fix: [alert] KubeletTooManyPods (cluster) (#937 )

Merge pull request 'fix(alertmanager): route InfoInhibitor to null and add the info inhibit rules' (#936 ) from fix/alertmanager-info-inhibitor into master
Reviewed-on: k.renner/homelab#936
2026-09-23 19:44:05 +00:00 · 2026-09-23 19:24:22 +00:00 · 2026-09-23 21:23:29 +02:00 · 2026-09-23 19:15:33 +00:00 · 2026-09-23 21:12:43 +02:00 · 2026-09-23 19:04:44 +00:00
75 changed files with 875 additions and 446 deletions
@@ -101,7 +101,7 @@ steps:
        bash scripts/ci/make-diag-kubeconfig.sh

  - name: alert-triage-agent
-    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.250
+    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.280
    environment:
      CLAUDE_CODE_OAUTH_TOKEN:
        from_secret: claude_code_oauth_token
@@ -350,17 +350,17 @@ steps:
          --destination gitea.renner.uno/gitea_admin/heating-can:$TAG \
          --destination gitea.renner.uno/gitea_admin/heating-can:latest

-  # netbird-proxy-services: the controller that reconciles NetBird reverse-proxy
+  # netbird-service-sync: the controller that reconciles NetBird reverse-proxy
  # services (public Ingress hosts -> TLS passthrough -> Traefik ClusterIP).
  # Runtime artifact like heating-can; same registry so the audit tracks it.
-  - name: build-netbird-proxy-services
+  - name: build-netbird-service-sync
    image: registry.gitlab.com/gitlab-ci-utils/container-images/kaniko:v1.25.18-debug
-    directory: ci/images/netbird-proxy-services
+    directory: ci/images/netbird-service-sync
    depends_on: []
    when:
      - event: deployment
      - event: push
-        path: ["ci/images/netbird-proxy-services/**", ".woodpecker/build-ci-images.yml"]
+        path: ["ci/images/netbird-service-sync/**", ".woodpecker/build-ci-images.yml"]
    environment:
      GITEA_USER: gitea_admin
      GITEA_TOKEN:
@@ -374,11 +374,11 @@ steps:
        /kaniko/executor \
          --context . \
          --dockerfile Containerfile \
-          --destination gitea.renner.uno/gitea_admin/netbird-proxy-services:$TAG \
-          --destination gitea.renner.uno/gitea_admin/netbird-proxy-services:latest
+          --destination gitea.renner.uno/gitea_admin/netbird-service-sync:$TAG \
+          --destination gitea.renner.uno/gitea_admin/netbird-service-sync:latest

  # wiki-index: embedding index over the Notes repo's wiki/, fed from Gitea and
-  # queried by ~/Notes/bin/wiki_rank.py. Runtime artifact like netbird-proxy-services.
+  # queried by ~/Notes/bin/wiki_rank.py. Runtime artifact like netbird-service-sync.
  - name: build-wiki-index
    image: registry.gitlab.com/gitlab-ci-utils/container-images/kaniko:v1.25.18-debug
    directory: ci/images/wiki-index
@@ -14,8 +14,8 @@
 #                 and  ci-claude:latest.  The Containerfile makes this layer
 #                 byte-reproducible, so rebuilding an unchanged version dedups
 #                 in the registry instead of orphaning a fresh ~260 MB copy.
-#   3. bump-pr  — (only when the version actually changed) rewrite the 5
-#                 consumer pins to  ci-claude:<version>  and open a PR. A human
+#   3. bump-pr  — (only when any pin is off the latest version) rewrite every
+#                 consumer pin to  ci-claude:<version>  and open a PR. A human
 #                 reviews and merges it; nothing auto-merges.
 #
 # TAG SCHEME — deliberately version tags (ci-claude:<claude-version>), NOT the dated
@@ -107,31 +107,22 @@ steps:
          [0-9]*.[0-9]*.[0-9]*) : ;;
          *) echo "npm returned an implausible version '$$NEW' — aborting."; exit 1 ;;
        esac
-        # The tag every consumer currently references. They are kept identical,
-        # so this must resolve to EXACTLY ONE non-empty tag; anything else
-        # (no match, or a partial/disagreeing set) aborts rather than risk a
-        # corrupting sed later. Counting with `grep -c .` (non-empty lines)
-        # instead of `wc -l` is deliberate: an empty result yields 0 here but
-        # 1 under wc -l, which would slip past the guard.
-        CUR=$$(grep -hoE 'ci-claude:[0-9A-Za-z._-]+' $$CONSUMERS | sed 's/^ci-claude://' | sort -u)
-        NTAGS=$$(printf '%s' "$$CUR" | grep -c . || true)
-        if [ -z "$$CUR" ] || [ "$$NTAGS" -ne 1 ]; then
-          echo "Expected exactly one ci-claude tag across the consumers, got [$$CUR]."
-          echo "Refusing to guess — fix the pins by hand."; exit 1
-        fi
-        echo "latest=$$NEW  current=$$CUR  event=$$CI_PIPELINE_EVENT"
+        # Distinct tags the consumers currently reference. Divergent pins are
+        # normal (a pipeline bumped on its own); the bump rewrites all of them.
+        CUR=$$(grep -hoE 'ci-claude:[0-9][0-9A-Za-z._-]*' $$CONSUMERS | sed 's/^ci-claude://' | sort -u | tr '\n' ' ' | sed 's/ $$//')
+        [ -n "$$CUR" ] || { echo "No ci-claude pin found in the consumers — aborting."; exit 1; }
+        echo "latest=$$NEW  current=[$$CUR]  event=$$CI_PIPELINE_EVENT"
        printf '%s' "$$NEW" > "$$CI_WORKSPACE/.ci-refresh/version"
        printf '%s' "$$CUR" > "$$CI_WORKSPACE/.ci-refresh/current"
-        # Build when the version changed, or on a push (Containerfile edit must
-        # ship even if the version is unchanged). Bump the pins only when the
-        # version the consumers reference is not the latest.
+        # Build when any pin is off latest, or on a push (Containerfile edit must
+        # ship even if the version is unchanged). Bump when any pin is off latest.
        if [ "$$CUR" != "$$NEW" ] || [ "$$CI_PIPELINE_EVENT" = "push" ]; then
          touch "$$CI_WORKSPACE/.ci-refresh/do-build"; echo "will build $$NEW"
        else
          echo "already on $$NEW — nothing to build"
        fi
        if [ "$$CUR" != "$$NEW" ]; then
-          touch "$$CI_WORKSPACE/.ci-refresh/do-bump"; echo "will open bump PR $$CUR -> $$NEW"
+          touch "$$CI_WORKSPACE/.ci-refresh/do-bump"; echo "will open bump PR [$$CUR] -> $$NEW"
        else
          echo "consumers already pin $$NEW — no PR"
        fi
@@ -175,6 +166,9 @@ steps:
      - |
        [ -f "$$CI_WORKSPACE/.ci-refresh/do-bump" ] || { echo "no version change — no PR"; exit 0; }
        set -eu
+        # Secret values can carry a trailing newline; it breaks the push URL and the auth header.
+        PUSH_TOKEN=$$(printf '%s' "$$PUSH_TOKEN" | tr -d '\r\n')
+        [ -n "$$PUSH_TOKEN" ] || { echo "PUSH_TOKEN is empty — aborting"; exit 1; }
        NEW=$$(cat "$$CI_WORKSPACE/.ci-refresh/version")
        CUR=$$(cat "$$CI_WORKSPACE/.ci-refresh/current")
        # Both were validated non-empty by the decide step; re-assert before a
@@ -190,9 +184,11 @@ steps:
        # is not a maintained list.
        CONSUMERS=$$(grep -rlE 'ci-claude:[0-9]' .woodpecker)
        [ -n "$$CONSUMERS" ] || { echo "No ci-claude consumers found — refusing to continue."; exit 1; }
-        # Exact-string swap of the current tag for the new one, consumers only.
+        # NEW goes into a sed replacement: allow only a plain x.y.z version.
+        printf '%s' "$$NEW" | grep -qE '^[0-9]+\.[0-9]+\.[0-9]+$$' || { echo "refusing non-x.y.z version '$$NEW'"; exit 1; }
+        # Every version-pinned tag becomes NEW, whatever it was; consumers only.
        for f in $$CONSUMERS; do
-          sed -i "s#ci-claude:$$CUR#ci-claude:$$NEW#g" "$$f"
+          sed -i "s#ci-claude:[0-9][0-9A-Za-z._-]*#ci-claude:$$NEW#g" "$$f"
        done
        # Author is claude-bot (the PAT's account); the commit subject makes
        # the automated-bump nature clear.
@@ -17,6 +17,6 @@ when:

 steps:
  - name: model-selection-tests
-    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.250
+    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.280
    commands:
      - sh scripts/ci/test-model-selection.sh
@@ -26,7 +26,7 @@ steps:
    # fetch-models.sh response mapping. No cluster, no secrets, no network — the
    # ci-claude image just provides sh + jq. Keeps the flag-injection regression
    # test from rotting (it exercises the exact injection shapes).
-    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.250
+    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.280
    commands:
      - sh scripts/ci/test-model-selection.sh

@@ -62,7 +62,7 @@ steps:
      - bash scripts/ci/make-diag-kubeconfig.sh

  - name: issue-fix-agent
-    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.250
+    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.280
    environment:
      CLAUDE_CODE_OAUTH_TOKEN:
        from_secret: claude_code_oauth_token
@@ -65,7 +65,7 @@ when:

 steps:
  - name: pr-review-agent
-    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.250
+    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.280
    environment:
      CLAUDE_CODE_OAUTH_TOKEN:
        from_secret: claude_code_oauth_token
@@ -70,7 +70,7 @@ steps:
    # this step is skipped — otherwise we'd fork-bomb PRs.
    when:
      - branch: master
-    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.250
+    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.280
    environment:
      CLAUDE_CODE_OAUTH_TOKEN:
        from_secret: claude_code_oauth_token
@@ -30,7 +30,7 @@ when:

 steps:
  - name: update-audit-agent
-    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.250
+    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.280
    environment:
      CLAUDE_CODE_OAUTH_TOKEN:
        from_secret: claude_code_oauth_token
@@ -1,12 +1,13 @@
 # Nightly wiki maintenance for the Notes vault (k.renner/notes), replacing the
-# workstation's wiki-maintenance.timer: distill the raw/ backlog into wiki/,
-# then audit wiki/. Each agent is its own `claude` run, not a subagent of an
-# orchestrator, so nothing depends on background tasks in print mode (the
-# workstation job died on their 600 s ceiling). The pipeline commits and pushes
-# as Claude Bot; the push webhook then refreshes the wiki-index service.
+# workstation's wiki-maintenance and wiki-consolidate timers: distill the raw/
+# backlog into wiki/ (wiki-distill agent), audit wiki/ with the vault's own
+# script (structural checks, no model), then consolidate (wiki-consolidate
+# agent: verifies a cohort of notes against sources, repos and the live
+# cluster and edits them). The pipeline commits and pushes as Claude Bot; the
+# push webhook then refreshes the wiki-index service.
 #
-# The agents and their retrieval scripts address the vault as ~/Notes, so each
-# agent step links $HOME/Notes to the checkout instead of rewriting them.
+# The agents address the vault as ~/Notes and the repos as ~/source/<repo>, so
+# each agent step links those into the workspace instead of rewriting them.
 when:
  - event: cron
    cron: wiki-maintenance
@@ -16,7 +17,7 @@ when:

 steps:
  - name: checkout-notes
-    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.278
+    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.280
    pull: true
    environment:
      WIKI_BOT_TOKEN:
@@ -30,9 +31,30 @@ steps:
        git -c http.extraheader="AUTHORIZATION: token $$TOKEN" \
          clone --depth 50 --branch main https://gitea.renner.uno/k.renner/notes.git notes
        echo "raw backlog: $$(ls notes/raw/*.md | grep -v _index.md | wc -l) file(s)"
+        # Full history for the consolidate agent's `git log` checks; the
+        # workspace clone is shallow and tree-filtered.
+        mkdir -p sources
+        git clone --quiet https://gitea.renner.uno/k.renner/homelab.git sources/homelab
+        git -c http.extraheader="AUTHORIZATION: token $$TOKEN" \
+          clone --quiet https://gitea.renner.uno/k.renner/homeassistant.git sources/homeassistant
+
+  - name: fetch-diag
+    # Read-only live-cluster access for the consolidate agent, same mechanism
+    # as issue-fix.yml: ./.ci-diag/kubectl plus a 1h ci-diagnostics token.
+    image: gitea.renner.uno/gitea_admin/ci-k8s:2026-09-08
+    environment:
+      SOPS_AGE_KEY:
+        from_secret: sops_age_key
+      S3_ACCESS_KEY_ID:
+        from_secret: hetzner_s3_access_key
+      S3_SECRET_ACCESS_KEY:
+        from_secret: hetzner_s3_secret_key
+    commands:
+      - bash scripts/ci/fetch-kubeconfig.sh
+      - bash scripts/ci/make-diag-kubeconfig.sh

