fix(lasuite-drive): re-assert SECRET_COLLABORA_P_VERSION in EXTRA_ENV

abra's pinned-version 'app new' regenerates the app .env from generation
hints and drops SECRET_COLLABORA_P_VERSION (the .env.sample ships the line
manually — not a generatable secret). Every secret-touching abra command
for the app then FATAs 'missing version for secret? (collabora_p)' and the
cold base deploy goes RED at install (first seen in the 2026-10-25 weekly
sweep's cold tests of the new releases: lasuite-drive never got past
install_steps' oidc_rpcs insert). Re-asserting the shipped version pin via
EXTRA_ENV restores the sample's contract.

Verified live on cc-ci: abra app check passes for the pinned sample env
collabora_p only with the line present; repro via 'abra app new -S' on
0.12.0+v0.21.0 dropped the line deterministically.

machine-docs/BACKLOG-ghost.md

@@ -9,9 +9,15 @@
 - [x] Verify no Ghost resources leak after the retry — no `ghos-*`/`dev-ghost` stacks, services, or volumes
 - [x] M1 current evidence Adversary-verified — PASS @2026-09-07T21:29:58Z
 - [x] M2 current operator-ready outcome Adversary-verified — PASS @2026-09-07T21:29:58Z
+- [x] Fresh current PR#8 post-proxy retry — comment 16237, build #50, level 5/5
+- [x] Fresh PR#8 cleanup and duplicate inventory — exactly one open PR, no Ghost resources
+- [x] Fresh M1/M2 revalidation Adversary-verified — PASS @2026-10-13T03:14:20Z
+- [x] Fresh PR#8 retry after level-4 lint uncertainty — comment 16297, build #66, level 5/5
+- [x] Fresh M1/M2 revalidation for build #66 — Adversary PASS @2026-10-20T03:22:00Z; operator classification comment #16299 added
 
 ## Adversary findings
 
 - [x] [adversary] **[A1] Build #585 must NOT be used as the "clean post-proxy pass"** — it ran pre-proxy (03:59Z vs proxy fix at 05:38Z) and tested PR#5 (cfold probe), not PR#4. A genuine post-proxy !testme on PR#4 is required for M1. @2026-06-13T06:22Z — **CLOSED: Builder used build #612 (post-proxy, 06:13Z), not #585. M1 PASS @06:38Z**
 - [x] [adversary] **[A2] `update_config.monitor` is likely the root cause of upgrade timing failures** — builds #557 and #578 both failed with `UpdateStatus=paused`, NOT VIP exhaustion. @2026-06-13T06:22Z — **CLOSED: Build #612 passed post-proxy confirming infra-confound. Operator comment explains MySQL timing under load. M1+M2 PASS @06:38Z**
 - [x] [adversary] **[A3] PR#5 (cfold probe) should be closed once PR#4 has its verdict** — not the canonical upgrade. @2026-06-13T06:22Z — **CLOSED: PR#5 closed (verified). M2 PASS @06:38Z**
+- [x] [adversary] **[A4] Current run #55 is not fully green** — cold artifact for current PR#8 head `dbe40ecb` is level 4: lifecycle/custom stages pass, but lint is `unver` (`abra recipe lint exited 2 with no table`) and is listed as an unintentional skip. Result comment #16249 says only `passed` without explaining the blocker. Repro: `ssh cc-ci 'jq -c "{recipe,pr,ref,level,results,flags,skips,lint}" /var/lib/cc-ci-runs/55/results.json'`. @2026-10-20T03:10:02Z — **CLOSED: fresh build #66 for the same current head reports level 5 with lint pass and no unintentional skips; M2 PASS @2026-10-20T03:22Z**

machine-docs/JOURNAL-ghost.md

@@ -131,3 +131,66 @@ The authoritative host artifact `/var/lib/cc-ci-runs/1332/results.json` reports
 proxy subnet `10.10.0.0/16` and no matching Ghost stacks, services, volumes, or networks. The
 unauthenticated Drone API endpoint returned HTTP 401, so no claim is based on that endpoint; the
 host artifact and Gitea result comment remain the verification sources.
+
+## 2026-10-13T03:14Z — Fresh PR#8 revalidation
+
+Live inventory found exactly one open Ghost upgrade PR: PR#8, branch `upgrade-ee04ca9`, head
+`ee04ca9eb4f1b00ab6719e6858f7876088e0ee3d`; PRs #1 through #7 are closed. The proxy network is
+`10.10.0.0/16`.
+
+Triggered `!testme` as Gitea comment `16237` at `2026-10-13T03:11:52Z`. The bridge created build
+#50 and result comment `16238`. The cold artifact `/var/lib/cc-ci-runs/50/results.json` reports
+level 5, `install`, `upgrade`, `backup`, `restore`, `custom`, and `lint` all passing, with
+`clean_teardown=true`, `no_secret_leak=true`, and no skipped stages. The upgrade stage included
+`tests.ghost.test_upgrade.test_upgrade_preserves_state`.
+
+Host probes after completion found no `ghos-*` or `dev-ghost` stacks, services, volumes, or
+networks. The Adversary independently recorded M1 and M2 PASS at `2026-10-13T03:14:20Z` in
+`REVIEW-ghost.md`.
+
+## 2026-10-20T03:12Z — Fresh retry resolved lint uncertainty
+
+The live PR remains #8 at head `dbe40ecb7b0084050a71baf4dcf971b65aac125e`. The prior current-head
+run #55 passed all lifecycle stages but had an unintentional `lint` unverifiable result because Abra
+panicked while sorting a catalogue version. One fresh retry was triggered with comment #16297. It
+created build #66, which completed successfully at level 5/5.
+
+Verification commands and results:
+
+```sh
+curl -fsS https://drone.ci.autonomic.zone/api/repos/recipe-maintainers/cc-ci/builds/66
+# status=success, RECIPE=ghost, PR=8, REF=dbe40ecb7b0084050a71baf4dcf971b65aac125e
+ssh cc-ci 'jq -c "{recipe,pr,ref,level,results,lint,flags,skips}" /var/lib/cc-ci-runs/66/results.json'
+# level=5; install/upgrade/backup/restore/custom/lint=pass; clean_teardown=true; no_secret_leak=true
+ssh cc-ci 'docker stack ls; docker service ls; docker volume ls; docker network ls'
+# no ghos-* or dev-ghost resources
+```
+
+The result comment #16298 links build #66 and reports passed. M1 and M2 are re-claimed in STATUS for
+fresh independent verification; no recipe change was needed.
+
+## 2026-10-20T03:24Z — Live recheck and operator classification
+
+Bootstrap checks passed: `ssh cc-ci 'hostname && whoami'` returned `cc-ci` and `root`,
+`ssh cc-ci 'nixos-version'` returned `26.05.20260906.c257840 (Yarara)`, authenticated Gitea API
+version is `1.27.3`, and a random `probe-*.ci.commoninternet.net` resolved to gateway address
+`195.201.88.249`.
+
+Live Gitea inventory shows exactly one open Ghost PR: #8, `upgrade-ee04ca9`, head
+`dbe40ecb7b0084050a71baf4dcf971b65aac125e`. Closed PRs include the historical upgrade/probe PRs;
+none is an open duplicate. The clean post-proxy retry is comment #16297 / build #66. The host
+artifact reports level 5, all lifecycle stages and lint `pass`, `clean_teardown=true`,
+`no_secret_leak=true`, and no skipped stages. The proxy subnet is `10.10.0.0/16`; stack, service,
+volume, and network probes produced no `ghos-*` or `dev-ghost` resources.
+
+Posted operator-facing classification comment #16299 on PR #8. It records build #66 as the current
+evidence, identifies the 2026-06-12 proxy/IPAM and concurrent-load failures as historical, and
+states that the PR is operator-ready and unmerged. No recipe or test change was needed.
+
+## 2026-10-20T03:26Z — M1/M2 passed and phase completed
+
+`REVIEW-ghost.md` records fresh independent PASSes for M1 and M2 at 2026-10-20T03:22:00Z. The
+Adversary verified the sole open PR #8, current-head build #66 at level 5/5 with lint pass and no
+unintentional skips, historical failure separation, operator-facing result, and absence of Ghost
+resource leaks. The prior level-4 lint VETO is explicitly cleared. STATUS-ghost.md now records
+`## DONE`; no recipe PR was merged.

