiba. **This is the honest gap and it must not be closed by pointing at the fixture tests**: the drive that produced the fixture is in DooPlex, which is Tier 2 and never a drill target, and the demo boxes are all-flash and healthy | **WATCHING — NEW 2026-08-14, NARROWED same day.** One item originally in this gap is now PROVEN LIVE: the **persisted state surviving a controller restart**. The v0.215.0→v0.216.0 redeploy destroyed and rebuilt the container, and the new one read back a `changed_at` written by the PREVIOUS version (`2026-08-14T07:23:14.640216851Z`, still intact at 09:31:35Z) instead of re-baselining — Scenario L on real hardware, not just the production-path unit test. **What remains unproven is the verdict itself, plus the stronger restart half: an already-ALERTED disk not re-alerting** | a real degrading disk, or an injection harness | **Closing condition:** a live disk reaching Hiba from counters, OR a deliberate injection through the REAL pipeline (agent `/disks` → controller check → hub event), not a hand-set verdict | CC |
@@ -208,8 +204,9 @@ stopping line that lies.
 | **R-1076** | Storage & devices | P4 | **A cheap SSD reports a vendor-less WWN, so a WWN-first disk key may not be unique.** MEASURED 2026-10-15 on the N100's system SSD (AirDisk 512 GB): `ID_WWN=0x5000000123456f1f` — NAA-5 with OUI `000000`, a pattern that looks like a placeholder; its serial `QDF922W009654S30EX` looks real. The agent's wipe binding prefers `wwn-` links (A/internal/storage/durable_device.go:32). Not known whether two units share the WWN. Evidence: `audits/storage-identity-2026-10-15/B-n100.txt`. | OPEN | — | Check a second unit of the same model (or vendor data); if shared, make the agent skip OUI-000000 WWNs (the design already keys on model+serial, `architecture/16-storage-physical-identity.md` §3). | CC |
 | **R-1077** | Storage & devices | P4 | **No USB enclosure that hides or fakes a disk's serial has been measured.** The storage-identity spike (2026-10-15) had none at hand and found none in earlier evidence; the design's `usb_bridge` unknown rule (`architecture/16-storage-physical-identity.md` §3) is therefore an assumption. Cheap single- and dual-bay bridges are known to report one bridge serial for any disk. Evidence: `audits/storage-identity-2026-10-15/README.md` case 4b. | OPEN | — | Buy or borrow one cheap USB enclosure (single and dual bay), run `tools/physdisk_probe.py` against it, and set the rule from what it reports. | Operator (a device) then CC |
 | **R-1094** | Storage & devices | P4 | **The whole-disk format refusal tells the household to format the free part — which is refused while the other part is in use.** SEEN 2026-10-16 (controller 0.318.0): 409 „Ezen a lemezen egy használatban lévő meghajtó van, ezért a teljes lemez nem formázható. Csak a szabad részét formázhatod." — since agent 0.158.1 a part next to a mounted part can only be USED (the root wrapper trusts no „Felhom's own" claim). Evidence: `audits/storage-identity-buildB-2026-10-15/live/14-samedisk-refusal-offer.txt`. | READY | — | Reword `err.web.meghajto_hasznalatban_formazas` (hu + en): „… nem formázható. A szabad részét használatba veheted." Next controller release; the copy gates run. | CC |
+| **R-1144** | Storage & devices | P3 | **„Végleges leszerelés — mégis" stops the drive's apps before the agent is asked; when the agent refuses, the drive stays (fixed tonight, R-1063) but its apps stay stopped.** `handleStorageDecommission` mode `anyway` calls `StopStack` for every app on the drive, then `finalizeDecommission`; on the agent's 403/409 the household reads the plain refusal, and its apps are down until started by hand. Read from code 2026-10-18 (controller v0.329.0, `internal/web/storage_handlers.go`); not pressed live. | OPEN — filed 2026-10-18 | — | Start again the apps this press stopped when the refusal is a policy one (or ask the agent before stopping); a test with a refusing fake agent; next controller release | CC |
 
-## Security & access — 21 rows (P1 1, P2 2, P3 12, P4 6)
+## Security & access — 22 rows (P1 1, P2 2, P3 13, P4 6)
 
 | ID | Category | Sev | What | State | Blocked on | Next action | Owner |
 |---|---|---|---|---|---|---|---|
@@ -217,8 +214,8 @@ stopping line that lies.