  - name: distill
-    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.278
+    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.280
    pull: true
    environment:
      CLAUDE_CODE_OAUTH_TOKEN:
@@ -44,25 +66,37 @@ steps:
    commands:
      - |
        set -eu
+        # An empty queue must not start the agent: its startup context alone
+        # costs more than the whole audit.
+        BACKLOG=$$(ls notes/raw/*.md | grep -v _index.md | wc -l)
+        if [ "$$BACKLOG" -eq 0 ]; then
+          echo "distill: raw/ is empty, skipping the agent"
+          exit 0
+        fi
        ln -sfn "$$PWD/notes" "$$HOME/Notes"
        mkdir -p "$$HOME/.config/wiki-index"
        printf '%s' "$$WIKI_INDEX_BASIC_AUTH" | tr -d '\r\n' > "$$HOME/.config/wiki-index/basic-auth"
        chmod 600 "$$HOME/.config/wiki-index/basic-auth"
        "$$HOME/Notes/bin/wiki_rank.py" --probe-index
-      - |
-        set -eu
        MODEL=$$(sh scripts/ci/pick-model.sh "distill raw source notes into atomic wiki notes: read sources, check the wiki for existing coverage, write or merge notes; careful writing, moderate reasoning")
        if [ -n "$$MODEL" ]; then echo "distill: selected model $$MODEL"; else echo "distill: using the CLI default model"; fi
        cd notes
        cat ../ci/agents/wiki-maintenance/distill.md | claude $${MODEL:+--model $$MODEL} \
          --agent wiki-distill \
          --permission-mode acceptEdits \
-          --allowedTools "Read,Edit,Write,Glob,Grep,Bash(/root/Notes/bin/wiki_rank.py:*),Bash(/root/Notes/bin/wiki_ask.py:*),Bash(mv /root/Notes/raw/:*),Bash(mkdir -p /root/Notes/:*)" \
+          --allowedTools "Read,Edit,Write,Glob,Grep,Bash(/root/Notes/bin/wiki_rank.py:*),Bash(/root/Notes/bin/wiki_ask.py:*),Bash(mv /root/Notes/raw/:*),Bash(mkdir -p /root/Notes/:*),Bash(~/Notes/bin/wiki_rank.py:*),Bash(~/Notes/bin/wiki_ask.py:*),Bash(mv ~/Notes/raw/:*),Bash(mkdir -p ~/Notes/:*)" \
          --settings ../ci/agents/wiki-maintenance/settings.json \
          -p

  - name: audit
-    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.278
+    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.280
+    pull: true
+    commands:
+      # Findings are report lines in output/, not a pipeline failure.
+      - python3 -B notes/bin/wiki_audit.py --vault "$$PWD/notes"
+
+  - name: consolidate
+    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.280
    pull: true
    environment:
      CLAUDE_CODE_OAUTH_TOKEN:
@@ -75,23 +109,43 @@ steps:
      - |
        set -eu
        ln -sfn "$$PWD/notes" "$$HOME/Notes"
+        ln -sfn "$$PWD/sources" "$$HOME/source"
        mkdir -p "$$HOME/.config/wiki-index"
        printf '%s' "$$WIKI_INDEX_BASIC_AUTH" | tr -d '\r\n' > "$$HOME/.config/wiki-index/basic-auth"
        chmod 600 "$$HOME/.config/wiki-index/basic-auth"
-      - |
-        set -eu
-        MODEL=$$(sh scripts/ci/pick-model.sh "structural audit of ~1700 markdown notes: frontmatter completeness, broken links, orphans; mechanical checks, little reasoning")
-        if [ -n "$$MODEL" ]; then echo "audit: selected model $$MODEL"; else echo "audit: using the CLI default model"; fi
+        if [ -f .ci-diag/kubeconfig ]; then
+          ln -sf "$$PWD/.ci-diag/kubectl" /usr/local/bin/kubectl
+          export KUBECONFIG="$$PWD/.ci-diag/kubeconfig"
+        else
+          echo "consolidate: no ci-diag kubeconfig, live cluster checks unavailable"
+        fi
        cd notes
-        cat ../ci/agents/wiki-maintenance/audit.md | claude $${MODEL:+--model $$MODEL} \
-          --agent wiki-audit \
+        # JSON output: the agent's report and the run's cost come out of one
+        # record at the end instead of a 45-minute silent log.
+        # One source for the Bash allowlist: the settings file (run 10608: the
+        # agent writes ~/Notes, the rules carried only /root/Notes).
+        BASH_RULES=$$(jq -r '[.permissions.allow[] | select(startswith("Bash("))] | join(",")' ../ci/agents/wiki-maintenance/consolidate-settings.json)
+        rc=0
+        cat ../ci/agents/wiki-maintenance/consolidate.md | claude \
+          --agent wiki-consolidate \
          --permission-mode acceptEdits \
-          --allowedTools "Read,Write,Glob,Grep,Bash(/root/Notes/bin/wiki_rank.py:*),Bash(/root/Notes/bin/wiki_ask.py:*)" \
-          --settings ../ci/agents/wiki-maintenance/settings.json \
-          -p
+          --allowedTools "Read,Edit,Write,Glob,Grep,Agent,$$BASH_RULES" \
+          --settings ../ci/agents/wiki-maintenance/consolidate-settings.json \
+          --output-format json \
+          -p > /tmp/consolidate.json || rc=$$?
+        # A killed or crashed claude leaves no valid JSON; report rc, not a jq error.
+        if [ -s /tmp/consolidate.json ] && jq -e . /tmp/consolidate.json > /dev/null; then
+          jq -r '.result // empty' /tmp/consolidate.json
+          jq -c '{duration_ms, num_turns, total_cost_usd, is_error}' /tmp/consolidate.json
+        else
+          echo "consolidate: no result record (exit $$rc)"; head -c 2000 /tmp/consolidate.json || true
+        fi
+        echo "consolidate: git log --oneline of the run:"
+        git log --oneline origin/main..HEAD
+        exit $$rc

  - name: push-notes
-    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.278
+    image: gitea.renner.uno/gitea_admin/ci-claude:2.1.280
    pull: true
    environment:
      WIKI_BOT_TOKEN:
@@ -102,7 +156,13 @@ steps:
        cd notes
        git config user.email "claude-bot@renner.uno"
        git config user.name "Claude Bot"
-        if git diff --quiet && git diff --cached --quiet && [ -z "$$(git ls-files --others --exclude-standard)" ]; then
+        # Only the vault content is committed; anything an agent leaves elsewhere
+        # in the checkout (run 10362: scratch scripts in the root) stays behind.
+        # The consolidate agent commits its own snapshots, so HEAD may already be
+        # ahead of origin/main with a clean tree.
+        git add -A -- wiki raw output .wiki-consolidation-state.json
+        git diff --cached --quiet || git commit -q -m "wiki maintenance $$(date -u +%F): distill + audit + consolidate (CI run $$CI_PIPELINE_NUMBER)"
+        if [ "$$(git rev-list --count origin/main..HEAD)" -eq 0 ]; then
          echo "wiki-maintenance: no changes"
          exit 0
        fi
@@ -112,21 +172,27 @@ steps:
        # leaves the PR open for a human and fails the run.
        BRANCH="wiki-maintenance/$$(date -u +%F)-$$CI_PIPELINE_NUMBER"
        git checkout -q -b "$$BRANCH"
-        git add -A
-        git commit -q -m "wiki maintenance $$(date -u +%F): distill + audit (CI run $$CI_PIPELINE_NUMBER)"
        TOKEN=$$(printf '%s' "$$WIKI_BOT_TOKEN" | tr -d '\r\n')
        git -c http.extraheader="AUTHORIZATION: token $$TOKEN" push origin "HEAD:$$BRANCH"
        API="https://gitea.renner.uno/api/v1/repos/k.renner/notes"
        jq -n --arg title "wiki maintenance $$(date -u +%F) (CI run $$CI_PIPELINE_NUMBER)" \
-              --arg body "Nightly distill + audit. Merged by the pipeline; open only if the merge failed." \
+              --arg body "Nightly distill + audit + consolidate. Merged by the pipeline; open only if the merge failed." \
              --arg head "$$BRANCH" '{title: $$title, body: $$body, head: $$head, base: "main"}' > /tmp/pr.json
        code=$$(curl -s -o /tmp/pr-resp.json -w '%{http_code}' -X POST \
          -H "Authorization: token $$TOKEN" -H "Content-Type: application/json" --data @/tmp/pr.json "$$API/pulls")
        echo "POST pull -> HTTP $$code"
        [ "$$code" -ge 200 ] && [ "$$code" -lt 300 ] || { cat /tmp/pr-resp.json; exit 1; }
        NUM=$$(jq -r '.number' /tmp/pr-resp.json)
+        # Merge without delete_branch_after_merge: run 10254 got HTTP 500 from that
+        # call although the merge itself had landed. Verify the PR state instead of
+        # trusting the status code, then drop the branch as a separate, best-effort step.
        code=$$(curl -s -o /tmp/merge-resp.json -w '%{http_code}' -X POST \
          -H "Authorization: token $$TOKEN" -H "Content-Type: application/json" \
-          --data '{"Do":"merge","delete_branch_after_merge":true}' "$$API/pulls/$$NUM/merge")
+          --data '{"Do":"merge"}' "$$API/pulls/$$NUM/merge")
        echo "POST merge #$$NUM -> HTTP $$code"
-        [ "$$code" -ge 200 ] && [ "$$code" -lt 300 ] || { cat /tmp/merge-resp.json; echo "PR #$$NUM left open"; exit 1; }
+        MERGED=$$(curl -s -H "Authorization: token $$TOKEN" "$$API/pulls/$$NUM" | jq -r '.merged')
+        echo "PR #$$NUM merged: $$MERGED"
+        [ "$$MERGED" = "true" ] || { cat /tmp/merge-resp.json; echo "PR #$$NUM left open"; exit 1; }
+        code=$$(curl -s -o /tmp/del-resp.json -w '%{http_code}' -X DELETE -H "Authorization: token $$TOKEN" \
+          "$$API/branches/$$(printf '%s' "$$BRANCH" | sed 's|/|%2F|g')")
+        echo "DELETE branch $$BRANCH -> HTTP $$code (best effort)"
@@ -1,5 +0,0 @@
-You are running as the scheduled wiki-maintenance job in CI. The Notes vault is checked out at ~/Notes (the repository k.renner/notes).
-
-Run the standard audit pass over ~/Notes/wiki/ and write the dated report to ~/Notes/output/. Do not modify any wiki note. Do not commit or push; the pipeline does that.
-
-End with two lines: notes scanned, and issues found by category.
@@ -0,0 +1,120 @@
+{
+  "//": "Permission scope for the wiki-consolidate agent in CI: read the vault and the",
+  "//2": "source repos cloned under /root/source, read the live cluster through the",
+  "//3": "ci-diagnostics kubeconfig, edit wiki/, output/ and the rotation state only.",
+  "//4": "Deny rules win; git and kubectl are read-only subcommands, no network tools.",
+  "permissions": {
+    "allow": [
+      "Read(//root/Notes/**)",
+      "Read(//root/source/**)",
+      "Edit(//root/Notes/wiki/**)",
+      "Edit(//root/Notes/output/**)",
+      "Edit(//root/Notes/.wiki-consolidation-state.json)",
+      "Bash(/root/Notes/bin/wiki-snapshot.sh:*)",
+      "Bash(/root/Notes/bin/wiki_rank.py:*)",
+      "Bash(/root/Notes/bin/wiki_ask.py:*)",
+      "Bash(~/Notes/bin/wiki-snapshot.sh:*)",
+      "Bash(~/Notes/bin/wiki_rank.py:*)",
+      "Bash(~/Notes/bin/wiki_ask.py:*)",
+      "Bash(~/Notes/bin/wiki-delete.sh:*)",
+      "Bash(git -C /root/Notes log:*)",
+      "Bash(git -C /root/Notes show:*)",
+      "Bash(git -C /root/Notes diff:*)",
+      "Bash(git -C /root/Notes status:*)",
+      "Bash(git -C /root/Notes rev-parse:*)",
+      "Bash(git -C /root/Notes ls-files:*)",
+      "Bash(git -C ~/Notes log:*)",
+      "Bash(git -C ~/Notes show:*)",
+      "Bash(git -C ~/Notes diff:*)",
+      "Bash(git -C ~/Notes status:*)",
+      "Bash(git -C ~/Notes rev-parse:*)",
+      "Bash(git -C ~/Notes ls-files:*)",
+      "Bash(kubectl get:*)",
+      "Bash(kubectl describe:*)",
+      "Bash(kubectl version:*)",
+      "Bash(kubectl api-resources:*)",
+      "Bash(kubectl explain:*)",
+      "Bash(git log:*)",
+      "Bash(git show:*)",
+      "Bash(git status:*)",
+      "Bash(git diff:*)",
+      "Bash(git rev-parse:*)",
+      "Bash(git ls-files:*)",
+      "Bash(git -C /root/source/homelab log:*)",
+      "Bash(git -C /root/source/homelab show:*)",
+      "Bash(git -C /root/source/homelab diff:*)",
+      "Bash(git -C /root/source/homelab status:*)",
+      "Bash(git -C /root/source/homelab rev-parse:*)",
+      "Bash(git -C /root/source/homelab ls-files:*)",
+      "Bash(git -C /root/source/homeassistant log:*)",
+      "Bash(git -C /root/source/homeassistant show:*)",
+      "Bash(git -C /root/source/homeassistant diff:*)",
+      "Bash(git -C /root/source/homeassistant status:*)",
+      "Bash(git -C /root/source/homeassistant rev-parse:*)",
+      "Bash(git -C /root/source/homeassistant ls-files:*)",
+      "Bash(git -C ~/source/homelab log:*)",
+      "Bash(git -C ~/source/homelab show:*)",
+      "Bash(git -C ~/source/homelab diff:*)",
+      "Bash(git -C ~/source/homelab status:*)",
+      "Bash(git -C ~/source/homelab rev-parse:*)",
+      "Bash(git -C ~/source/homelab ls-files:*)",
+      "Bash(git -C ~/source/homeassistant log:*)",
+      "Bash(git -C ~/source/homeassistant show:*)",
+      "Bash(git -C ~/source/homeassistant diff:*)",
+      "Bash(git -C ~/source/homeassistant status:*)",
+      "Bash(git -C ~/source/homeassistant rev-parse:*)",
+      "Bash(git -C ~/source/homeassistant ls-files:*)",
+      "Bash(/root/Notes/bin/wiki-delete.sh:*)",
+      "Bash(ls:*)",
+      "Bash(date:*)",
+      "Bash(wc:*)",
+      "Bash(stat:*)",
+      "Bash(head:*)",
+      "Bash(tail:*)"
+    ],
+    "deny": [
+      "Read(//proc/**)",
+      "Read(//etc/**)",
+      "Read(//root/.config/**)",
+      "Read(//root/.kube/**)",
+      "Read(//woodpecker/**/.ci-diag/**)",
+      "Read(//root/Notes/.git/**)",
+      "Read(//root/source/*/.git/**)",
+      "Read(//root/Notes/_archive/**)",
+      "Edit(//root/Notes/raw/**)",
+      "Edit(//root/Notes/_archive/**)",
+      "Edit(//root/source/**)",
+      "Bash(kubectl apply:*)",
+      "Bash(kubectl delete:*)",
+      "Bash(kubectl patch:*)",
+      "Bash(kubectl edit:*)",
+      "Bash(kubectl scale:*)",
+      "Bash(kubectl rollout:*)",
+      "Bash(kubectl exec:*)",
+      "Bash(kubectl cp:*)",
+      "Bash(kubectl drain:*)",
+      "Bash(kubectl cordon:*)",
+      "Bash(kubectl label:*)",
+      "Bash(kubectl annotate:*)",
+      "Bash(git push:*)",
+      "Bash(git reset:*)",
+      "Bash(git checkout:*)",
+      "Bash(git clean:*)",
+      "Bash(git rebase:*)",
+      "Bash(git config:*)",
+      "Bash(git -C /root/Notes config:*)",
+      "Bash(git -C ~/Notes config:*)",
+      "Bash(git -C /root/source/homelab config:*)",
+      "Bash(git -C /root/source/homeassistant config:*)",
+      "Bash(git -C ~/source/homelab config:*)",
+      "Bash(git -C ~/source/homeassistant config:*)",
+      "Bash(rm:*)",
+      "Bash(rm -rf:*)",
+      "Bash(curl:*)",
+      "Bash(wget:*)",
+      "Bash(ssh:*)",
+      "Bash(tea:*)",
+      "Bash(woodpecker:*)"
+    ]
+  }
+}
@@ -0,0 +1,5 @@
+You are running as the scheduled wiki-maintenance job in CI. The Notes vault is checked out at ~/Notes (the repository k.renner/notes); the homelab and homeassistant repositories are checked out under ~/source/. `kubectl` reads the live cluster with a read-only identity.
+
+Run the wiki-consolidate agent now for today's scheduled consolidation pass. Follow its instructions exactly: snapshot the vault to git first, abort if that fails, select today's rotating cohort of 20 from the state file, verify claims against both their sources and the live systems, act on the verdicts, write the dated report to ~/Notes/output/, update the state file, and snapshot again at the end. Do not push; the pipeline pushes the run's commits.
+
+Report the counts and the coverage progress when you are done.
@@ -1,5 +1,5 @@
 {
-  "//": "Permission scope for the wiki-maintenance agents (wiki-distill, wiki-audit) in CI.",
+  "//": "Permission scope for the wiki-maintenance agent (wiki-distill) in CI.",
  "//2": "The step's environment holds a Gitea push token, a TypeSafe key and the wiki-index",
  "//3": "client credential, so the Read tool is kept away from /proc (env dumps), /root outside",
  "//4": "the vault symlink target, /etc and the checkout's .git/. The vault itself is the work.",
@@ -15,6 +15,10 @@
      "Bash(/root/Notes/bin/wiki_ask.py:*)",
      "Bash(mv /root/Notes/raw/:*)",
      "Bash(mkdir -p /root/Notes/:*)",
+      "Bash(~/Notes/bin/wiki_rank.py:*)",
+      "Bash(~/Notes/bin/wiki_ask.py:*)",
+      "Bash(mv ~/Notes/raw/:*)",
+      "Bash(mkdir -p ~/Notes/:*)",
      "Bash(ls:*)",
      "Bash(date:*)",
      "Bash(wc:*)",
@@ -1,6 +1,6 @@
 FROM alpine:3.21@sha256:48b0309ca019d89d40f670aa1bc06e426dc0931948452e8491e3d65087abc07d