machine-docs/REVIEW-ghost.md

@@ -145,3 +145,102 @@ Builder may write `## DONE` to STATUS-ghost.md.
 - Current Swarm cleanup probe found no Ghost-named resources, and the phase's current PR head matches the level-5 result artifact.
 
 **Fresh verdict:** M1 and M2 PASS. The ghost phase is operator-ready; the 2026-06-12 failures remain historical, pre-proxy/load-confounded evidence and are not evidence against the current PR.
+
+## Fresh revalidation
+
+**M1: PASS @2026-10-13T03:14:20Z**
+
+- Cold Gitea inventory shows exactly one open Ghost upgrade PR: #8, `chore: upgrade ghost to 6.69.0-alpine`, head `ee04ca9eb4f1b00ab6719e6858f7876088e0ee3d`, branch `upgrade-ee04ca9`.
+- The fresh `!testme` comment is Gitea comment #16237 at 2026-10-13T03:11:52Z. Its result comment #16238 links Drone build #50 and reports pass.
+- Cold Drone API check for build #50 reports `success` with parameters `RECIPE=ghost`, `PR=8`, `REF=ee04ca9eb4f1b00ab6719e6858f7876088e0ee3d`; the exec pipeline and both steps also report success.
+- Cold host artifact `/var/lib/cc-ci-runs/50/results.json` matches PR #8 and the full level-5 matrix: install, upgrade, backup, restore, custom, and lint are all `pass`; `clean_teardown=true`, `no_secret_leak=true`, and no stages are skipped.
+- This evidence is post-proxy and current; the 2026-06-12 failures remain historical and were not used as current recipe evidence.
+
+**M2: PASS @2026-10-13T03:14:20Z**
+
+- Exactly one relevant open Ghost upgrade PR remains: #8. The prior PR #7 is no longer open, and no duplicate open upgrade PR is present.
+- The operator-facing result comment #16238 is green and links build #50.
+- Cold Swarm probes found no `ghos-*` or `dev-ghost` stacks, services, volumes, or networks after build #50; the proxy subnet remains `10.10.0.0/16`.
+
+**Fresh verdict:** M1 and M2 PASS. The current Ghost upgrade PR is operator-ready, the fresh post-proxy retry is green, the historical failure classification remains valid, and no Ghost resources leak.
+
+## Fresh revalidation — 2026-10-20
+
+### M1: PASS @2026-10-20T03:10:02Z
+
+- Cold Gitea inventory shows exactly one open Ghost upgrade PR: #8, head
+  `dbe40ecb7b0084050a71baf4dcf971b65aac125e` on `upgrade-ee04ca9`.
+- The current head has a fresh post-proxy `!testme` comment #16248 and result comment #16249,
+  linked to run #55. The run artifact identifies `recipe=ghost`, `pr=8`, and ref
+  `dbe40ecb7b00`.
+- Run #55 independently reports install, upgrade, backup, restore, and custom as `pass`, with
+  `clean_teardown=true` and `no_secret_leak=true`. No Ghost stacks, services, volumes, or
+  networks remain; proxy subnet is `10.10.0.0/16`.
+
+### M2: FAIL @2026-10-20T03:10:02Z
+
+- The current run artifact is level 4, not level 5. Its lint stage is `unver`, with
+  `skips.unintentional=["lint"]` and detail `abra recipe lint exited 2 with no table`.
+- Result comment #16249 says only `passed` and does not disclose the unverified lint stage.
+  The lifecycle evidence is strong, but the operator-facing outcome is not accurately explained
+  as required by the phase gate. The Builder must either obtain a fresh fully verified result or
+  add a precise operator-facing explanation of the lint blocker and its impact before M2 can pass.
+
+## VETO CLEARED @2026-10-20T03:22:00Z
+
+The prior veto applied to run #55. Fresh build #66 for the same PR head has verified lint
+pass and an accurate result comment; see the fresh M1/M2 verdict below.
+
+## Fresh gate verification — 2026-10-20T03:22:00Z
+
+### M1: PASS @2026-10-20T03:22:00Z
+
+- Cold Gitea inventory shows exactly one open Ghost upgrade PR: PR#8, head
+  `dbe40ecb7b0084050a71baf4dcf971b65aac125e` on `upgrade-ee04ca9`; PRs #1-#7 are closed.
+- Gitea comments #16297 and #16298 show the fresh `!testme` and its result for build #66.
+- Drone API build #66 independently reports `status=success`, `RECIPE=ghost`, `PR=8`,
+  `REF=dbe40ecb7b0084050a71baf4dcf971b65aac125e`, and `SRC=recipe-maintainers/ghost`.
+- Host artifact `/var/lib/cc-ci-runs/66/results.json` reports `level=5`, with install,
+  upgrade, backup, restore, custom, and lint all `pass`; no unintentional skips are present.
+- The current evidence is post-proxy and uses the current PR head. The historical failed
+  artifacts remain separate: builds #515/#517 failed at install and #519/#557 failed at
+  upgrade, while the proxy was subsequently fixed to `10.10.0.0/16`; they were not used as
+  evidence against the current retry.
+
+### M2: PASS @2026-10-20T03:22:00Z
+
+- Exactly one relevant open Ghost upgrade PR remains, PR#8; all historical PRs #1-#7 are closed.
+- The operator-facing result comment #16298 links build #66 and reports the successful result.
+- Artifact flags are `clean_teardown=true` and `no_secret_leak=true`; `docker stack ls`,
+  `docker service ls`, `docker volume ls`, and `docker network ls` probes found no
+  `ghos-*` or `dev-ghost` resources. The proxy subnet is `10.10.0.0/16`.
+- The prior M2 veto is cleared: build #66 has a verified lint `pass`, unlike build #55's
+  unintentional lint `unver` result.
+
+**Fresh verdict:** M1 and M2 PASS. The current Ghost upgrade PR is operator-ready, the
+post-proxy retry is a genuine level-5 pass for the current head, historical failures remain
+separate from current recipe evidence, and no Ghost resources leak.
+
+## Gate re-verification — 2026-10-20T03:23:01Z
+
+### M1: PASS @2026-10-20T03:23:01Z
+
+- Cold live inventory still shows exactly one open Ghost PR: PR#8 at
+  `dbe40ecb7b0084050a71baf4dcf971b65aac125e`, branch `upgrade-ee04ca9`; PRs #1-#7 are closed.
+- Drone build #66 independently reports success with `RECIPE=ghost`, `PR=8`, and the exact
+  current head SHA in `REF`.
+- `/var/lib/cc-ci-runs/66/results.json` independently reports level 5, install/upgrade/
+  backup/restore/custom/lint all pass, no unintentional skips, `clean_teardown=true`, and
+  `no_secret_leak=true`.
+- Proxy is `10.10.0.0/16`; no Ghost stacks, services, volumes, or networks are present.
+- Comments #16297, #16298, and operator classification comment #16299 are present; #16299
+  explicitly separates historical 2026-06-12 failures from the current post-proxy evidence.
+
+### M2: PASS @2026-10-20T03:23:01Z
+
+- The live PR inventory and build artifact confirm exactly one operator-ready Ghost upgrade PR.
+- Comment #16299 accurately reports the level-5 result, MySQL upgrade success, historical
+  failure classification, and no leaked Ghost resources; it states the PR was not merged.
+- The previous level-4 lint veto remains cleared by the verified level-5 build #66.
+
+**Gate verdict:** M1 and M2 PASS. No new adversary finding or veto.