 | **R-985** | Security & access | P3 | **Our zero-knowledge wording is stronger than the product: on a live box the operator can reach the off-site backup key (through the vaulted root credential and the operator tunnel) without the customer's recovery code.** Design-shaped; decision for the operator in `STATUS.md`. Fix options: change the wording, or remove operator root reach to the key. Security review 2026-10-12 (`audits/security-review-2026-10-12/`); detail held off-git for the operator (`/mnt/5_hdd/felhom.eu/evidence/security-review-2026-10-12-PRIVATE/`). | WAITING-ON-OPERATOR — design-proposal | operator decision | Decide the wording or the change | operator |
 | **R-1042** | Security & access | P3 | **The boxes' Cloudflare keys cannot change zone settings, so Cloudflare's own „Always Use HTTPS” cannot be switched on.** Measured 2026-10-14 on all three zones (`audits/https-only-2026-10-14/``1A-1-zones-and-keys.txt`): each `cf_api_token` is valid and sees only its own zone, but GET/PATCH `settings/always_use_https`, `ssl`, `security_header` answer 403; the redirect and transform rule phases too. The https-only fix therefore lives on the box (decision 246). A second layer at Cloudflare needs „Zone Settings: Edit” (and, for headers, „Transform Rules”) added to each customer's key — the hub's zone-reach check stays. | **WAITING-ON-OPERATOR** — filed 2026-10-14 | a key permission only the operator can grant | Optional: widen each customer key, then switch „Always Use HTTPS” on per zone | operator |
 | **R-1104** | Security & access | P3 | **Check https-only on Tester 2 when it reports again.** R-1033 (plain http on every public name) is closed on the three boxes that report (35 names → 301, 2026-10-16). Tester 2 (`sajatfelhom.hu`) has been off since 2026-10-04 on controller 0.292.0; its addresses answer Cloudflare 530 (tunnel down), so nothing is served in plain http while it is off. When it comes back it takes the floor (≥ 0.312.0 carries the redirect). | OPEN — filed 2026-10-16 | Tester 2 reporting again | Run `scripts/https_only_check.py` with Tester 2 in it; close | CC |
-| **R-1140** | Security & access | P3 | **The hub's ingress access logs cover about 13 hours, so a question like R-954's („did anyone use the old key?") cannot be answered next time.** Measured 2026-10-17: the ingress-nginx controller logs to the container log only; k3s keeps `container-log-max-size=10Mi` × `max-files=5` — 03:30 to 16:18 today; 268 `hub.felhom.eu` lines in that window. 90 days needs either a DooPlex k3s setting (all containers, ≈ 9 GB raw / ≈ 1.3 GB compressed at today's rate) or a hub-only access log kept on the hub's own volume (≈ 500 lines a day, ≈ 15 MB for 90 days). Both change DooPlex or the hub beyond this session. | WAITING-ON-OPERATOR — filed 2026-10-17 | operator decision | Pick: (A) a hub-only access log on the hub PVC, 90 days (recommended: small, no DooPlex change), or (B) longer k3s container logs | operator |
-| **R-1141** | Security & access | P3 | **`felhom-host-install --rotate-recovery` exists only as part of a full installer re-run, which also reinstalls the vouched agent and the root files.** On a box running a newer agent than the vouched one (demo-hp on the candidate, 2026-10-17) it would move the agent back; on a household box it is a much larger act than a password change. The 2026-10-17 rotation therefore ran the installer's own `step_break_glass` ALONE (`audits/agent-allowlist-rotation-2026-10-17/P3/`). | OPEN — filed 2026-10-17 | — | A `--rotate-recovery-only` mode (step 4b and nothing else), or a signed op; the runbook `break-glass.md` §5 points at it | CC |