-ARG TALOS_VERSION=v1.13.9
+ARG TALOS_VERSION=v1.13.10
 ARG VELERO_VERSION=v1.16.0
 ARG KUBECTL_VERSION=v1.36.3
 ARG FLUX_VERSION=2.9.5
@@ -18,7 +18,7 @@ RUN apk add --no-cache \

 RUN curl -sfL -o /usr/local/bin/talosctl \
      "https://github.com/siderolabs/talos/releases/download/${TALOS_VERSION}/talosctl-linux-amd64" \
- && echo "7e1d4b7d5846964bdcf63a794e3c8161bb6ef2983d5ace58ea5322f3bf32a27e  /usr/local/bin/talosctl" | sha256sum -c - \
+ && echo "28e9ccd9aebd690fa03d8cf2582221d87ea38b2a1704f411d75a03589c58f2aa  /usr/local/bin/talosctl" | sha256sum -c - \
 && chmod +x /usr/local/bin/talosctl

 RUN curl -sfL -o /usr/local/bin/kubectl \
@@ -1,4 +1,4 @@
-# netbird-proxy-services: reconciles NetBird reverse-proxy services (public
+# netbird-service-sync: reconciles NetBird reverse-proxy services (public
 # domains -> TLS passthrough -> Traefik ClusterIP) from Ingresses carrying the
 # public label. Standard library only, so the image is the interpreter plus one
 # file; runs unprivileged as its own user.
@@ -1,5 +1,5 @@
 #!/usr/bin/env python3
-"""netbird-proxy-services: reconcile NetBird reverse-proxy services from Ingresses.
+"""netbird-service-sync: reconcile NetBird reverse-proxy services from Ingresses.

 Source of truth: every Ingress carrying the public label (dns.renner.uno/public).
 For each host on such an Ingress the NetBird management gets one reverse-proxy
@@ -30,11 +30,11 @@ import urllib.error
 import urllib.parse
 import urllib.request

-LOG = logging.getLogger("netbird-proxy-services")
+LOG = logging.getLogger("netbird-service-sync")

 NB = os.environ.get("NB_MANAGEMENT_URL", "https://netbird.renner.uno").rstrip("/")
 BAO_ADDR = os.environ.get("BAO_ADDR", "http://openbao.openbao.svc.cluster.local:8200").rstrip("/")
-BAO_ROLE = os.environ.get("BAO_ROLE", "netbird-proxy-services")
+BAO_ROLE = os.environ.get("BAO_ROLE", "netbird-service-sync")
 BAO_PAT_PATH = os.environ.get("BAO_PAT_PATH", "homelab/data/infrastructure/secret-rotators-bootstrap/netbird")
 BAO_PAT_KEY = os.environ.get("BAO_PAT_KEY", "pat")
 SERVICE_CIDR = os.environ.get("SERVICE_CIDR", "10.96.0.0/12")
@@ -156,7 +156,7 @@ def ensure_network(nb):
        if n["name"] == NETWORK_NAME:
            return n["id"]
    n = nb.post("networks", {"name": NETWORK_NAME,
-                             "description": "managed by netbird-proxy-services; public ingress via TLS passthrough"})
+                             "description": "managed by netbird-service-sync; public ingress via TLS passthrough"})
    LOG.info("created network %s (%s)", NETWORK_NAME, n["id"])
    return n["id"]

@@ -166,7 +166,7 @@ def ensure_resource(nb, net):
        if r.get("address") == SERVICE_CIDR:
            return r["id"]
    r = nb.post(f"networks/{net}/resources", {"name": "cluster service CIDR", "address": SERVICE_CIDR,
-                                              "description": "managed by netbird-proxy-services",
+                                              "description": "managed by netbird-service-sync",
                                              "enabled": True, "groups": []})
    LOG.info("created subnet resource %s (%s)", SERVICE_CIDR, r["id"])
    return r["id"]
@@ -14,7 +14,7 @@ FROM alpine:3.21@sha256:48b0309ca019d89d40f670aa1bc06e426dc0931948452e8491e3d650
 # kubernetes/talos/patches/cluster.yaml. The phase reads the version from that
 # file at run time and passes it to the provisioner; this pin only decides which
 # client talks to it.
-ARG TALOS_VERSION=v1.13.9
+ARG TALOS_VERSION=v1.13.10
 ARG KUBECTL_VERSION=v1.36.2
 ARG FLUX_VERSION=2.9.2
 ARG KOPIA_VERSION=0.23.1
@@ -49,7 +49,7 @@ RUN apk add --no-cache \

 RUN curl -sfL -o /usr/local/bin/talosctl \
      "https://github.com/siderolabs/talos/releases/download/${TALOS_VERSION}/talosctl-linux-amd64" \
- && echo "7e1d4b7d5846964bdcf63a794e3c8161bb6ef2983d5ace58ea5322f3bf32a27e  /usr/local/bin/talosctl" | sha256sum -c - \
+ && echo "28e9ccd9aebd690fa03d8cf2582221d87ea38b2a1704f411d75a03589c58f2aa  /usr/local/bin/talosctl" | sha256sum -c - \
 && chmod +x /usr/local/bin/talosctl

 RUN curl -sfL -o /usr/local/bin/kubectl \
@@ -1,117 +1,99 @@
 # Traffic classes on the k8s nodes

-Issue #576. Three of the four classes are separated; management is not.
+Issue #576. Five classes, each on its own VLAN.

 ## Where each class runs

 | class | link | mechanism |
 |---|---|---|
 | Ceph replication | `ceph-san`, VLAN 49 | the whole link is handed to the OSD pod (Multus `host-device`) |
-| world→pod | `ingress0`, VLAN 47 | BGP session sourced there, so the advertised next hop is that address |
-| pod→world | `ingress0`, VLAN 47 | routing rule on the pod source prefix → table 47 |
-| pod↔pod | `pod0`, VLAN 48 | routing rule on pod source **and** destination → table 48 |
-| management | `mgmt0`, VLAN 40 | not started |
+| pod↔pod | `pod0`, VLAN 48 | pod-sourced traffic is routed from table 47; bird puts the peers' pod CIDRs there via `pod0` |
+| world↔pod | `ingress0`, VLAN 47 | BGP session sourced there (next hop of every LoadBalancer VIP); pod egress takes table 47's default |
+| node↔node | `node0`, VLAN 46 | the node InternalIP and etcd live there; so do apiserver↔kubelet, cilium-health and host→remote-pod traffic |
+| world↔node | `mgmt0`, VLAN 40 | API VIP for kubectl, talosctl from outside, node egress (DNS, NTP, image pulls) |

 Hard constraint, set by the owner: inter-node and pod traffic never traverses
-the UDM. Everything above stays on the 10G switch.
+the UDM. Everything but world↔node stays on the 10G switch. VLAN 46 and 48 are
+statically addressed, without a gateway; VLAN 46 is isolated in UniFi.

-## Why policy routing rather than moving the node IP
+## Pod traffic: one table

-Cilium reaches a peer's pods at that peer's Kubernetes InternalIP, in native
-and tunnel mode alike. `--ipv4-node` is overwritten from the Node object, and
-`nodePort.directRoutingDevice` selects the BPF NodePort device, not a next hop.
-Both were established by test (#837, #838), not by reading.
+Two rules, one per family, send every packet with a pod source to table 47
+(`talos/patches/site-homelab.yaml`, priorities 1000/1001, below main). The
+table decides by longest prefix; no rule carves out exceptions.

-So the InternalIP decides where pod traffic goes — unless the lookup happens in
-a different table. The rules do that, and nothing about the node's identity
-moves. The routing decision runs before Cilium's masquerade hook, so a rule
-matching the pod source sees the pod address, and the masquerade then rewrites
-it to the egress link's.
+bird (`kubernetes/infrastructure/bird/`) owns table 47 and fills it with:

-## Rule order
+- the node's attached networks — `mgmt0`, `ingress0`, `pod0`, `node0` —
+  taken from the interfaces;
+- the peers' pod CIDRs, learned over an iBGP mesh on `pod0`, next hop the
+  peer's `pod0` address;
+- the node's own pod CIDR via `cilium_host`, originated rather than read back
+  from the kernel (reading it back made BIRD depend on the shape of Cilium's
+  route, and KRT rejects the IPv4 one);
+- the egress default via the UDM on `ingress0`, from the node annotations
+  `net.renner.uno/egress-gw-v4/v6`.

-Lower priority runs first.
+`node0` has to be in the table: a pod reaching a node address — the
+`kubernetes` ClusterIP, which socket LB rewrites to the apiserver's advertise
+address, or a hostNetwork pod — would otherwise take the default to the UDM.

-| prio | match | table |
-|---|---|---|
-| 980 | src pod, dst pod | 48 |
-| 981 | src pod, dst pod (v4) | 48 |
-| 990 | dst `fd00::/8` | main |
-| 991 | dst `10.0.0.0/8` | main |
-| 1000 | src pod | 47 |
-| 1001 | src pod (v4) | 47 |
+The routing decision runs before Cilium's masquerade hook, so the rule sees the
+pod address and the masquerade then rewrites it to the egress link's.