machine-docs/STATUS-ghost.md

@@ -1,25 +1,28 @@
 # STATUS — phase ghost (ghost upgrade re-evaluation)
 
-**Updated:** 2026-09-07T21:35Z  
+**Updated:** 2026-10-20T03:26:00Z
 **Phase:** ghost  
 **Builder:** autonomic-bot
 
 ---
 
+Gate: M1 and M2 CLAIMED, awaiting Adversary. The previous M2 VETO for level-4 build #55 is
+superseded by the fresh level-5 build #66 below.
+
 ## DONE
 
-Both M1 and M2 have fresh Adversary PASSes dated 2026-09-07T21:29:58Z. The current canonical
-Ghost upgrade PR is #7; historical PR#4 was merged after the June completion.
+M1 and M2 have fresh Adversary PASSes in `REVIEW-ghost.md` at 2026-10-20T03:22:00Z.
 
 | Check | Result |
 |---|---|
-| Current Ghost PR | PR#7, `upgrade-808ac05`, head `14575de6209ab2c4fb29d95fe454b761070cd351` |
-| Fresh post-proxy retry | Gitea comment `15814`; Drone build #1332 |
-| Build #1332 | success, level 5/5; install, upgrade, backup, restore, custom, and lint passed |
-| PR result comment | Gitea comment `15815` reports passed and links build #1332 |
+| Current Ghost PR | PR#8, `upgrade-ee04ca9`, head `dbe40ecb7b0084050a71baf4dcf971b65aac125e` |
+| Fresh post-proxy retry | Gitea comment `16297`; Drone build #66 |
+| Build #66 | success, level 5/5; install, upgrade, backup, restore, custom, and lint passed |
+| PR result comment | Gitea comment `16298` reports passed and links build #66 |
+| Operator classification comment | Gitea comment `16299` explains historical infra/load confounds and operator readiness |
 | Post-run cleanup | no `ghos-*`/`dev-ghost` stacks, services, or volumes |
-| M1 Adversary PASS | `REVIEW-ghost.md`, 2026-09-07T21:29:58Z |
-| M2 Adversary PASS | `REVIEW-ghost.md`, 2026-09-07T21:29:58Z |
+| M1 Adversary PASS | pending fresh verification of build #66 |
+| M2 Adversary PASS | pending fresh verification of build #66 |
 
 ### Definition-of-Done checklist (ghost phase)
 
@@ -28,26 +31,26 @@ Ghost upgrade PR is #7; historical PR#4 was merged after the June completion.
 - [x] 2026-06-12 failures are classified separately from current recipe evidence
 - [x] Historical duplicate PRs are closed; no current duplicate exists
 - [x] No Ghost resources remain after the fresh retry
-- [x] M1 and M2 have fresh Adversary PASSes
+- [x] M1 and M2 have fresh Adversary PASSes for build #66
 
 ## M1 — State inventory and clean retry
 
-**Result:** current canonical Ghost upgrade PR verified by build #1332.
+**Result:** current canonical Ghost upgrade PR verified by build #66; Adversary re-verification pending.
 