+| **R-1140** | Security & access | P3 | **The hub's ingress access logs cover about 13 hours, so a question like R-954's („did anyone use the old key?") cannot be answered next time.** Measured 2026-10-17: the ingress-nginx controller logs to the container log only; k3s keeps `container-log-max-size=10Mi` × `max-files=5` — 03:30 to 16:18 today; 268 `hub.felhom.eu` lines in that window. 90 days needs either a DooPlex k3s setting (all containers, ≈ 9 GB raw / ≈ 1.3 GB compressed at today's rate) or a hub-only access log kept on the hub's own volume (≈ 500 lines a day, ≈ 15 MB for 90 days). Both change DooPlex or the hub beyond this session. | **NARROWED 2026-10-18 — built, not deployed.** Hub v0.158.0 (image `sha256:e4203978…`, felhom.eu `7d64038f`): the hub writes its own access log, one line per request (time, method, path without query, status, duration, source, the credential KIND and identity kind — never a value; `/bind/<token>` and 32+-character segments redacted), `/data/access-log/access-YYYY-MM-DD.log`, deleted after 90 days (`05` §16.4). **Rate corrected:** ≈ 1,870 hub lines a day through the ingress (measured 2026-10-17, most are `/wait` long-polls), not ≈ 500 — so ≈ 0.2 MB a day, ≈ 19 MB for 90 days. Red-proofed (`TestAccessLog_NeverWritesASecret` leaks a fake token with the raw URL). The night does not deploy the hub (decision 162). | the operator's yes to deploy hub v0.158.0 | Deploy hub v0.158.0 (daytime, on the operator's yes); read one day's file on the PVC; close | CC |
+| **R-1141** | Security & access | P3 | **`felhom-host-install --rotate-recovery` exists only as part of a full installer re-run, which also reinstalls the vouched agent and the root files.** On a box running a newer agent than the vouched one (demo-hp on the candidate, 2026-10-17) it would move the agent back; on a household box it is a much larger act than a password change. The 2026-10-17 rotation therefore ran the installer's own `step_break_glass` ALONE (`audits/agent-allowlist-rotation-2026-10-17/P3/`). | **NARROWED 2026-10-18 — built on `main`, not published.** `felhom-host-install.sh --rotate-recovery-only` runs step 4b alone (felhom.eu `7d64038f`): reads the hub, host id and host key from the agent config, refuses byo and every other mode flag, asks the hub's `/healthz` before the password changes, never touches the agent or root files. The default install path is unchanged in behaviour (harness verdicts and 9 transcripts identical old vs new; demo-hp `--preflight-only` identical; `--rotate-recovery-only --dry-run` on demo-hp changed nothing). `step_break_glass` moved, byte-identical, ahead of the argument parse. `break-glass.md` §5 points at it. Publishes only with the next `installer-v…` tag. | the next installer tag | Cut the next `installer-v…` tag (with the other untagged installer changes); close | CC |
 | **R-955** | Security & access | P4 | **The cross-box check of the hub API is done in source and tests but not fully proven: if a later check masks a missing id check, a box's key could read or change another box's hub record without any test noticing.** Read 2026-10-12: every box-facing `/api/v1` route takes the box from its own key or refuses a different id; 14 routes already had a refusal test, 5 got one (`crossbox_refusal_test.go`). Left: red-prove the 4 new tests that were not red-proved (as the restore-directive one was), and the live read-only check with one demo box's key against the other demo box's id. | OPEN — filed 2026-10-12 by the security review | — | Red-prove the 4 tests; run the live GET check on the two demo boxes | CC |