-980/981 match source **and** destination. Source alone would also catch the
-node's own traffic to a pod, which would then leave with a `pod0` source — an
-address `CiliumNode.spec.addresses` does not list, so Cilium classifies it as
-`world` rather than `host`.
-
-990/991 are Cilium's `ipv4/ipv6NativeRoutingCIDR`, and they are what keeps
-cluster-internal traffic out of table 47. Without them a pod's packet to a
-service matches the source rule, finds only a default route, and leaves for the
-UDM — measured, in-cluster DNS stopped resolving.
-
-Traffic to a pod on the same node falls through on its own: table 48 holds only
-the peers' CIDRs, so the lookup misses and 990/991 send it to main.
-
-## Where each table comes from
-
- **47** — two default routes declared per node in `talos/patches/k8s-*.yaml`.
-  The IPv6 gateway is the UDM's configured address on the segment, not the
-  RA-learned link-local; `accept_ra_defrtr` stays `0` on `ingress0` so the RA
-  default does not land in main.
- **48** — filled by the `bird` DaemonSet: an iBGP mesh over `pod0`, one route
-  per peer pod CIDR with the peer's `pod0` address as next hop.
-
-## The bird mesh
-
-`kubernetes/infrastructure/bird/`. Each node advertises only its own pod CIDR
-and installs what it learns into table 48. Cilium's routes in main are
-untouched, which is why the two coexist.
+Node-originated traffic does not match the rules and uses main, where Cilium
+routes each peer's pod CIDR via the peer's InternalIP — on `node0`.

 No address is written into this repo twice. Each node publishes its `pod0`
-addresses as `nodeAnnotations`, beside the `VLANConfig` that assigns them; the
-pod CIDRs come from the Node objects. An init container renders both from the
-API with `kubectl -o go-template-file`, excludes the node itself with a field
-selector, and ends with `bird -p` so BIRD's own parser gates startup.
+addresses as `nodeAnnotations` beside the `VLANConfig` that assigns them; the
+pod CIDRs come from the Node objects. bird's init container renders both from
+the API and ends with `bird -p`, so BIRD's own parser gates startup.

-The node's own CIDR is **originated**, not read back from the kernel. Reading
-it back made BIRD depend on the shape Cilium installs, and the IPv4 route is
-rejected outright: its next hop sits inside the prefix it describes, which KRT
-refuses with `strange next-hop`. The IPv6 route is a device route and passed,
-so the failure was silent on one family only.
+## Node↔node
+
+`kubelet.nodeIP.validSubnets` and `cluster.etcd.advertisedSubnets` point at
+VLAN 46. Talos picks one InternalIP per family, IPv4 first — the service CIDR
+is IPv4, and the apiserver's advertise address must match it.
+
+The API VIP stays on `mgmt0` for kubectl. No in-cluster client uses it:
+kubelet, scheduler, controller-manager and Cilium go through KubePrism
+(`localhost:7445`), which balances over every address of every control-plane
+member; pods use the ClusterIP.
+
+Talos never updates an existing etcd member's peer URLs
+(siderolabs/talos#12720). Changing `advertisedSubnets` means, per member:
+stage the config, `talosctl etcd forfeit-leadership`, `talosctl etcd leave`,
+reboot. `leave` removes only `/var/lib/etcd`; the node rejoins as a learner
+and is promoted. EPHEMERAL does not need a wipe.
+
+## Moving node IPs: the mixed state
+
+While some nodes have their InternalIP on the new network and some on the old,
+host→remote-pod traffic leaves with a node address that is not an InternalIP:
+the kernel picks the source from the route to the peer's InternalIP. Cilium
+classifies that as `world`, and every ingress policy that admits only
+`host`/`remote-node` (rook-ceph, monitoring, …) drops it. Kernel Ceph clients
+lose the OSDs on the other side; RBD mounts hang.
+
+Bridge it for the duration with a temporary `CiliumClusterwideNetworkPolicy`
+admitting the node subnets of both networks, both families, with
+`enableDefaultDeny: false` so it restricts nothing else, and delete it once
+every node has moved. Keep all OSD nodes schedulable while one is drained:
+draining two takes Ceph below its data availability.

 ## Failure behaviour

-Benign at every layer. An empty or absent table 48 makes 980/981 fall through
-to main, and pod traffic returns to VLAN 40 — unseparated, not broken. The same
-holds if the DaemonSet cannot start, which is the case during a cold rebuild
-until Gitea serves its image.
-
-## Management, the remaining class
-
-Not started. The levers differ per service and none of them is the node IP:
-
-| service | can leave the node IP? | lever |
-|---|---|---|
-| etcd | yes | `cluster.etcd.advertisedSubnets`; needs a reboot per node |
-| kube-apiserver endpoint | yes | it is a VIP — `Layer2VIPConfig.link` |
-| Talos API | n/a | `apid` answers on every node address; a firewall question |
-| kubelet | no | it *is* the node IP |
-
-Two approaches were rejected by the owner and are not to be proposed again:
-moving the node IP and etcd to `pod0` (#838), and relabelling VLAN 40 as the
-cluster VLAN while moving etcd and the API VIP to a new management VLAN.
-Replacing the CNI (#839, Antrea) was shelved.
-
-Note that the reasoning which produced that list also concluded pod↔pod was
-blocked, which policy routing disproved. Treat it as a record of decisions, not
-as an exhaustive analysis.
+If bird stops, table 47 empties, the rules find nothing and fall through to
+main: pod traffic returns to the node networks, unseparated rather than
+misrouted. The same holds during a cold rebuild until Gitea serves bird's
+image. The default route lives in table 47, not in the Talos config, so a
+default cannot outlive the specific routes and send pod traffic to the UDM.

 ## Sources

-Cilium v1.20.1 `pkg/datapath/linux/node.go`, `pkg/node/types/node.go`,
-`pkg/node/sync/local_node_sync.go`; BIRD 2.17 user manual (kernel protocol,
-`learn all`, `strict bind`, `direct`); Talos v1.13 `RoutingRuleConfig` and
-`machine.nodeAnnotations`. PRs #831, #833, #835, #836, #837, #842, #852, #854,
-#855; closed unmerged: #830, #838, #839.
+Cilium v1.20 `pkg/datapath/linux/node.go`; BIRD 2.17 user manual (kernel
+protocol, `strict bind`, `direct`); Talos v1.13 `RoutingRuleConfig`,
+`machine.nodeAnnotations`, configuration patching (lists are appended,
+`$patch: delete`), etcd `LeaveCluster`; KubePrism docs;
+siderolabs/talos#12720. PRs #831, #833, #835, #836, #837, #842, #852, #854,
+#855, #857, #858, #918, #919, #922; closed unmerged: #830, #838, #839.
@@ -107,7 +107,7 @@ config, and the nodes are addressed by name.
 ## Step 1b — boot all three machines from install media

 ```
-https://factory.talos.dev/image/0f6fb8bd8bb53013e9743b7c7b099d96fb9c32fd2464ced32693c2dc0ea5feb9/v1.13.9/metal-amd64.iso
+https://factory.talos.dev/image/0f6fb8bd8bb53013e9743b7c7b099d96fb9c32fd2464ced32693c2dc0ea5feb9/v1.13.10/metal-amd64.iso
 ```

 Write it to a USB stick and boot each MS-01 from it. Each comes up in
@@ -2,7 +2,7 @@ apiVersion: networking.k8s.io/v1
 kind: Ingress
 metadata:
  name: gitea
-  namespace: apps
+  namespace: gitea
  labels:
    app.kubernetes.io/name: gitea
  annotations:
@@ -2,7 +2,7 @@ apiVersion: traefik.io/v1alpha1
 kind: IngressRouteTCP
 metadata:
  name: gitea-ssh
-  namespace: apps
+  namespace: gitea
  labels:
    app.kubernetes.io/name: gitea
 spec:
@@ -6,7 +6,6 @@ kind: Kustomization
 # and verified. See MIGRATION-gitea-woodpecker.md.
 resources:
  - statefulset.yaml
-  - pdb.yaml
  - service.yaml
  - ingressroute-ssh.yaml
  - ingress.yaml
@@ -1,28 +1,8 @@
-apiVersion: cilium.io/v2
-kind: CiliumNetworkPolicy
-metadata:
-  name: gitea-allow-metrics-scrape
-  namespace: apps
-spec:
-  # Prometheus (monitoring ns) scrapes gitea's /metrics on :3000. The apps
-  # default-deny-ingress only allows same-ns / home-assistant / host, so the
-  # scrape is dropped (context deadline exceeded). This became necessary only
-  # after the LXC->in-cluster cutover — the old static target was external, so
-  # no in-cluster policy applied. Scoped to the gitea pod + :3000.
-  endpointSelector:
-    matchLabels:
-      app.kubernetes.io/name: gitea
-  ingress:
-    - fromEndpoints:
-        - matchLabels:
-            "k8s:io.kubernetes.pod.namespace": monitoring
-      toPorts:
-        - ports:
-            - port: "3000"
-              protocol: TCP
---
+# Prometheus scrape and Traefik are admitted by the namespace baseline in
+# infrastructure/network-policies/gitea-ns.yaml.
+#
 # Secret rotation: the flux-git rotator (openbao ns) reaches gitea to mint a new
-# flux PAT. It uses the in-cluster Service (gitea.apps.svc:3000) — the public
+# flux PAT. It uses the in-cluster Service (gitea.gitea.svc:3000) — the public
 # gitea.renner.uno name hairpins off the ingress LB from a pod (same failure as
 # pve.renner.uno). Scoped to the gitea pod :3000 and the rotator label; the
 # egress half is rotator-gitea-flux-egress in network-policies/openbao-ns.yaml.
@@ -30,7 +10,7 @@ apiVersion: cilium.io/v2
 kind: CiliumNetworkPolicy
 metadata:
  name: gitea-allow-flux-rotator
-  namespace: apps
+  namespace: gitea
 spec:
  endpointSelector:
    matchLabels:
@@ -54,7 +34,7 @@ apiVersion: cilium.io/v2
 kind: CiliumNetworkPolicy
 metadata:
  name: gitea-allow-review-bot-rotator
-  namespace: apps
+  namespace: gitea
 spec:
  endpointSelector:
    matchLabels:
@@ -1,23 +0,0 @@
---
-# Eviction protection for the single-replica Gitea StatefulSet.
-#
-# Gitea runs on SQLite (single writer) and its repos + DB live on a single
-# ReadWriteOnce PVC, so it is deliberately replicas:1 and cannot fail over.
-# With no PDB, a node drain / maintenance evicts the pod abruptly, mid-write,
-# with zero coordination. minAvailable:1 on a 1-replica workload makes the
-# eviction API refuse *voluntary* evictions: `kubectl drain` blocks and
-# surfaces this pod, forcing a conscious decision (drain with
-# --disable-eviction, or scale down first) instead of yanking the live DB.
-# It does NOT block involuntary disruptions (node crash, OOM).
-apiVersion: policy/v1
-kind: PodDisruptionBudget
-metadata:
-  name: gitea
-  namespace: apps
-  labels:
-    app.kubernetes.io/name: gitea
-spec:
-  minAvailable: 1
-  selector:
-    matchLabels:
-      app.kubernetes.io/name: gitea
@@ -2,7 +2,7 @@ apiVersion: v1
 kind: Service
 metadata:
  name: gitea
-  namespace: apps
+  namespace: gitea
  labels:
    app.kubernetes.io/name: gitea
 spec:
@@ -2,7 +2,7 @@ apiVersion: apps/v1
 kind: StatefulSet
 metadata:
  name: gitea
-  namespace: apps
+  namespace: gitea
  labels:
    app.kubernetes.io/name: gitea
 spec:
@@ -152,7 +152,7 @@ spec:
          # NOTE: volumeClaimTemplates is immutable on an existing StatefulSet,
          # so bumping this value cannot be `kubectl apply`-ed in place — Flux
          # will reject the whole object. Landing this size change requires a
-          # one-time `kubectl delete statefulset gitea -n apps --cascade=orphan`
+          # one-time `kubectl delete statefulset gitea -n gitea --cascade=orphan`
          # (pod + PVC keep running) so Flux re-creates the STS from this
          # template and re-adopts them. After the cleanup_packages GC reclaims
          # the ~10G of orphaned blobs, real usage is ~9G, so 40Gi is headroom.
@@ -4,7 +4,9 @@ metadata:
  name: gf-pg
  namespace: monitoring
 spec:
-  instances: 1
+  instances: 2
+  smartShutdownTimeout: 10
+  stopDelay: 60
  imageName: ghcr.io/cloudnative-pg/postgresql:16
  storage:
    storageClass: ceph-block
@@ -30,7 +30,9 @@ metadata:
  name: ha-pg
  namespace: home-assistant
 spec:
-  instances: 1
+  instances: 2
+  smartShutdownTimeout: 10
+  stopDelay: 60
  imageName: ghcr.io/cloudnative-pg/postgresql:16
  storage:
    storageClass: ceph-block
@@ -124,8 +126,7 @@ spec:
      # padding rather than data.
      #
      # The cost is RPO: up to 30min of WAL is unrecoverable if this cluster and
-      # its volume are lost outright. This is a single-instance cluster, so the
-      # archive is the only off-node copy.
+      # its volumes are lost outright.
      archive_timeout: "30min"
      # Fewer checkpoints => fewer full-page writes, since a page is only
      # fully-logged on its first modification after a checkpoint.
@@ -59,8 +59,7 @@ spec:
      # padding rather than data.
      #
      # The cost is RPO: up to 30min of WAL is unrecoverable if this cluster and
-      # its volume are lost outright. This is a single-instance cluster, so the
-      # archive is the only off-node copy.
+      # its volumes are lost outright.
      archive_timeout: "30min"
      # Write-reduction tuning. These databases are largely idle, and the
      # Postgres defaults favour throughput over storage that wears out.
@@ -61,6 +61,9 @@ spec:
      enabled: true
      storageClass: ceph-block
      size: 50Gi
+    # A minAvailable:1 PDB on a single replica blocks every node drain.
+    pdb:
+      create: false
    # Bootstrap admin/org/bucket on the fresh PVC. Admin password + token come
    # from the influxdb-bootstrap Secret (OpenBao via ESO), deterministic across
    # rebuilds so the bootstrap token is the live admin token. `influx setup`
@@ -3,8 +3,8 @@ kind: Kustomization
 resources:
  - gatus/
  - external-dns/
-  - gitea/
  - ical-proxy/
+  # gitea/ has its own `gitea` Kustomization (clusters/homelab/gitea.yaml, #878).
  # woodpecker/ moved to its own dedicated `woodpecker` Kustomization
  # (clusters/homelab/woodpecker.yaml) so its wp-pg CNPG cluster's barman
  # dependency + Cluster health-gating don't contaminate this shared apps KS.
@@ -57,10 +57,14 @@ spec:
          # the ClusterSecretStore goes InvalidProviderConfig and no
          # ExternalSecret can refresh.
          #
-          # No window: below quorum the secret store is down and every
-          # ExternalSecret stalls with it, so waiting only delays the page.
+          # Short window: readiness is only a proxy for quorum. A one-second
+          # node NotReady flap makes the node lifecycle controller mark that
+          # node's pods not-ready for ~30s while raft keeps its leader and
+          # serves (issue #920, during a rolling reboot with one replica
+          # already evicted). A really sealed or crashed majority persists,
+          # and existing Secrets keep their last values, so 2m costs little.
          expr: max by (namespace, statefulset) (kube_statefulset_status_replicas_ready{namespace="openbao", statefulset="openbao"}) < 2
-          for: 0m
+          for: 2m
          labels:
            severity: critical
          annotations:
@@ -91,12 +91,11 @@ spec:
        # Additional scrape targets not covered by ServiceMonitors.
        additionalScrapeConfigs:
          # Gitea exposes /metrics anonymously on its main HTTP port. The
-          # `gitea_*` metrics power the gitea-overview dashboard. Gitea now runs
-          # in-cluster (StatefulSet in apps ns; LXC 114 retired), so scrape the
-          # ClusterIP Service directly instead of the dead LXC hostname.
+          # `gitea_*` metrics power the gitea-overview dashboard. Scrapes the
+          # ClusterIP Service of the StatefulSet in the gitea namespace.
          - job_name: gitea
            static_configs:
-              - targets: ["gitea.apps.svc.cluster.local:3000"]
+              - targets: ["gitea.gitea.svc.cluster.local:3000"]
            metrics_path: /metrics