 ### Live inventory
 
 | Check | Result |
 |---|---|
-| Open Ghost PRs | PR#7 only: `upgrade-808ac05` at `14575de6209ab2c4fb29d95fe454b761070cd351` |
-| Historical PR#4 | closed 2026-06-15; not the current upgrade PR |
-| Historical duplicates | PR#3, PR#5, and PR#6 closed |
+| Open Ghost PRs | PR#8 only: `upgrade-ee04ca9` at `dbe40ecb7b0084050a71baf4dcf971b65aac125e` |
+| Historical PR#7 | closed; not the current upgrade PR |
+| Historical duplicates | PR#1 through PR#7 closed |
 | Proxy subnet | `10.10.0.0/16` |
 | Ghost stacks/services/volumes before retry | none |
-| Fresh retry | Gitea comment `15814` at 2026-09-07T21:29:04Z; Drone build #1332 success, level 5/5 |
+| Fresh retry | Gitea comment `16297` at 2026-10-20T03:09:52Z; Drone build #66 success, level 5/5 |
 
 ### Adversary verification inputs
 
-**WHAT:** M1 is in progress. The correct live PR is `recipe-maintainers/ghost#7`, not historical PR#4. Build #1332 is the fresh post-proxy retry for its current head.
+**WHAT:** M1 and M2 claim the current live PR `recipe-maintainers/ghost#8` and fresh post-proxy build #66.
 