 | **R-924** | Security & access | P2 | **The operator signing keys exist only on DooPlex, with no passphrase and in no backup — a loss of DooPlex means no box ever accepts a signed agent update, config bundle or OS step again without an on-site re-enrollment.** MEASURED 2026-10-09: `/mnt/5_hdd/felhom.eu/felhom-op-operational`, `felhom-rec-recovery`, `felhom_op_ed25519` (0600 `kisfenyo`); `ssh-keygen -y -P ''` opens all three (no passphrase); `/mnt/5_hdd/felhom.eu` is in no backup set, the off-site job does not carry them. `04` §1 says the operational key is "passphrase-protected" and the recovery key "kept offline" — today neither holds (custody ruling 1 of 2026-09-16 put both on DooPlex for this phase; `04` §3.1). Both keys share one fate, so the cold key cannot do its one job (re-pin after losing the operational key). `runbooks/total-loss-of-dooplex.md` S6. | **NARROWED 2026-10-09 — operator ruling: option A** (print both AND copy them). The three key files and their public halves ride the nightly encrypted DooPlex→ep0 copy (`scripts/dooplex-offsite/`, a missing key refuses the push); the Sunday restore test proves each private key still derives its own public half (never printed). **LEFT:** the operator prints S6a/S6b (`runbooks/break-glass-sheet.md`); the keys still have no passphrase (custody ruling 1 of 2026-09-16 stands until the first sale). | operator: print | Operator: print the two keys; then close | operator |
 | **R-132** | Security & access | P3 | **`curl -w '%{redirect_url}'` reconstructs the request URL WITH its basic-auth credential** — so a `-u ":$HUB_PW"` call that never put the password in a URL still printed it **Merged 2026-10-05 from R-350 (duplicate):** (1) 2026-08-20 occurrence: POST /configuration/artifacts answers 303; leak lives only in the CC transcript under ~/.claude/projects/ on DooPlex, not in git/evidence (checked then). (2) `-v` and `--libcurl` also re-render the credential, not only %{redirect_url}; confirm redirects with %{http_code} + follow-up GET. (3) Rotation path: hub /configuration form (current_password/new_password/confirm_password); DB override wins over ConfigMap (break-glass); CC can rotate file-to-file without printing (operator-present-one-time-secrets) if asked. | **WAITING-ON-OPERATOR** — **ACTION: rotate `HUB_PW`** **Folded R-580 2026-10-03** (the same `curl -w %{redirect_url}` credential echo, seen again 2026-09-18). **Note 2026-10-12 (security review):** `manifests/hub.yaml` also carries the hub login's bcrypt hash in this public repository (the break-glass field); when `HUB_PW` is rotated, decide whether that field should stay empty. | — | Happened on 2026-07-31 while red-proofing the R-120 gate: the hub operator password was written to the session transcript by the write-out format, not by the request. `-u` is safe; the *reporting* was not. Rule: read the redirect from `-D -` and grep `^Location:`, never `%{redirect_url}`, on any authenticated call. Rotate the hub password (`/configuration` → Login password; ConfigMap `auth.password_hash` is the reset path) and update `~/.config/credentials` | Viktor |
@@ -234,6 +231,7 @@ stopping line that lies.
 | **R-1054** | Security & access | P3 | **The hub's operator/global API key is still one bearer that can do everything on `/api/v1`, including reading every box's break-glass root password (`GET /admin/hosts/{id}/recovery-credential`).** Hub v0.152.0 gave the web surface scoped tokens (`05` §17) but tokens are refused on `/api/v1` by design. Who uses the global key today is not listed in one place. Fix: list its consumers; move each operator use to a scoped token on the web side (or a token class on the API), and keep the global key only where a box path needs it — or retire it. | OPEN — filed 2026-10-15 by the token build | — | List the global key's consumers (pod env, scripts, runbooks) | CC |
 | **R-1118** | Security & access | P3 | **CC's secrets moved off the plain `~/.config/credentials` into an encrypted store with a broker and guards; what is left is retiring the old file.** BUILT 2026-10-17 (store B, operator ruling 11:28): private repo `admin/cc-secrets` (`ARCHITECTURE.md`, the design verbatim in `docs/`), `sec` (no get), PreToolUse refusals + PostToolUse masking for Bash and Read in `~/.claude/settings.json` (every CC session on DooPlex), nightly transcript sweep with mail. 55 tests, 18 red-proofs, 36-row child-session proof in both launch modes, fresh-session proof (`audits/cc-creds-build-2026-10-17/`). 25 names in the store (23 imported, `RESEND_API` from CC's environment, a self-test value). `12-agent-tooling.md` §5 points at it. | NARROWED — built; the old file's retirement is R-1134 | the week with no reader (R-1134) | `sec doctor` stays all OK; R-1134 | CC |