    # Grafana
@@ -305,6 +304,10 @@ spec:
            - matchers:
                - alertname = "KubeCPUOvercommit"
              receiver: 'null'
+            # Helper alert that only drives the info inhibit rule below.
+            - matchers:
+                - alertname = "InfoInhibitor"
+              receiver: 'null'
            # Flux notification-controller posts info-severity events
            # here (e.g. image-automation commits). Keep them visible
            # but don't repeat-spam every 12h on the same event.
@@ -314,6 +317,23 @@ spec:
              group_wait: 1m
              group_interval: 10m
              repeat_interval: 24h
+        # From kube-prometheus manifests/alertmanager-secret.yaml.
+        inhibit_rules:
+          - source_matchers:
+              - severity = "critical"
+            target_matchers:
+              - severity =~ "warning|info"
+            equal: ['namespace', 'alertname']
+          - source_matchers:
+              - severity = "warning"
+            target_matchers:
+              - severity = "info"
+            equal: ['namespace', 'alertname']
+          - source_matchers:
+              - alertname = "InfoInhibitor"
+            target_matchers:
+              - severity = "info"
+            equal: ['namespace']
        receivers:
          - name: email
            email_configs:
@@ -11,7 +11,6 @@ resources:
  - protonmail-bridge.yaml
  - ftp-scanner.yaml
  - syncthing.yaml
-  - pdb.yaml
  - network-policy.yaml
  # pprl-pg CNPG Postgres + data/media/consume/redis PVCs (moved here from the
  # retired seed-paperless KS; each carries prune: disabled so ownership
@@ -27,6 +27,8 @@ data:
  PAPERLESS_ALLOWED_HOSTS: "*"
  PAPERLESS_TIME_ZONE: "Europe/Berlin"
  PAPERLESS_OCR_LANGUAGE: "deu+eng"
+  # Default is cpu_count / TASK_WORKERS, 20 on k8s-1; OCR memory scales with it.
+  PAPERLESS_THREADS_PER_WORKER: "5"
  PAPERLESS_CONSUMER_RECURSIVE: "true"
  PAPERLESS_CONSUMER_DELETE_DUPLICATES: "true"
  PAPERLESS_CONSUMER_ENABLE_BARCODES: "true"
@@ -273,7 +275,7 @@ spec:
              cpu: 200m
              memory: 512Mi
            limits:
-              memory: 2Gi
+              memory: 4Gi
          securityContext:
            runAsNonRoot: false
            allowPrivilegeEscalation: true
@@ -1,37 +0,0 @@
---
-# Eviction protection for the single-replica stateful paperless sidecars.
-#
-# protonmail-bridge and syncthing each hold state on a single RWO PVC and run
-# replicas:1. minAvailable:1 makes the eviction API refuse voluntary evictions,
-# so a node drain blocks and surfaces these pods instead of evicting them
-# abruptly. Involuntary disruptions (node crash) are not affected.
-#
-# NOTE: the other single-replica paperless pods (paperless, redis, gotenberg,
-# tika, ftp-scanner) are intentionally omitted — they are caches / stateless
-# converters where a blocking PDB would only wedge node drains with no
-# recoverability benefit.
-apiVersion: policy/v1
-kind: PodDisruptionBudget
-metadata:
-  name: protonmail-bridge
-  namespace: paperless
-  labels:
-    app.kubernetes.io/name: protonmail-bridge
-spec:
-  minAvailable: 1
-  selector:
-    matchLabels:
-      app.kubernetes.io/name: protonmail-bridge
---
-apiVersion: policy/v1
-kind: PodDisruptionBudget
-metadata:
-  name: syncthing
-  namespace: paperless
-  labels:
-    app.kubernetes.io/name: syncthing
-spec:
-  minAvailable: 1
-  selector:
-    matchLabels:
-      app.kubernetes.io/name: syncthing
@@ -25,7 +25,7 @@ spec:
          type: RuntimeDefault
      containers:
        - name: redis
-          image: redis:8.10.1-alpine # {"$imagepolicy": "flux-system:redis-paperless"}
+          image: redis:8.10.2-alpine # {"$imagepolicy": "flux-system:redis-paperless"}
          args:
            - redis-server
            - --save
@@ -4,7 +4,7 @@ metadata:
  name: tr-pg
  namespace: trmnl
 spec:
-  instances: 1
+  instances: 2
  smartShutdownTimeout: 10
  stopDelay: 60
  # PG 18, NOT the repo-standard 16. Terminus ships a committed
@@ -47,8 +47,7 @@ spec:
      # padding rather than data.
      #
      # The cost is RPO: up to 30min of WAL is unrecoverable if this cluster and
-      # its volume are lost outright. This is a single-instance cluster, so the
-      # archive is the only off-node copy.
+      # its volumes are lost outright.
      archive_timeout: "30min"
      # Write-reduction tuning. These databases are largely idle, and the
      # Postgres defaults favour throughput over storage that wears out.
@@ -22,7 +22,9 @@ metadata:
    # migration deliberately withheld this while a rollback still existed.
    kustomize.toolkit.fluxcd.io/prune: disabled
 spec:
-  instances: 1
+  instances: 2
+  smartShutdownTimeout: 10
+  stopDelay: 60
  imageName: ghcr.io/cloudnative-pg/postgresql:16
  storage:
    storageClass: ceph-block
@@ -91,8 +93,7 @@ spec:
      # padding rather than data.
      #
      # The cost is RPO: up to 30min of WAL is unrecoverable if this cluster and
-      # its volume are lost outright. This is a single-instance cluster, so the
-      # archive is the only off-node copy.
+      # its volumes are lost outright.
      archive_timeout: "30min"
  resources:
    requests:
@@ -3,7 +3,6 @@ kind: Kustomization
 resources:
  - rbac.yaml
  - server.yaml
-  - pdb.yaml
  - agent.yaml
  - service.yaml
  - ingress.yaml
@@ -22,7 +21,7 @@ resources:
  # OpenBao access, no NetworkPolicy needed. server.yaml/agent.yaml keep their
  # Reloader annotations; externalsecret.yaml keeps the 10m refreshInterval.
  # --- namespace migration, stages 1-2 (issue #433) ---
-  # Stage 2 has moved the workloads above (server, agent, service, ingress, pdb,
+  # Stage 2 has moved the workloads above (server, agent, service, ingress,
  # rbac, secrets, policies) into the `woodpecker` namespace and pointed the
  # server's DSN at wp-pg-rw.woodpecker.svc. Stage 3 retired the `apps`
  # originals; these four now describe the surviving cluster. History and the
@@ -1,27 +0,0 @@
---
-# Eviction protection for the single-replica Woodpecker server StatefulSet.
-#
-# The server is SQLite-backed (single writer) on a single RWO PVC, so it is
-# deliberately replicas:1. minAvailable:1 makes the eviction API refuse
-# voluntary evictions so a node drain blocks and surfaces this pod rather
-# than yanking the live CI DB mid-write. Involuntary disruptions are not
-# affected.
-#
-# The selector MUST include app.kubernetes.io/component: server: the agent
-# Deployment in this same namespace shares app.kubernetes.io/name: woodpecker
-# (component: agent), and a name-only selector would wrongly also cover the
-# stateless, disposable agents.
-apiVersion: policy/v1
-kind: PodDisruptionBudget
-metadata:
-  name: woodpecker-server
-  namespace: woodpecker
-  labels:
-    app.kubernetes.io/name: woodpecker
-    app.kubernetes.io/component: server
-spec:
-  minAvailable: 1
-  selector:
-    matchLabels:
-      app.kubernetes.io/name: woodpecker
-      app.kubernetes.io/component: server
@@ -1946,6 +1946,7 @@ if phase_runs wait-apps; then
    # apps waits on infrastructure-ceph-cluster (Ceph mon/mgr/OSD bring-up
    # is genuinely 5–10 min on a fresh OSD-wiped cluster).
    wait_kustomization apps                "apps"                                                90
+    wait_kustomization gitea               "Gitea (incl. data-gitea-0 PVC)"                      90
    wait_kustomization influxdb            "InfluxDB (HR install + token Jobs)"                  600
    wait_kustomization woodpecker          "Woodpecker CI (incl. wp-pg CNPG backend)"            600
    wait_kustomization authentik           "Authentik (incl. ak-pg CNPG backend)"                600
@@ -0,0 +1,20 @@
+apiVersion: kustomize.toolkit.fluxcd.io/v1
+kind: Kustomization
+metadata:
+  name: gitea
+  namespace: flux-system
+spec:
+  # Gitea in its own `gitea` namespace (#878), formerly under the shared `apps` KS.
+  interval: 10m
+  retryInterval: 2m
+  timeout: 15m
+  sourceRef:
+    kind: GitRepository
+    name: flux-system
+  path: ./kubernetes/apps/gitea
+  prune: true
+  wait: true
+  dependsOn:
+    - name: infrastructure
+    - name: infrastructure-crds
+    - name: infrastructure-ceph-cluster
@@ -23,7 +23,7 @@ protocol device { scan time 10; }
 # down. The usual advice is not to export these to the kernel because it
 # maintains them itself — true of main, not of a table the kernel never fills.
 protocol direct attached {
-  interface "mgmt0", "ingress0", "pod0";
+  interface "mgmt0", "ingress0", "pod0", "node0";
  ipv4 { table master4; };
  ipv6 { table master6; };
 }
@@ -48,8 +48,8 @@ spec:
        tier: node
        app: bird
    spec:
-      # Carries pod east-west once the routing rules land. Without it table 48
-      # stays empty and the rules fall through to main, i.e. back to VLAN 40.
+      # Owns table 47, where all pod traffic is routed. Without it the table
+      # stays empty and the rules fall through to main, unseparated.
      priorityClassName: system-node-critical
      # Writes host routes and peers on host addresses.
      hostNetwork: true
@@ -118,14 +118,21 @@ spec:
      # traffic out to the VIP and back (which fails with L2 announcements).
      lbExternalClusterIP: true
    cni:
-      # Multus (kubernetes/infrastructure/multus) runs alongside Cilium and
-      # writes 00-multus.conf to /etc/cni/net.d so kubelet picks multus
-      # first; multus then delegates to Cilium's 05-cilium.conflist via
-      # clusterNetwork. The default `exclusive: true` makes cilium-cni's
-      # install step rename anything alphabetically before its own config
-      # to *.cilium_bak on every reconcile, which kept nuking multus.
-      # Setting it false lets multus's config coexist.
+      # Multus is the only writer of /etc/cni/net.d (#832): Cilium's conflist
+      # goes to /etc/cni/net.d/multus/, which the container runtime does not
+      # read, and Multus delegates to it via clusterNetwork. Without this a
+      # pod created after a reboot, before Multus rewrote 00-multus.conf, came
+      # up with Cilium alone. See kubernetes/infrastructure/multus.
+      confPath: /etc/cni/net.d/multus
+      # The default `exclusive: true` renames other configs in the conf dir to
+      # *.cilium_bak on every reconcile, which kept nuking multus.
      exclusive: false
+    # One agent at a time: the chart default of 2 restarts two of three
+    # nodes' datapath at once.
+    updateStrategy:
+      type: RollingUpdate
+      rollingUpdate:
+        maxUnavailable: 1
    # Talos-specific settings
    securityContext:
      capabilities:
@@ -1,11 +1,16 @@
 apiVersion: source.toolkit.fluxcd.io/v1
-kind: HelmRepository
+kind: OCIRepository
 metadata:
  name: k8s-gateway
  namespace: kube-system
 spec:
  interval: 24h
-  url: https://k8s-gateway.github.io/k8s_gateway/
+  url: oci://codeberg.org/k8s-gateway/charts/k8s-gateway
+  ref:
+    semver: ">=2.4.0"
+  layerSelector:
+    mediaType: "application/vnd.cncf.helm.chart.content.v1.tar+gzip"
+    operation: copy
 ---
 # In-cluster authoritative resolver for renner.uno.
 #
@@ -28,14 +33,10 @@ metadata:
  namespace: kube-system
 spec:
  interval: 30m
-  chart:
-    spec:
-      chart: k8s-gateway
-      version: ">=2.4.0"
-      sourceRef:
-        kind: HelmRepository
-        name: k8s-gateway
-        namespace: kube-system
+  chartRef:
+    kind: OCIRepository
+    name: k8s-gateway
+    namespace: kube-system
  install:
    remediation:
      retries: 3
@@ -19,7 +19,7 @@ resources:
  - cloudnative-pg/
  - mariadb-operator/
  - netbird-ingress/
-  - netbird-proxy-services/
+  - netbird-service-sync/
  - coredns-config/
  - k8s-gateway/
  - external-dns-unifi/
@@ -111,10 +111,18 @@ metadata:
  name: multus
  namespace: kube-system
 ---
+# Multus is the only writer of /etc/cni/net.d (#832). Cilium writes its conflist
+# to /etc/cni/net.d/multus/, a subdirectory the container runtime does not read
+# (libcni skips directories); Multus picks it up there via multusAutoconfigDir
+# and writes 00-multus.conf. Until that file exists the runtime has no network
+# config at all, so no pod can be created with Cilium alone after a reboot.
+#
+# The "-subdir" names date from the node-by-node rollout; renaming would roll
+# Multus again for nothing.
 kind: ConfigMap
 apiVersion: v1
 metadata:
-  name: multus-daemon-config
+  name: multus-daemon-config-subdir
  namespace: kube-system
  labels:
    tier: node
@@ -127,7 +135,7 @@ data:
        "logLevel": "verbose",
        "logToStderr": true,
        "cniConfigDir": "/host/etc/cni/net.d",
-        "multusAutoconfigDir": "/host/etc/cni/net.d",
+        "multusAutoconfigDir": "/host/etc/cni/net.d/multus",
        "multusConfigFile": "auto",
        "socketDir": "/host/run/multus/"
    }
@@ -135,16 +143,16 @@ data:
 apiVersion: apps/v1
 kind: DaemonSet
 metadata:
-  name: kube-multus-ds
+  name: kube-multus-ds-subdir
  namespace: kube-system
  labels:
    tier: node
    app: multus
-    name: multus
+    name: multus-subdir
 spec:
  selector:
    matchLabels:
-      name: multus
+      name: multus-subdir
  updateStrategy:
    type: RollingUpdate
  template:
@@ -152,7 +160,7 @@ spec:
      labels:
        tier: node
        app: multus
-        name: multus
+        name: multus-subdir
    spec:
      # Upstream ships this DaemonSet with no priority class, so it runs at 0
      # and can preempt nothing. Any event that makes the pod need a fresh
@@ -224,6 +232,10 @@ spec:
          volumeMounts:
            - name: cni
              mountPath: /host/etc/cni/net.d
+            # Exists before Multus starts, so it waits for Cilium's conflist
+            # instead of exiting on a missing directory.
+            - name: cni-cilium-conf
+              mountPath: /host/etc/cni/net.d/multus
            # multus-daemon expects that cnibin path must be identical between pod and container host.
            # e.g. if the cni bin is in '/opt/cni/bin' on the container host side, then it should be mount to '/opt/cni/bin' in multus-daemon,
            # not to any other directory, like '/opt/bin' or '/usr/bin'.
@@ -279,6 +291,10 @@ spec:
        - name: cni
          hostPath:
            path: /etc/cni/net.d
+        - name: cni-cilium-conf
+          hostPath:
+            path: /etc/cni/net.d/multus
+            type: DirectoryOrCreate
        - name: cnibin
          hostPath:
            path: /opt/cni/bin
@@ -287,7 +303,7 @@ spec:
            path: /
        - name: multus-daemon-config
          configMap:
-            name: multus-daemon-config
+            name: multus-daemon-config-subdir
            items:
              - key: daemon-config.json
                path: daemon-config.json
@@ -39,9 +39,11 @@ spec:
        name: whereabouts-reconciler
    spec:
      serviceAccountName: whereabouts
-      # Every node here is a control plane node.
+      # Every node here is a control plane node. Keyed: a keyless Exists would
+      # also tolerate node.kubernetes.io/unschedulable and ignore cordons.
      tolerations:
-        - operator: Exists
+        - key: node-role.kubernetes.io/control-plane
+          operator: Exists
          effect: NoSchedule
      containers:
        - name: whereabouts-reconciler
@@ -179,6 +179,22 @@ kind: Namespace
 metadata:
  name: woodpecker
 ---
+# Gitea, moving out of the shared `apps` namespace (issue #878). Declared here
+# like `woodpecker` above, so it exists before infrastructure-crds applies its
+# CiliumNetworkPolicies. No pod-security labels: the cluster default applies.
+apiVersion: v1
+kind: Namespace
+metadata:
+  name: gitea
+---
+# netbird-service-sync, moving out of `openbao` (issue #876). Declared here like
+# `woodpecker` above, so it exists before infrastructure-crds applies its
+# CiliumNetworkPolicies. No pod-security labels: the cluster default applies.
+apiVersion: v1
+kind: Namespace
+metadata:
+  name: netbird-service-sync
+---
 # Local LLM inference (kubernetes/apps/llama-server). Declared here, like
 # `woodpecker` above, so the `infrastructure` KS creates it before
 # `infrastructure-crds` applies the namespace's CiliumNetworkPolicies.
@@ -1,28 +1,28 @@
 # Reconciles NetBird reverse-proxy services from Ingresses that carry
 # dns.renner.uno/public: one TLS-passthrough service per host on :443,
 # targeting Traefik's ClusterIP through a subnet resource over the service
-# CIDR (ci/images/netbird-proxy-services/controller.py).
+# CIDR (ci/images/netbird-service-sync/controller.py).
 #
-# openbao namespace, like the rotators: its default-deny already allows egress
-# to OpenBao, DNS, the apiserver and `world` (netbird.renner.uno), and the
-# NetBird PAT is read from OpenBao on every pass through the
-# netbird-proxy-services role — policies/netbird-proxy-services.hcl, applied
-# once by hand like the rotator roles. Until that role exists the pod logs an
-# auth error each minute and does nothing.
+# Own namespace; its egress (DNS, apiserver, OpenBao, NetBird API) is in
+# network-policies/netbird-service-sync-ns.yaml. The NetBird PAT is read from
+# OpenBao on every pass through the netbird-service-sync role —
+# policies/netbird-service-sync.hcl, applied once by hand like the rotator
+# roles. Until that role exists the pod logs an auth error each minute and does
+# nothing.
 apiVersion: v1
 kind: ServiceAccount
 metadata:
-  name: netbird-proxy-services
-  namespace: openbao
+  name: netbird-service-sync
+  namespace: netbird-service-sync
  labels:
-    app.kubernetes.io/name: netbird-proxy-services
+    app.kubernetes.io/name: netbird-service-sync
 ---
 # Read-only, cluster-wide: public Ingresses live in many namespaces, the
 # Traefik Service in ingress/.
 apiVersion: rbac.authorization.k8s.io/v1
 kind: ClusterRole
 metadata:
-  name: netbird-proxy-services-read
+  name: netbird-service-sync-read
 rules:
  - apiGroups: ["networking.k8s.io"]
    resources: ["ingresses"]
@@ -34,23 +34,23 @@ rules:
 apiVersion: rbac.authorization.k8s.io/v1
 kind: ClusterRoleBinding
 metadata:
-  name: netbird-proxy-services-read
+  name: netbird-service-sync-read
 roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: ClusterRole
-  name: netbird-proxy-services-read
+  name: netbird-service-sync-read
 subjects:
  - kind: ServiceAccount
-    name: netbird-proxy-services
-    namespace: openbao
+    name: netbird-service-sync
+    namespace: netbird-service-sync
 ---
 apiVersion: apps/v1
 kind: Deployment
 metadata:
-  name: netbird-proxy-services
-  namespace: openbao
+  name: netbird-service-sync
+  namespace: netbird-service-sync
  labels:
-    app.kubernetes.io/name: netbird-proxy-services
+    app.kubernetes.io/name: netbird-service-sync
 spec:
  replicas: 1
  # One writer against the management API at a time.
@@ -58,13 +58,13 @@ spec:
    type: Recreate
  selector:
    matchLabels:
-      app.kubernetes.io/name: netbird-proxy-services
+      app.kubernetes.io/name: netbird-service-sync
  template:
    metadata:
      labels:
-        app.kubernetes.io/name: netbird-proxy-services
+        app.kubernetes.io/name: netbird-service-sync
    spec:
-      serviceAccountName: netbird-proxy-services
+      serviceAccountName: netbird-service-sync
      securityContext:
        runAsNonRoot: true
        # Numeric, or the kubelet refuses the image's named user with
@@ -80,7 +80,7 @@ spec:
          # image directory (branch and master alike) and the registry keeps
          # only the newest manifest per tag, so the digest must come from
          # the MASTER build of the code change.
-          image: gitea.renner.uno/gitea_admin/netbird-proxy-services:2026-09-07@sha256:caffa308d802c6fcc6aea919ac94657c08b993651baa6b8d18de395bcc40523e
+          image: gitea.renner.uno/gitea_admin/netbird-service-sync:2026-09-23@sha256:64f6057c508640e1dbc50457dc26e9b2de058a111fd46ca096021d7444b5bbff
          securityContext:
            allowPrivilegeEscalation: false
            readOnlyRootFilesystem: true
@@ -92,7 +92,7 @@ spec:
            - name: BAO_ADDR
              value: http://openbao.openbao.svc.cluster.local:8200
            - name: BAO_ROLE
-              value: netbird-proxy-services
+              value: netbird-service-sync
            # Must match cluster.network.serviceSubnets in talos/patches/cluster.yaml.
            - name: SERVICE_CIDR
              value: "10.96.0.0/12"
@@ -167,7 +167,7 @@ spec:
            "k8s:io.kubernetes.pod.namespace": ingress
            app.kubernetes.io/name: traefik
        - matchLabels:
-            "k8s:io.kubernetes.pod.namespace": apps
+            "k8s:io.kubernetes.pod.namespace": gitea
            app.kubernetes.io/name: gitea
      toPorts:
        - ports:
@@ -45,16 +45,6 @@ spec:
        - ports:
            - port: "1883"
              protocol: TCP
-    # gitea delivers the notes repo's push webhook to wiki-index (ai ns) on :8080.
-    # wiki-index's own ingress policy scopes this to the gitea pod.
-    - toEndpoints:
-        - matchLabels:
-            "k8s:io.kubernetes.pod.namespace": ai
-            app.kubernetes.io/name: wiki-index
-      toPorts:
-        - ports:
-            - port: "8080"
-              protocol: TCP
    - toEntities:
        - host
        - remote-node
@@ -0,0 +1,104 @@
+---
+# Baseline policy for the `gitea` namespace (issue #878).
+#
+# endpointSelector: {} — every rule applies to every pod in the namespace.
+# Workload-scoped grants go in their own policy with their own selector.
+apiVersion: cilium.io/v2
+kind: CiliumNetworkPolicy
+metadata:
+  name: default-deny
+  namespace: gitea
+spec:
+  endpointSelector: {}
+  ingress:
+    - fromEndpoints:
+        - matchLabels:
+            "k8s:io.kubernetes.pod.namespace": gitea
+    # Prometheus scrape of /metrics.
+    - fromEndpoints:
+        - matchLabels:
+            "k8s:io.kubernetes.pod.namespace": monitoring
+    # Kubelet probes and hostNetwork siblings.
+    - fromEntities:
+        - host
+        - remote-node
+    # ICMP from any cluster pod; required once an L4 policy selects the pod.
+    # See woodpecker-ns.yaml for the rationale. INGRESS ONLY.
+    - fromEndpoints:
+        - matchExpressions:
+            - key: io.kubernetes.pod.namespace
+              operator: Exists
+      icmps:
+        - fields:
+            - { type: 0,   family: IPv4 }
+            - { type: 3,   family: IPv4 }
+            - { type: 8,   family: IPv4 }
+            - { type: 11,  family: IPv4 }
+            - { type: 12,  family: IPv4 }
+            - { type: 1,   family: IPv6 }
+            - { type: 2,   family: IPv6 }
+            - { type: 3,   family: IPv6 }
+            - { type: 4,   family: IPv6 }
+            - { type: 128, family: IPv6 }
+            - { type: 129, family: IPv6 }
+            - { type: 133, family: IPv6 }
+            - { type: 134, family: IPv6 }
+            - { type: 135, family: IPv6 }
+            - { type: 136, family: IPv6 }
+            - { type: 143, family: IPv6 }
+  egress:
+    - toEndpoints:
+        - matchLabels:
+            "k8s:io.kubernetes.pod.namespace": gitea
+    # Webhooks to hosts served by Traefik (ci.renner.uno).
+    - toEndpoints:
+        - matchLabels:
+            "k8s:io.kubernetes.pod.namespace": ingress
+            app.kubernetes.io/name: traefik
+    - toEndpoints:
+        - matchLabels:
+            "k8s:io.kubernetes.pod.namespace": kube-system
+            k8s-app: kube-dns
+      toPorts:
+        - ports:
+            - port: "53"
+              protocol: UDP
+            - port: "53"
+              protocol: TCP
+    # Push webhook of the notes repo to wiki-index.
+    - toEndpoints:
+        - matchLabels:
+            "k8s:io.kubernetes.pod.namespace": ai
+            app.kubernetes.io/name: wiki-index
+      toPorts:
+        - ports:
+            - port: "8080"
+              protocol: TCP
+    # Push mirror to GitHub and other external targets.
+    - toEntities:
+        - host
+        - remote-node
+        - world
+---
+# Traefik serves git.renner.uno / gitea.renner.uno (:3000) and the gitea-ssh
+# entrypoint (:2222). Lives beside the baseline so both land in one reconcile.
+apiVersion: cilium.io/v2
+kind: CiliumNetworkPolicy
+metadata:
+  name: gitea-allow-traefik
+  namespace: gitea
+spec:
+  endpointSelector:
+    matchLabels:
+      app.kubernetes.io/name: gitea
+  ingress:
+    - fromEndpoints:
+        - matchLabels:
+            "k8s:io.kubernetes.pod.namespace": ingress
+            app.kubernetes.io/name: traefik
+      toPorts:
+        - ports:
+            - port: "3000"
+              protocol: TCP
+            - port: "2222"
+              protocol: TCP
@@ -22,6 +22,8 @@ resources:
  - external-secrets-ns.yaml
  - woodpecker-ci-ns.yaml
  - woodpecker-ns.yaml
+  - gitea-ns.yaml
+  - netbird-service-sync-ns.yaml
  - trmnl-ns.yaml
  - wordpress-ns.yaml
  - ai-ns.yaml
@@ -0,0 +1,43 @@
+---
+# Baseline policy for the `netbird-service-sync` namespace (issue #876).
+# The controller serves nothing; it only calls out: the Kubernetes API
+# (Ingress/Service reads), OpenBao (NetBird PAT) and the NetBird management API.
+apiVersion: cilium.io/v2
+kind: CiliumNetworkPolicy
+metadata:
+  name: default-deny
+  namespace: netbird-service-sync
+spec:
+  endpointSelector: {}
+  ingress:
+    - fromEntities:
+        - host
+        - remote-node
+  egress:
+    - toEndpoints:
+        - matchLabels:
+            "k8s:io.kubernetes.pod.namespace": kube-system
+            k8s-app: kube-dns
+      toPorts:
+        - ports:
+            - port: "53"
+              protocol: UDP
+            - port: "53"
+              protocol: TCP
+    - toEntities:
+        - kube-apiserver
+    - toEndpoints:
+        - matchLabels:
+            "k8s:io.kubernetes.pod.namespace": openbao
+            app.kubernetes.io/name: openbao
+      toPorts:
+        - ports:
+            - port: "8200"
+              protocol: TCP
+    # NetBird management API (netbird.renner.uno).
+    - toEntities:
+        - world
+      toPorts:
+        - ports:
+            - port: "443"
+              protocol: TCP
@@ -41,6 +41,16 @@ spec:
        - ports:
            - port: "8200"
              protocol: TCP
+    # netbird-service-sync logs in via auth/kubernetes/role/netbird-service-sync
+    # and reads the NetBird PAT on every pass.
+    - fromEndpoints:
+        - matchLabels:
+            "k8s:io.kubernetes.pod.namespace": netbird-service-sync
+            app.kubernetes.io/name: netbird-service-sync
+      toPorts:
+        - ports:
+            - port: "8200"
+              protocol: TCP
    # (Secret-rotation CronJobs run in THIS namespace since 2026-07-18, so their
    # OpenBao access is covered by the intra-namespace ingress rule above — no
    # per-app rule needed.)
@@ -127,7 +137,7 @@ spec:
 # Egress complement to gitea-allow-flux-rotator (apps/gitea/networkpolicy.yaml).
 # The flux-git rotator mints a PAT against the in-cluster gitea Service; the
 # openbao default-deny only permits TCP egress to openbao/kube-dns/apiserver/
-# world, and gitea (apps ns) is in-cluster, so it needs an explicit allow.
+# world, and gitea (gitea ns) is in-cluster, so it needs an explicit allow.
 # Scoped to the flux-git rotator label and gitea :3000.
 apiVersion: cilium.io/v2
 kind: CiliumNetworkPolicy
@@ -142,7 +152,7 @@ spec:
  egress:
    - toEndpoints:
        - matchLabels:
-            "k8s:io.kubernetes.pod.namespace": apps
+            "k8s:io.kubernetes.pod.namespace": gitea
            app.kubernetes.io/name: gitea
      toPorts:
        - ports:
@@ -150,10 +160,10 @@ spec:
              protocol: TCP
 ---
 # Egress for the review-bot rotator. Its CronJob has two steps: the init step
-# mints review-bot's PAT against gitea (apps :3000), the main step upserts the
+# mints review-bot's PAT against gitea (gitea :3000), the main step upserts the
 # Woodpecker org secret against the woodpecker server (woodpecker :8000). Both run
 # under the same pod labels, so one egress policy covers both hops. Complements
-# gitea-allow-review-bot-rotator (apps) + woodpecker-allow-rotators (woodpecker).
+# gitea-allow-review-bot-rotator (gitea) + woodpecker-allow-rotators (woodpecker).
 apiVersion: cilium.io/v2
 kind: CiliumNetworkPolicy
 metadata:
@@ -167,7 +177,7 @@ spec:
  egress:
    - toEndpoints:
        - matchLabels:
-            "k8s:io.kubernetes.pod.namespace": apps
+            "k8s:io.kubernetes.pod.namespace": gitea
            app.kubernetes.io/name: gitea
      toPorts:
        - ports:
@@ -1,23 +0,0 @@
-# OpenBao policy: netbird-proxy-services
-#
-# The reverse-proxy-services controller (Deployment netbird-proxy-services in
-# the openbao namespace) reads the NetBird PAT on every reconcile pass and
-# calls the management API with it. Read-only on that one path; it shares the
-# bootstrap credential with rotator-netbird but not that role's PATCH rights.
-#
-# NOT applied by Flux. Apply ONCE with an admin token:
-#
-#   bao policy write netbird-proxy-services kubernetes/infrastructure/openbao/policies/netbird-proxy-services.hcl
-#   bao write auth/kubernetes/role/netbird-proxy-services \
-#     bound_service_account_names=netbird-proxy-services \
-#     bound_service_account_namespaces=openbao \
-#     policies=netbird-proxy-services ttl=15m
-#
-# The PAT itself is the one rotator-netbird already uses
-# (homelab/infrastructure/secret-rotators-bootstrap/netbird, key `pat`).
-#
-# Do NOT set token_no_default_policy=true (kv-v2 preflight needs the default policy).
-
-path "homelab/data/infrastructure/secret-rotators-bootstrap/netbird" {
-  capabilities = ["read"]
-}
@@ -0,0 +1,24 @@
+# OpenBao policy: netbird-service-sync
+#
+# The controller that syncs public Ingresses into NetBird reverse-proxy services
+# (Deployment netbird-service-sync in the netbird-service-sync namespace) reads
+# the NetBird PAT on every reconcile pass and calls the management API with it.
+# Read-only on that one path; it shares the bootstrap credential with
+# rotator-netbird but not that role's PATCH rights.
+#
+# NOT applied by Flux. Apply ONCE with an admin token:
+#
+#   bao policy write netbird-service-sync kubernetes/infrastructure/openbao/policies/netbird-service-sync.hcl
+#   bao write auth/kubernetes/role/netbird-service-sync \
+#     bound_service_account_names=netbird-service-sync \
+#     bound_service_account_namespaces=netbird-service-sync \
+#     policies=netbird-service-sync ttl=15m
+#
+# The PAT itself is the one rotator-netbird already uses
+# (homelab/infrastructure/secret-rotators-bootstrap/netbird, key `pat`).
+#
+# Do NOT set token_no_default_policy=true (kv-v2 preflight needs the default policy).