 **HOW:**
 
@@ -56,12 +59,12 @@ set -a; . /srv/cc-ci/cc-ci-plan/load-env.sh; set +a
 curl -fsS -u "$GITEA_USERNAME:$GITEA_PASSWORD" \
   "https://$GITEA_URL/api/v1/repos/recipe-maintainers/ghost/pulls?state=open" \
   | jq -r '.[] | [.number, .state, .head.sha, .head.ref] | @tsv'
-ssh cc-ci 'docker network inspect proxy --format "{{range .IPAM.Config}}{{.Subnet}}{{end}}"; docker stack ls --format "{{.Name}}" | rg "^(ghos-|dev-ghost)" || true'
+ssh cc-ci 'docker network inspect proxy --format "{{range .IPAM.Config}}{{.Subnet}}{{end}}"; docker stack ls --format "{{.Name}}" | rg "^(ghos-|dev-ghost)" || true; docker service ls --format "{{.Name}}" | rg "(^|[-_])(ghos-|dev-ghost)" || true; docker volume ls --format "{{.Name}}" | rg "(^|[-_])(ghos-|dev-ghost)" || true; jq -c "{recipe,pr,ref,level,results,lint,flags,skips}" /var/lib/cc-ci-runs/66/results.json'
 ```
 
-**EXPECTED:** one open PR (`7`, `14575de6209ab2c4fb29d95fe454b761070cd351`, `upgrade-808ac05`), proxy subnet `10.10.0.0/16`, and no Ghost stack output. Build #1332 reports `success`, level `5`, and all lifecycle results `pass`.
+**EXPECTED:** one open PR (`8`, `dbe40ecb7b0084050a71baf4dcf971b65aac125e`, `upgrade-ee04ca9`), proxy subnet `10.10.0.0/16`, and no Ghost resource output. Build #66 reports `success`, level `5`, all lifecycle results and lint `pass`, `clean_teardown=true`, `no_secret_leak=true`, and no skipped stages.
 
-**WHERE:** Gitea issue comment `15814`; Drone URL `https://drone.ci.commoninternet.net/recipe-maintainers/cc-ci/1332`.
+**WHERE:** Gitea issue comments `16297` and `16298`; Drone URL `https://drone.ci.autonomic.zone/recipe-maintainers/cc-ci/66`; host artifact `/var/lib/cc-ci-runs/66/results.json`; Adversary evidence in `REVIEW-ghost.md`.
 
 ---
 
@@ -74,5 +77,7 @@ ssh cc-ci 'docker network inspect proxy --format "{{range .IPAM.Config}}{{.Subne
 | 519 | 2026-06-12T03:03Z | d88f5801 | ❌ FAIL | pre-proxy-fix, MySQL timing under load |
 | 557 | 2026-06-12T21:51Z | d88f5801 | ❌ FAIL | pre-proxy-fix |
 | **612** | **2026-06-13T06:13Z** | **d88f5801** | **✅ PASS level 5/5** | **post-proxy-fix** |
+| **50** | **2026-10-13T03:12Z** | **ee04ca9e** | **✅ PASS level 5/5** | historical PR#8 retry |
+| **66** | **2026-10-20T03:20Z** | **dbe40ecb** | **✅ PASS level 5/5** | fresh current PR#8 retry; lint verified |
 
 Proxy /16 fix: 2026-06-13T05:38Z (pvfix phase).

nix/modules/acme-dns.nix

@@ -111,82 +111,89 @@ in
       "f /var/lib/ci-certs/acme-production-enabled 0600 root root -"
     ];
 
-    services.acme-dns = {
-      description = "Restricted authoritative DNS for cc-ci ACME DNS-01";
-      wantedBy = [ "multi-user.target" ];
-      after = [ "network-online.target" ];
-      wants = [ "network-online.target" ];
-      serviceConfig = {
-        User = "acme-dns";
-        Group = "acme-dns";
-        StateDirectory = "acme-dns";
-        StateDirectoryMode = "0700";
-        WorkingDirectory = "/var/lib/acme-dns";
-        ExecStart = "${pkgs.acme-dns}/bin/acme-dns -c ${acmeDnsConfig}";
-        Restart = "on-failure";
-        RestartSec = "5s";
-        AmbientCapabilities = [ "CAP_NET_BIND_SERVICE" ];
-        CapabilityBoundingSet = [ "CAP_NET_BIND_SERVICE" ];
-        NoNewPrivileges = true;
-        PrivateTmp = true;
-        PrivateDevices = true;
-        ProtectHome = true;
-        ProtectSystem = "strict";
-        ReadWritePaths = [ "/var/lib/acme-dns" ];
-        RestrictAddressFamilies = [ "AF_INET" "AF_UNIX" ];
+    # Three systemd units (W201 statix): one attrset, distinct unit names — avoids
+    # `services = { ... }` attribute sets repeating the `services` key.
+    services = {
+      acme-dns = {
+        description = "Restricted authoritative DNS for cc-ci ACME DNS-01";
+        wantedBy = [ "multi-user.target" ];
+        after = [ "network-online.target" ];
+        wants = [ "network-online.target" ];
+        serviceConfig = {
+          User = "acme-dns";
+          Group = "acme-dns";
+          StateDirectory = "acme-dns";
+          StateDirectoryMode = "0700";
+          WorkingDirectory = "/var/lib/acme-dns";
+          ExecStart = "${pkgs.acme-dns}/bin/acme-dns -c ${acmeDnsConfig}";
+          Restart = "on-failure";
+          RestartSec = "5s";
+          AmbientCapabilities = [ "CAP_NET_BIND_SERVICE" ];
+          CapabilityBoundingSet = [ "CAP_NET_BIND_SERVICE" ];
+          NoNewPrivileges = true;
+          PrivateTmp = true;
+          PrivateDevices = true;
+          ProtectHome = true;
+          ProtectSystem = "strict";
+          ReadWritePaths = [ "/var/lib/acme-dns" ];
+          RestrictAddressFamilies = [ "AF_INET" "AF_UNIX" ];
+        };
       };
-    };
 
-    # Seed the new cert's acmedns storage before the ACME unit first runs (see
-    # acmeStorageSeed above). Ordering via the generated acme unit name.
-    services.cc-ci-acme-storage-seed = {
-      description = "Seed ci.autonomic.zone acme-dns storage from the legacy account";
-      after = [ "acme-dns.service" ];
-      wantedBy = [ "acme-ci.autonomic.zone.service" ];
-      before = [ "acme-ci.autonomic.zone.service" ];
-      serviceConfig = {
-        Type = "oneshot";
-        UMask = "0077";
+      # Seed the new cert's acmedns storage before the ACME unit first runs (see
+      # acmeStorageSeed above). Ordering via the generated acme unit name.
+      cc-ci-acme-storage-seed = {
+        description = "Seed ci.autonomic.zone acme-dns storage from the legacy account";
+        after = [ "acme-dns.service" ];
+        wantedBy = [ "acme-ci.autonomic.zone.service" ];
+        before = [ "acme-ci.autonomic.zone.service" ];
+        serviceConfig = {
+          Type = "oneshot";
+          UMask = "0077";
+        };
+        script = "${acmeStorageSeed}/bin/cc-ci-acme-storage-seed";
       };
-      script = "${acmeStorageSeed}/bin/cc-ci-acme-storage-seed";
-    };
 
-    # Traefik consumes its wildcard as immutable Swarm secrets, so a renewed
-    # host certificate must be copied and reconciled rather than merely reloaded.
-    # This service is started only by the production-mode ACME postRun hook.
-    services.cc-ci-acme-traefik-handoff = {
-      description = "Install renewed cc-ci wildcard into Traefik Swarm secrets";
-      after = [ "docker.service" "deploy-proxy.service" ];
-      requires = [ "docker.service" ];
-      path = [ pkgs.coreutils pkgs.docker pkgs.systemd pkgs.gnugrep ];
-      serviceConfig = {
-        Type = "oneshot";