 | **R-1121** | Security & access | P3 | **`~/rotated-secrets-2026-10-09.txt` still exists on DooPlex** — 632 bytes, modified 2026-10-09 13:08 +0200, mode 600 (SPIKE Q6: `test -e` + `stat` only, not opened). Left by the R-925 rotation work. If any value in it is still valid, it is one more plaintext secrets file in the home folder, readable by every CC session. | WAITING-ON-OPERATOR | what the file holds — only the operator knows | Operator: confirm every value in it is revoked, then `shred -u` it; or move the live ones into the R-1118 store | operator |
+| **R-1145** | Security & access | P3 | **The installer's step 4b sets the new console root password (chpasswd) BEFORE it vaults it at the hub; if the vault then fails, root@pam holds a password nobody has.** The console is the last door into a broken box. Found 2026-10-18 while building `--rotate-recovery-only` (R-1141), which now asks the hub's `/healthz` first — that narrows it, it does not make the vault-then-set order. The full install path keeps the old order. | OPEN — filed 2026-10-18 | — | Vault first (or stage), set the password only after the hub answers 2xx, and keep the old one on any failure; harness case with a failing vault | CC |
 
 ## Box system & updates — 12 rows (P2 1, P3 7, P4 4)
 
@@ -344,7 +342,7 @@ stopping line that lies.
 | **R-1135** | Process & tooling | P4 | **CI's failure mail links the run at the in-cluster Gitea address** (`http://gitea.gitea-system.svc.cluster.local:3000/...`), which a reader outside the cluster cannot open. Seen in the forced test mail 2026-10-17 (job 1997). | OPEN — filed 2026-10-17 | — | Use the public Gitea URL in the alarm step's link | CC |
 | **R-1136** | Process & tooling | P4 | **felhom.eu CI runs take ~6.9 minutes and complete, so the 5-minute job timeout R-265 assumes does not hold there** (runs 1996, 1997; `script-tests` alone ~356 s). Found by the R-1129 proof helper 2026-10-17. | OPEN — filed 2026-10-17 | — | Measure the runner's real timeout; correct R-265's assumption or speed up `script-tests` | CC |
 | **R-1137** | Process & tooling | P4 | **A CI rerun may overwrite a failed job's log** — failure mails arrived for controller commits `c02c9d83` and `29a3ce8c` while their jobs (1944, 1959) now read green. UNPROVEN (a reading, 2026-10-17). | OPEN — filed 2026-10-17 | — | Confirm with one deliberate red + rerun on a scratch repo; if true, the CI check must read the first attempt | CC |
-| **R-1143** | Process & tooling | P4 | **The controller `build.sh` pushes `:latest` with EVERY build, release candidates included**, and the legacy `controller/docker-compose.yml` follows `:latest`. Nothing current pulls it (the update button takes the highest x.y.z tag), but a legacy-shaped box would take an rc. Seen 2026-10-17 (rc.1–rc.5 each moved `:latest`). | OPEN — filed 2026-10-17 | — | Push `:latest` only for a release tag, or not at all; pin it in `test_build_push_once.sh` | CC |
+| **R-1146** | Process & tooling | P4 | **`scripts/hostinstall-mode-harness.sh` is red on `main` and no gate runs it.** 2 of 47 fail on the unchanged default path: shellcheck SC2066 at the `for f in "$d/pve-enterprise.list"` loop, and GL8-F6 (cause not read). `repo_gates.py` runs only `hostinstall_gates.py`, so nobody sees it. Found 2026-10-18 (R-1141 work). | OPEN — filed 2026-10-18 | — | Fix the two (the loop is on the default path: ship it with the next installer tag), then register the harness in `repo_gates.py` with a decoy | CC |
 
 <!-- DUE-CHECKS-BEGIN — machine-readable. Parsed by scripts/due_checks_gate.py.
      One row per dated check. The R-number must have a row above. Dates are UTC.