+
+path "homelab/data/infrastructure/secret-rotators-bootstrap/netbird" {
+  capabilities = ["read"]
+}
@@ -6,7 +6,7 @@
 # newest. A token that fails verification is deleted and the run aborts without
 # storing, so Flux is never stranded.
 #
-# gitea is reached at the in-cluster Service gitea.apps.svc:3000 — the public
+# gitea is reached at the in-cluster Service gitea.gitea.svc:3000 — the public
 # name hairpins off the ingress LB from a pod. Needs BOTH netpol halves
 # (gitea-allow-flux-rotator + rotator-gitea-flux-egress).
 #
@@ -76,7 +76,7 @@ spec:
                - name: BAO_PATH
                  value: homelab/infrastructure/flux-git
                - name: GITEA_URL
-                  value: http://gitea.apps.svc.cluster.local:3000
+                  value: http://gitea.gitea.svc.cluster.local:3000
                - name: GITEA_VERIFY_REPO
                  value: k.renner/homelab
                - name: GITEA_SCOPES
@@ -24,7 +24,7 @@
 # synced. Quarterly cadence + backoffLimit:1 makes a two-cycle failure unlikely,
 # but a failed sync should be fixed before the next run rather than left.
 #
-# gitea is reached in-cluster at gitea.apps.svc:3000, woodpecker at
+# gitea is reached in-cluster at gitea.gitea.svc:3000, woodpecker at
 # woodpecker.woodpecker.svc:8000 (it moved out of `apps` in the namespace
 # migration — see kubernetes/apps/woodpecker/MIGRATION-namespace.md). Needs
 # the netpol halves: gitea-allow-review-bot-rotator + woodpecker-allow-rotators
@@ -99,7 +99,7 @@ spec:
                - name: BAO_PATH
                  value: homelab/infrastructure/review-bot
                - name: GITEA_URL
-                  value: http://gitea.apps.svc.cluster.local:3000
+                  value: http://gitea.gitea.svc.cluster.local:3000
                # Private repo review-bot collaborates on: the pull check is a
                # real access verification (git-upload-pack on a PRIVATE repo
                # needs a token with access), not the "any valid token" weakness
@@ -225,15 +225,8 @@ metadata:
  labels:
    app.kubernetes.io/part-of: homelab-backup
 spec:
-  # Daily at 07:15 local (next free stagger slot after home-assistant-daily
-  # at 06:45). Covers every stateful workload in the shared `apps`
-  # namespace — currently Gitea (data-gitea-0, the Flux GitOps source of
-  # truth) and Woodpecker (data-woodpecker-server-0), plus gatus-data.
-  # Added 2026-07-14 (issue #186): the apps namespace had no Velero
-  # schedule at all, so the in-cluster Gitea/Woodpecker PVCs migrated from
-  # the legacy LXCs (see MIGRATION-gitea-woodpecker.md) were unprotected.
-  # No CNPG cluster runs in this namespace (both Gitea and Woodpecker use
-  # SQLite), so no labelSelector exclusion is needed.
+  # Daily at 07:15 local. Every PVC in the shared `apps` namespace (gatus-data).
+  # No CNPG cluster runs here, so no labelSelector exclusion is needed.
  schedule: "15 7 * * *"
  useOwnerReferencesInBackup: false
  template:
@@ -247,6 +240,26 @@ spec:
 ---
 apiVersion: velero.io/v1
 kind: Schedule
+metadata:
+  name: gitea-daily
+  namespace: velero
+  labels:
+    app.kubernetes.io/part-of: homelab-backup
+spec:
+  # Daily at 07:45 local. data-gitea-0 (SQLite, repos, LFS, packages).
+  schedule: "45 7 * * *"
+  useOwnerReferencesInBackup: false
+  template:
+    includedNamespaces:
+      - gitea
+    includedResources:
+      - persistentvolumeclaims
+    storageLocation: default
+    ttl: 720h0m0s
+    snapshotMoveData: true
+---
+apiVersion: velero.io/v1
+kind: Schedule
 metadata:
  name: trmnl-daily
  namespace: velero
@@ -21,17 +21,29 @@ spec:
    modules:
      - name: pg_autoscaler
        enabled: true
-  # Rotates daemon keys to aes256k for CVE-2025-30156 (#527). Opt-in; disabled
-  # by default. Increment keyGeneration to rotate again.
-  #
-  # Daemon keys only — aes256k over the CSI mount path needs kernel 7.0+ and
-  # these nodes run 6.18, so CSI stays on aes and the client-key checks stay
-  # raised. keyType is unset so Rook picks it.
+  # CephX keys on aes256k for CVE-2025-30156 (#527). Increment keyGeneration
+  # to rotate again. CSI kernel mounts need aes256k in libceph: Talos
+  # >= v1.13.10 backports it to 6.18 (siderolabs/pkgs 2da2cdc).
  security:
    cephx:
+      # auth_allowed_ciphers. Every key must be aes256k first, or the cluster
+      # locks itself out; recovery is in Rook's cephx-key-rotation doc.
+      allowedCiphers:
+        - aes256k
      daemon:
        keyRotationPolicy: KeyGeneration
        keyGeneration: 2
+      csi:
+        keyRotationPolicy: KeyGeneration
+        keyGeneration: 2
+        keyType: aes256k
+        # 0 deletes prior keys. Set 1 before the next rotation so PVCs
+        # mounted with the old key keep working until they are remounted.
+        keepPriorKeyCountMax: 0
+      rbdMirrorPeer:
+        keyRotationPolicy: KeyGeneration
+        keyGeneration: 2
+        keyType: aes256k
  # CephFS cloner pool raised from the default 4. Velero clones a
  # VolumeSnapshot per PVC, and 4 threads stall a multi-PVC backup with
  # `rados: "all cloner threads are busy"`.
@@ -96,7 +96,7 @@ spec:
        # that traffic is always a pod IP — collapsing all external clients
        # to one source. The NetBird reverse-proxy services are TLS
        # passthrough and carry the WAN client address in a PROXY protocol
-        # v2 header (options.proxy_protocol, set by netbird-proxy-services);
+        # v2 header (options.proxy_protocol, set by netbird-service-sync);
        # X-Forwarded-For stays trusted for HTTP-mode services. Both are
        # accepted from the pod CIDRs only; connections without a header
        # (direct LAN/mesh) keep their RemoteAddr.
@@ -24,7 +24,7 @@ machine:
  # The Kubernetes version is a separate decision, moved by
  # `talosctl upgrade-k8s`.
  install:
-    image: factory.talos.dev/metal-installer/0f6fb8bd8bb53013e9743b7c7b099d96fb9c32fd2464ced32693c2dc0ea5feb9:v1.13.9
+    image: factory.talos.dev/metal-installer/0f6fb8bd8bb53013e9743b7c7b099d96fb9c32fd2464ced32693c2dc0ea5feb9:v1.13.10
  features:
    rbac: true
    # stableHostname is NOT here: it is the v1alpha1 spelling of what the
@@ -57,7 +57,19 @@ machine:
    extraConfig:
      imageGCHighThresholdPercent: 70
      imageGCLowThresholdPercent: 60
+      # Default 110 leaves no N-1 headroom: with ~207 pods on three nodes, a
+      # rolling drain puts the survivors at 105/110 and KubeletTooManyPods
+      # fires (#937). The ceiling is IP space, not the kubelet: Cilium runs
+      # ipam.mode=kubernetes on the node's /24 podCIDR (~253 usable after
+      # cilium_host), so 200 keeps a margin for pods still terminating.
+      maxPods: 200
 cluster:
+  # etcd peers on node0, VLAN 46 (#576). Pinned: Talos never updates an
+  # existing member's peer URLs, so changing this subnet means leave + rejoin
+  # per member (talosctl etcd leave, reboot).
+  etcd:
+    advertisedSubnets:
+      - 10.2.46.0/24
  # controlPlane.endpoint is NOT here: `talosctl gen config` takes it as its
  # second argument, and bootstrap.sh already passes https://k8s.renner.uno:6443
  # there. Setting it again in this patch restated the same value in a second
@@ -17,6 +17,8 @@ machine:
    # pod0 is a closed L2 fabric between the three nodes: statically
    # addressed, no gateway, nothing to learn from an RA.
    net.ipv6.conf.pod0.accept_ra: "0"
+    # node0 is closed the same way.
+    net.ipv6.conf.node0.accept_ra: "0"
    # ingress0 keeps RAs, at the default accept_ra of 2: an IPv6 default
    # router is a link-local address learned only from the RA, and with
    # net.ipv6.conf.all.forwarding=1 — which Cilium sets — Linux ignores RAs
@@ -122,8 +124,8 @@ kind: LinkConfig
 name: uplink
 up: true
 ---
-# Management, VLAN 40. Carries the node IP, etcd, the API and Ceph's public
-# network.
+# Management, VLAN 40: world↔node. The API VIP, admin access to the nodes, and
+# node egress (DNS, NTP, image pulls). Node↔node traffic is on node0 (#576).
 apiVersion: v1alpha1
 kind: VLANConfig
 name: mgmt0
@@ -177,6 +179,19 @@ addresses:
  - address: 10.2.48.11/24
  - address: fd02:48::11/64
 ---
+# Node network, VLAN 46: traffic among the three nodes themselves (#576) — the
+# node IP, etcd, apiserver↔kubelet. Closed like pod0: statically addressed, no
+# gateway, never routed via the UDM.
+apiVersion: v1alpha1
+kind: VLANConfig
+name: node0
+vlanID: 46
+parent: uplink
+up: true
+addresses:
+  - address: fd02:46::11/64
+  - address: 10.2.46.11/24
+---
 # Ingress network, VLAN 47. World-to-pod: the BGP session, the next hop for
 # every advertised LoadBalancer VIP, and the masquerade source for pod egress.
 # Publicly routed, which is what separates it from pod0.
@@ -16,6 +16,7 @@ machine:
    # ingress0 keeps RAs: it learns its IPv6 router and its address from them,
    # and only the default route is held back — see k8s-1.yaml.
    net.ipv6.conf.pod0.accept_ra: "0"
+    net.ipv6.conf.node0.accept_ra: "0"
    net.ipv6.conf.ingress0.accept_ra_defrtr: "0"
    # Answer ARP only on the link that owns the address — see k8s-1.yaml (#793).
    net.ipv4.conf.all.arp_ignore: "1"
@@ -82,7 +83,7 @@ kind: LinkConfig
 name: uplink
 up: true
 ---
-# Management, VLAN 40. Node IP, etcd, the API and Ceph's public network.
+# Management, VLAN 40: world↔node — see k8s-1.yaml.
 apiVersion: v1alpha1
 kind: VLANConfig
 name: mgmt0
@@ -122,6 +123,17 @@ addresses:
  - address: 10.2.48.12/24
  - address: fd02:48::12/64
 ---
+# Node network, VLAN 46 — see k8s-1.yaml.
+apiVersion: v1alpha1
+kind: VLANConfig
+name: node0
+vlanID: 46
+parent: uplink
+up: true
+addresses:
+  - address: fd02:46::12/64
+  - address: 10.2.46.12/24
+---
 # Ingress network, VLAN 47. World-to-pod, publicly routed, statically
 # addressed — see k8s-1.yaml for why it is not leased.
 apiVersion: v1alpha1
@@ -12,6 +12,7 @@ machine:
    # ingress0 keeps RAs: it learns its IPv6 router and its address from them,
    # and only the default route is held back — see k8s-1.yaml.
    net.ipv6.conf.pod0.accept_ra: "0"
+    net.ipv6.conf.node0.accept_ra: "0"
    net.ipv6.conf.ingress0.accept_ra_defrtr: "0"
    # Answer ARP only on the link that owns the address — see k8s-1.yaml (#793).
    net.ipv4.conf.all.arp_ignore: "1"
@@ -109,6 +110,17 @@ addresses:
  - address: 10.2.48.13/24
  - address: fd02:48::13/64
 ---
+# Node network, VLAN 46 — see k8s-1.yaml.
+apiVersion: v1alpha1
+kind: VLANConfig
+name: node0
+vlanID: 46
+parent: uplink
+up: true
+addresses:
+  - address: fd02:46::13/64
+  - address: 10.2.46.13/24
+---
 # Ingress network, VLAN 47. World-to-pod, publicly routed, statically
 # addressed — see k8s-1.yaml for why it is not leased.
 apiVersion: v1alpha1
@@ -7,19 +7,15 @@
 # part of the config that could not be true anywhere but here.
 machine:
  kubelet:
-    # Constrain kubelet's auto-picked node IP to the mgmt LAN. Without this,
-    # kubelet picks the first non-loopback IP, which could land on the v30iot
-    # NIC (ens19, 10.2.30.0/24). Control-plane traffic and hostNetwork pod IPs
-    # must stay on mgmt LAN.
-    # fd40::/64 is the mgmt LAN ULA, and it is listed so the Node object carries
-    # an IPv6 InternalIP at all. Without one, Cilium falls back to detecting the
-    # node's IPv6 from mgmt0 and picks the SLAAC address from the dynamic
-    # provider prefix, which it then probes until the agent restarts (#789).
-    # The GUA is not a control-plane address: BGP peers on fd40::/64 too.
+    # The node IP lives on node0, VLAN 46: the closed node↔node network (#576).
+    # Both families are listed so the Node object carries an IPv6 InternalIP at
+    # all; without one Cilium detects the node's IPv6 itself and picks a SLAAC
+    # address from the dynamic provider prefix (#789). Talos picks one address
+    # per family, IPv4 first, whatever the order here.
    nodeIP:
      validSubnets:
-        - 10.2.40.0/24
-        - fd40::/64
+        - fd02:46::/64
+        - 10.2.46.0/24
  # The UDM on the ingress segment. Published here rather than as a route
  # because bird owns table 47 and originates the default from it; this is the
  # one place the address is stated.
@@ -0,0 +1,30 @@
+#!/bin/sh
+# Turn the TypeSafe picker's probability distribution into one model id.
+#
+# The judgment (how well each catalog model fits the task) comes from the
+# model; the policy lives here, in code: among the candidates whose probability
+# is within MARGIN of the best one, take the most recently released. Several
+# same-family models (Sonnet 4.5 / 4.6 / 5) split the probability for a
+# mid-weight task, and the newest of a comparable set is the one to run.
+#
+# Usage: choose-model.sh <catalog-table-file> < <systemone-response.json>
+#   The table is fetch-models.sh --table output: id \t name \t released=YYYY-MM-DD ...
+#   Prints the chosen id, or nothing when the answer is NONE / unusable.
+# Env: CHOOSE_MODEL_MARGIN (default 0.7): candidate if p >= MARGIN * p_best.
+
+set -eu
+TABLE_FILE="${1:?usage: choose-model.sh <catalog-table-file> < response.json}"
+MARGIN="${CHOOSE_MODEL_MARGIN:-0.7}"
+
+jq -r --rawfile table "$TABLE_FILE" --argjson margin "$MARGIN" '
+  ($table | split("\n") | map(select(length > 0) | split("\t")
+     | {key: .[0], value: ((.[2:] | map(select(startswith("released="))) | .[0] // "released=") | ltrimstr("released="))})
+   | from_entries) as $released
+  | (.answers.model.probabilities // {}) | to_entries
+  | map(select(.key != "NONE")) | sort_by(-.value)
+  | if length == 0 then empty
+    else (.[0].value) as $best
+      | map(select(.value >= $margin * $best))
+      | sort_by([($released[.key] // ""), .value])
+      | .[-1].key
+    end' 2>&1
@@ -51,11 +51,18 @@ if [ -n "${TYPESAFE_API_KEY:-}" ]; then
  if [ -n "${PICK_MODEL_TYPESAFE_DRYRUN:-}" ]; then printf '%s\n' "$REQ"; exit 0; fi
  RESP="$(printf '%s' "$REQ" | curl -sS --max-time 30 -X POST https://api.typesafe.ai/v1/systemone \
           -H "Authorization: Bearer $TYPESAFE_API_KEY" -H 'Content-Type: application/json' --data-binary @- || true)"
-  CHOICE="$(printf '%s' "$RESP" | jq -r '.answers.model.choice // empty' 2>&1 | head -1)"
-  CONF="$(printf '%s' "$RESP" | jq -r '.answers.model.confidence // empty' 2>&1 | head -1)"
+  # Policy in code (choose-model.sh): among candidates within a margin of the
+  # best probability, the most recently released wins. Same-family models split
+  # the probability for mid-weight tasks (live catalog: Sonnet 4.5 chosen at
+  # 0.23 over Sonnet 4.6).
+  TABLE_FILE="$(mktemp)"
+  printf '%s\n' "$TABLE" > "$TABLE_FILE"
+  CHOICE="$(printf '%s' "$RESP" | sh "$HERE/choose-model.sh" "$TABLE_FILE" | head -1)"
+  rm -f "$TABLE_FILE"
+  TOP="$(printf '%s' "$RESP" | jq -r '(.answers.model.probabilities // {}) | to_entries | sort_by(-.value) | .[0:4] | map("\(.key)=\(.value)") | join(" ")' 2>&1 | head -1)"
  VALID="$(printf '%s\n' "$TABLE" | sh "$HERE/valid-model.sh" "$CHOICE")"
  if [ -n "$VALID" ]; then
-    log "typesafe chose '$VALID' (confidence ${CONF:-?}) for: $GUIDANCE"
+    log "typesafe chose '$VALID' (top: ${TOP:-?}) for: $GUIDANCE"
    printf '%s\n' "$VALID"
    exit 0
  fi
@@ -97,9 +97,27 @@ else
  printf 'FAIL  typesafe request: keys [%s] task [%s]\n' "$keys" "$task"; fails=$((fails + 1))
 fi