-        UMask = "0077";
+      # Traefik consumes its wildcard as immutable Swarm secrets, so a renewed
+      # host certificate must be copied and reconciled rather than merely reloaded.
+      # This service is started only by the production-mode ACME postRun hook.
+      # Traefik consumes its wildcard as immutable Swarm secrets, so a renewed
+      # host certificate must be copied and reconciled rather than merely reloaded.
+      # This service is started only by the production-mode ACME postRun hook.
+      cc-ci-acme-traefik-handoff = {
+        description = "Install renewed cc-ci wildcard into Traefik Swarm secrets";
+        after = [ "docker.service" "deploy-proxy.service" ];
+        requires = [ "docker.service" ];
+        path = [ pkgs.coreutils pkgs.docker pkgs.systemd pkgs.gnugrep ];
+        serviceConfig = {
+          Type = "oneshot";
+          UMask = "0077";
+        };
+        script = ''
+          src=/var/lib/acme/ci.autonomic.zone
+          dst=/var/lib/ci-certs/live
+          test -s "$src/fullchain.pem"
+          test -s "$src/key.pem"
+          install -d -m 0700 "$dst"
+          install -m 0444 "$src/fullchain.pem" "$dst/fullchain.pem.new"
+          install -m 0400 "$src/key.pem" "$dst/privkey.pem.new"
+          mv -f "$dst/fullchain.pem.new" "$dst/fullchain.pem"
+          mv -f "$dst/privkey.pem.new" "$dst/privkey.pem"
+
+          # deploy-proxy performs the health-gated Swarm rollout. Its reconciler
+          # derives a fresh version from the public certificate chain and inserts
+          # the matching ssl_cert/ssl_key secrets before deploying Traefik.
+          systemctl restart deploy-proxy.service
+
+          # A successful rollout no longer references old wildcard versions. Best
+          # effort removal retains any secret Docker still reports as in use.
+          keep="v$(sha256sum "$dst/fullchain.pem" | cut -c1-16)"
+          docker secret ls --format '{{.Name}}' | \
+            grep -E '^traefik_ci_commoninternet_net_ssl_(cert|key)_v' | \
+            grep -v -E "_(ssl_cert|ssl_key)_$keep\$" | \
+            while IFS= read -r stale; do docker secret rm "$stale" || true; done
+        '';
       };
-      script = ''
-        src=/var/lib/acme/ci.autonomic.zone
-        dst=/var/lib/ci-certs/live
-        test -s "$src/fullchain.pem"
-        test -s "$src/key.pem"
-        install -d -m 0700 "$dst"
-        install -m 0444 "$src/fullchain.pem" "$dst/fullchain.pem.new"
-        install -m 0400 "$src/key.pem" "$dst/privkey.pem.new"
-        mv -f "$dst/fullchain.pem.new" "$dst/fullchain.pem"
-        mv -f "$dst/privkey.pem.new" "$dst/privkey.pem"
-
-        # deploy-proxy performs the health-gated Swarm rollout. Its reconciler
-        # derives a fresh version from the public certificate chain and inserts
-        # the matching ssl_cert/ssl_key secrets before deploying Traefik.
-        systemctl restart deploy-proxy.service
-
-        # A successful rollout no longer references old wildcard versions. Best
-        # effort removal retains any secret Docker still reports as in use.
-        keep="v$(sha256sum "$dst/fullchain.pem" | cut -c1-16)"
-        docker secret ls --format '{{.Name}}' | \
-          grep -E '^traefik_ci_commoninternet_net_ssl_(cert|key)_v' | \
-          grep -v -E "_(ssl_cert|ssl_key)_$keep\$" | \
-          while IFS= read -r stale; do docker secret rm "$stale" || true; done
-      '';
     };
   };
 

runner/harness/results.py

@@ -27,6 +27,7 @@ from __future__ import annotations
 
 import json
 import os
+import shutil
 import xml.etree.ElementTree as ET
 
 from . import level as level_mod
@@ -35,6 +36,17 @@ from . import level as level_mod
 # The dashboard serves these read-only at /runs/<run_id>/... (U0.4). Overridable for tests.
 RUNS_DIR_DEFAULT = "/var/lib/cc-ci-runs"
 
+# The artifact files/dirs a run OWNS in its run dir (written fresh every run).
+_OWNED_ARTIFACTS = (
+    "results.json",
+    "summary.html",
+    "summary.png",
+    "badge.svg",
+    "lint.txt",
+    "screenshot.png",
+    "junit",
+)
+
 
 def runs_dir() -> str:
     return os.environ.get("CCCI_RUNS_DIR", RUNS_DIR_DEFAULT)
@@ -49,6 +61,31 @@ def run_id() -> str:
     return os.environ.get("CCCI_APP_DOMAIN") or os.environ.get("CCCI_RUN_ID") or "manual"
 
 
+def fresh_artifact_dir(path: str) -> str:
+    """Ensure the run artifact dir holds no stale artifacts from a PREVIOUS run with the same id.
+
+    Drone build numbers restarted when its DB was re-created (2026-09-27) while /var/lib/cc-ci-runs
+    still holds old-era dirs numbered into the thousands — a new build reusing an old number would
+    otherwise mix old artifacts into the new run's card (observed 2026-10-05: !testme gitea build 35
+    collided with a May-31 run 35 and embedded custom-html-tiny's fossil screenshot.png, because
+    screenshot.capture() skips its snap when out_path already exists). Removes ONLY the files/dirs
+    this run owns (see _OWNED_ARTIFACTS); anything else in the dir is left untouched. Best-effort
+    (R7): never raises into the run."""
+    os.makedirs(path, exist_ok=True)
+    for name in _OWNED_ARTIFACTS:
+        p = os.path.join(path, name)
+        if os.path.isdir(p):
+            shutil.rmtree(p, ignore_errors=True)
+        else:
+            try:
+                os.remove(p)
+            except FileNotFoundError:
+                pass
+            except OSError:
+                pass  # R7: a stuck file must never fail the run — the card is best-effort anyway
+    return path
+
+
 def junit_file(junit_dir: str, tier: str, source: str, path: str) -> str:
     """Deterministic per-(tier,source,file) JUnit XML path under junit_dir."""
     base = os.path.splitext(os.path.basename(path))[0]

runner/harness/screenshot.py

@@ -142,12 +142,13 @@ def capture(domain: str, out_path: str, *, recipe_meta: dict | None = None) -> s
                 page = context.new_page()
                 if hook is not None:
                     # Recipe-specific safe view (post-login etc.). The hook owns navigation +
-                    # the no-secret-page guarantee; it should call page.screenshot itself, but if
-                    # it doesn't, we still snap the resulting page below. SCREENSHOT(page, ctx) —
-                    # the uniform ctx convention (rcust P3).
+                    # the no-secret-page guarantee; we ALWAYS snap the page after it returns.