+# choose-model.sh: newest among comparable candidates, best otherwise, NONE -> nothing.
+TB="$(mktemp)"
+printf '%s\n' 'claude-sonnet-4-5	Claude Sonnet 4.5	released=2025-09-29	context=200000' \
+              'claude-sonnet-4-6	Claude Sonnet 4.6	released=2026-02-17	context=200000' \
+              'claude-opus-5	Claude Opus 5	released=2026-07-24	context=1000000' \
+              'claude-haiku-4-5	Claude Haiku 4.5	released=2025-10-15	context=200000' > "$TB"
+choose() { printf '%s' "$1" | sh "$HERE/choose-model.sh" "$TB" || true; }
+got="$(choose '{"answers":{"model":{"probabilities":{"claude-sonnet-4-5":0.30,"claude-sonnet-4-6":0.25,"claude-opus-5":0.10,"NONE":0.05}}}}')"
+if [ "$got" = "claude-sonnet-4-6" ]; then printf 'ok    newer sibling within margin wins\n'; else printf 'FAIL  comparable: got [%s]\n' "$got"; fails=$((fails + 1)); fi
+got="$(choose '{"answers":{"model":{"probabilities":{"claude-opus-5":0.90,"claude-haiku-4-5":0.05,"claude-sonnet-4-6":0.04,"NONE":0.01}}}}')"
+if [ "$got" = "claude-opus-5" ]; then printf 'ok    clear best is kept\n'; else printf 'FAIL  clear best: got [%s]\n' "$got"; fails=$((fails + 1)); fi
+got="$(choose '{"answers":{"model":{"probabilities":{"claude-haiku-4-5":0.60,"claude-sonnet-4-5":0.35,"NONE":0.05}}}}')"
+if [ "$got" = "claude-haiku-4-5" ]; then printf 'ok    older best outside margin is kept\n'; else printf 'FAIL  margin: got [%s]\n' "$got"; fails=$((fails + 1)); fi
+got="$(choose '{"answers":{"model":{"probabilities":{"NONE":1.0}}}}')"
+if [ -z "$got" ]; then printf 'ok    NONE yields nothing\n'; else printf 'FAIL  NONE: got [%s]\n' "$got"; fails=$((fails + 1)); fi
+rm -f "$TB"
+
 if [ "$fails" -eq 0 ]; then
  echo "All model-selection tests passed."
 else
  echo "$fails test(s) FAILED."
  exit 1
 fi
+