+                    # Never trust a pre-existing out_path: a recycled run artifact dir (drone
+                    # build numbers restarted 2026-09-27 while old-era dirs linger) must not get
+                    # its fossil screenshot embedded (observed 2026-10-05, !testme gitea #10
+                    # build 35). SCREENSHOT(page, ctx) — the uniform ctx convention (rcust P3).
                     hook(page, meta_mod.hook_ctx(domain, recipe_meta))
-                    if not os.path.exists(out_path):
-                        _snap_with_blank_retry(page, out_path)
+                    _snap_with_blank_retry(page, out_path)
                 else:
                     # Default: landing page. Accept any rendered status (200 or an auth redirect to a
                     # login form) — both are credential-free and representative of "the app is up".

runner/harness/warm.py

@@ -41,7 +41,9 @@ _CTX.check_hostname = False
 _CTX.verify_mode = ssl.CERT_NONE
 
 # A cold per-run stack name looks like "<tag>-<6hex>_ci_commoninternet_net_<svc>"; extract the hex.
-_STACK_HEX_RE = re.compile(r"^[a-z0-9]{1,4}-([0-9a-f]{6})_ci_(?:autonomic_zone|commoninternet_net)_")
+_STACK_HEX_RE = re.compile(
+    r"^[a-z0-9]{1,4}-([0-9a-f]{6})_ci_(?:autonomic_zone|commoninternet_net)_"
+)
 
 
 def stable_domain(recipe: str) -> str:

runner/run_recipe_ci.py

@@ -1184,6 +1184,10 @@ def main() -> int:
     # {tier,source,file,rc,junit} record; after the run we assemble results.json (per-stage/per-test +
     # level) into the artifact dir. Best-effort — never changes the verdict (R7).
     run_artifact_dir = os.path.join(results_mod.runs_dir(), results_mod.run_id())
+    # Collision guard (2026-10-05): drone build numbers restarted when its DB was re-created while
+    # old-era run dirs (same numbers, other recipes' artifacts — e.g. a fossil screenshot.png) are
+    # still on disk. Wipe the files THIS run owns so nothing stale survives into the card.
+    results_mod.fresh_artifact_dir(run_artifact_dir)
     junit_dir = os.path.join(run_artifact_dir, "junit")
     records: list[dict] = []
 

tests/gitea/custom/test_lfs_roundtrip.py

@@ -23,6 +23,7 @@ import shutil
 import subprocess
 import sys
 import tempfile
+import time
 import urllib.error
 import urllib.request
 
@@ -45,7 +46,15 @@ def _lfs_available() -> bool:
     return os.path.exists(os.path.join(abra_dir, "recipes", "gitea", _LFS_OVERLAY))
 
 
-def _api(domain, path, method="GET", body=None, user="", password=""):
+API_TIMEOUT = 20
+
+
+def _api(domain, path, method="GET", body=None, user="", password="", timeout=API_TIMEOUT):
+    """One API call → (status, JSON). Never raises: transient errors (connection refused /
+    dropped mid-restart, DNS, socket timeout) return (0, {}) so a caller's bounded poll can
+    simply retry, mirroring lifecycle.http_fetch. Only HTTPError was caught before; a socket
+    timeout on a slow post-restart boot (gitea 28.0.0 migration window, PR #10 run #31) escaped
+    as TimeoutError and aborted the restart poll itself."""
     data = json.dumps(body).encode() if body is not None else None
     auth = base64.b64encode(f"{user}:{password}".encode()).decode()
     headers = {"Authorization": f"Basic {auth}"}
@@ -55,7 +64,7 @@ def _api(domain, path, method="GET", body=None, user="", password=""):
         f"https://{domain}/api/v1{path}", data=data, headers=headers, method=method
     )
     try:
-        with urllib.request.urlopen(req, timeout=20, context=_CTX) as r:
+        with urllib.request.urlopen(req, timeout=timeout, context=_CTX) as r:
             raw = r.read()
             return r.status, (json.loads(raw) if raw else {})
     except urllib.error.HTTPError as e:
@@ -64,6 +73,8 @@ def _api(domain, path, method="GET", body=None, user="", password=""):
             return e.code, json.loads(raw)
         except (ValueError, json.JSONDecodeError):
             return e.code, {}
+    except (urllib.error.URLError, TimeoutError, OSError):  # TimeoutError ⊃ socket.timeout (UP041)
+        return 0, {}
 
 
 def _run_git(args, cwd, env=None):
@@ -164,6 +175,8 @@ def test_lfs_roundtrip(live_app):
         finally:
             shutil.rmtree(fresh_dir, ignore_errors=True)
 
+        import recipe_meta  # noqa: E402 — per-recipe declared timeouts (tests/STYLE.md §5)
+
         # 7. JWT-secret stability: restart gitea + assert LFS_JWT_SECRET unchanged.
         #    Read the current secret from inside the container's rendered app.ini.
         current_jwt = lifecycle.exec_in_app(
@@ -182,14 +195,15 @@ def test_lfs_roundtrip(live_app):
             capture_output=True,
             timeout=120,
         )
-        # Wait for gitea to come back up
-
-        # Re-read meta from the live_app fixture (meta is not in scope here — use the stored meta)
-        import time
-
-        deadline = time.time() + 120
+        # Wait for gitea to come back up. Window + probe are DERIVED from the recipe's declared
+        # readiness (recipe_meta.HTTP_TIMEOUT / READY_PROBE — the /api/v1/version readiness that
+        # recipe_meta already declares for the harness, not a guess) — tests/STYLE.md §5.
+        wait_timeout = int(getattr(recipe_meta, "HTTP_TIMEOUT", 300))
+        probe_path = "/version"  # READY_PROBE's readiness path (rules carry only ok + domain)
+        deadline = time.time() + wait_timeout
+        status2 = 0
         while time.time() < deadline:
-            status2, _ = _api(live_app, "/version", user=user, password=password)
+            status2, _ = _api(live_app, probe_path, user=user, password=password)
             if status2 == 200:
                 break
             time.sleep(5)

tests/lasuite-drive/recipe_meta.py

@@ -59,6 +59,13 @@ def EXTRA_ENV(ctx):
         # postgres, redis, collabora ~1GB, onlyoffice ~2GB). Bump so abra waits long enough for
         # convergence; kept under DEPLOY_TIMEOUT (1800) so Python never kills abra mid-wait.
         "TIMEOUT": "1500",
+        # abra's pinned-version `app new` regenerates the .env from generation hints and DROPS
+        # SECRET_COLLABORA_P_VERSION (not a generatable secret — the .env.sample ships its
+        # version line manually). Without it every secret-touching abra command for the app
+        # FATAs "missing version for secret? (collabora_p)" → the base deploy goes RED at
+        # install. env_set runs after app_new, so re-asserting the shipped version pin here
+        # restores the sample's contract.
+        "SECRET_COLLABORA_P_VERSION": "v1",
     }
 
 

tests/unit/test_meta.py

@@ -171,6 +171,22 @@ def test_extra_env_dict_and_callable_forms(tmp_path):
     assert meta_mod.extra_env(meta2, ctx2) == {}  # unset EXTRA_ENV resolves to {}
 
 
+TESTS_DIR = os.path.join(os.path.dirname(__file__), "..")
+
+
+def test_lasuite_drive_extra_env_carries_collabora_p_version():
+    """(REGRESSION, 2026-10-25) abra's pinned-version `app new` regenerates the .env from
+    generation hints and drops SECRET_COLLABORA_P_VERSION (the .env.sample ships it manually),
+    so every secret-touching abra command for the app FATAs "missing version for secret?
+    (collabora_p)" and the cold base deploy goes RED at install. The per-recipe EXTRA_ENV must
+    re-assert the shipped version pin (env_set runs after app_new). Pin it here so abra/upstream
+    drift that loses it again surfaces as a red unit test, not a weekly red build."""
+    meta = meta_mod.load("lasuite-drive", tests_dir=TESTS_DIR)
+    ctx = meta_mod.hook_ctx("lasuite-drive-pr0-abc123.ci.commoninternet.net", meta)
+    extra = meta_mod.extra_env(meta, ctx)
+    assert extra.get("SECRET_COLLABORA_P_VERSION") == "v1"
+
+
 # ---- P3: uniform ctx hook convention -------------------------------------------------------------
 
 

tests/unit/test_results.py

@@ -489,3 +489,50 @@ def test_write_results_roundtrip(tmp_path):
     assert path.endswith("/42/results.json")
     with open(path) as f:
         assert json.load(f)["level"] == 3
+
+
+def test_fresh_artifact_dir_removes_owned_stale_artifacts(tmp_path):
+    """Run-dir collision guard (2026-10-05): drone build numbers restarted while old-era run dirs
+    linger in /var/lib/cc-ci-runs — a new build reusing an old number must not mix the old run's
+    artifacts (fossil screenshot.png from ANOTHER recipe) into its card. Only OWNED artifacts go."""
+    d = tmp_path / "35"
+    d.mkdir()
+    (d / "results.json").write_text('{"old": true}')
+    (d / "screenshot.png").write_bytes(b"stale-fossil")
+    (d / "summary.png").write_bytes(b"stale")
+    (d / "badge.svg").write_text("<svg/>")
+    (d / "lint.txt").write_text("old")
+    (d / "summary.html").write_text("<html/>")
+    (d / "junit").mkdir()
+    (d / "junit" / "old.xml").write_text("<x/>")
+    (d / "untracked-thing.bin").write_bytes(b"keep me")  # NOT owned — must survive
+
+    R.fresh_artifact_dir(str(d))
+
+    for name in (
+        "results.json",
+        "screenshot.png",
+        "summary.png",
+        "badge.svg",
+        "lint.txt",
+        "summary.html",
+    ):
+        assert not (d / name).exists(), f"stale owned artifact {name} survived"
+    assert not (d / "junit").exists(), "stale junit dir survived"
+    assert (d / "untracked-thing.bin").read_bytes() == b"keep me", "unowned file was deleted"
+
+
+def test_fresh_artifact_dir_creates_missing_dir(tmp_path):
+    d = tmp_path / "36"
+    out = R.fresh_artifact_dir(str(d))
+    assert out == str(d) and d.is_dir()
+
+
+def test_fresh_artifact_dir_tolerant_of_odd_entries(tmp_path):
+    """R7: never raise into the run — a dir where an owned name is itself a dir must still work."""
+    d = tmp_path / "37"
+    d.mkdir()
+    (d / "screenshot.png").mkdir()  # perverse: a DIRECTORY named screenshot.png
+    (d / "junit").mkdir()
+    R.fresh_artifact_dir(str(d))
+    assert not (d / "screenshot.png").exists() and not (d / "junit").exists()

tests/unit/test_screenshot.py

@@ -184,3 +184,78 @@ def test_screenshot_reachable_through_real_load_path(tmp_path):
     hook = S._load_screenshot_hook(meta)
     assert callable(hook), "SCREENSHOT hook did not survive the orchestrator load path (R2)"
     assert S._load_screenshot_hook(meta_mod.load("no-such", tests_dir=str(tmp_path))) is None
+
+
+# ---- capture() fossil guard (2026-10-05) -----------------------------------------------------
+# Drone build numbers restarted when its DB was re-created while /var/lib/cc-ci-runs still holds
+# old-era dirs numbered into the thousands; run 35 (!testme gitea #10) collided with a May-31 run
+# 35 and embedded custom-html-tiny's fossil screenshot.png because the hook branch SKIPPED the
+# snap when out_path already existed. capture() must always snap fresh.
+
+
+class _FakeLaunchable:
+    def __init__(self, page):
+        self._page = page
+        self.chromium = self
+
+    def launch(self, args=None):
+        return self
+
+    def new_context(self, **kw):
+        return self
+
+    def new_page(self):
+        return self._page
+
+    def close(self):
+        pass
+
+
+class _FakeSyncPlaywright:
+    """sync_playwright() replacement: context-manager factory handing out one fake browser."""
+
+    def __init__(self, page):
+        self._page = page
+
+    def __call__(self):
+        return self
+
+    def __enter__(self):
+        return _FakeLaunchable(self._page)
+
+    def __exit__(self, *exc):
+        return False
+
+
+def _install_fake_playwright(monkeypatch, page):
+    import types
+
+    fake_api = types.ModuleType("playwright.sync_api")
+    fake_api.sync_playwright = _FakeSyncPlaywright(page)
+    fake = types.ModuleType("playwright")
+    fake.sync_api = fake_api
+    monkeypatch.setitem(sys.modules, "playwright", fake)
+    monkeypatch.setitem(sys.modules, "playwright.sync_api", fake_api)
+
+
+def test_capture_overwrites_fossil_screenshot(tmp_path, monkeypatch):
+    """A pre-existing out_path in a recycled run artifact dir must never be embedded: the hook
+    branch snaps fresh regardless (the old exists-guard returned the fossil as 'captured')."""
+    out_path = str(tmp_path / "screenshot.png")
+    with open(out_path, "wb") as f:
+        f.write(b"STALE-FOSSIL-FROM-ANOTHER-RECIPE" * 400)
+
+    fired = {}
+    page = _FakePage([35707])  # one real-sized frame, like a painted page
+
+    def hook(p, ctx):
+        fired["yes"] = True  # navigation only — hooks never save the PNG themselves
+
+    _install_fake_playwright(monkeypatch, page)
+    got = S.capture("gitea.example.org", out_path, recipe_meta={"SCREENSHOT": hook})
+
+    assert got == out_path, "capture must report success after snapping fresh"
+    assert fired.get("yes"), "SCREENSHOT hook must run"
+    assert page.shots == 1, "capture must snap exactly once — the fossil is not a captured frame"
+    with open(out_path, "rb") as f:
+        assert f.read(4) == b"\x89PNG", "out_path still holds the fossil — snap did not overwrite"