docs: Valia-sites domain language + ADR-0018 (off-infra Pages, in-cluster sync)

Grill session with Viktor: his mother Valia will keep asking for 1-page
site hosting, so the pattern is being made repeatable. Decisions: all
Valia sites serve off-infra on Cloudflare Pages (survive homelab
outages); one shared in-cluster CronJob mirrors her Drive folders every
10 min and redeploys on change; English subdomain names picked by
Viktor; failed-Job-only visibility; stem95su migrates onto the pattern.
CONTEXT.md gains Valia site / Content folder / Entry file; full
rationale and rejected options in ADR-0018.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

CONTEXT.md

@@ -237,6 +237,20 @@ _Avoid_: expecting Diun to deploy; conflating with **Keel**.
 **Anubis**:
 A PoW reverse-proxy issuing a 30-day JWT cookie, used in front of public content-bearing sites without app-level auth (blog, wiki, landing pages). Never in front of Git, WebDAV, CalDAV, or API endpoints (clients can't solve PoW).
 
+### Externally-authored sites
+
+**Valia site**:
+A small public static site authored by Valia (Viktor's mother, external to the infra) and hosted for her under `<name>.viktorbarzin.me`. Its source of truth is a **Content folder** she owns; the live site is a mirror of that folder, fresh within ~10 minutes. Hosted **off-infra** (Cloudflare Pages) by decision: a homelab outage freezes content but never takes her sites down. Viktor picks the English subdomain name per site at registration (her folder names stay Bulgarian). Current instances: `stem95su`, `bridge`.
+_Avoid_: "school site" (the family may grow beyond school projects); treating the deployed copy as editable — edits land only in the **Content folder**.
+
+**Content folder**:
+The Google Drive folder (or subfolder) Valia shares with `vbarzin@gmail.com` holding one **Valia site**'s files. Strictly read-only from the infra side — nothing ever writes back to her Drive. Empty or half-uploaded folder states must never wipe a live site.
+_Avoid_: syncing a folder root when the servable content lives in a subfolder (stem95su serves `stem claude/files/`, not the folder root).
+
+**Entry file**:
+The HTML file a **Valia site** serves at `/`. Defaults to `index.html`; per-site override when she names it differently (stem95su: `stem_board.html`). The override is a registration-time setting, not a constraint on her authoring.
+_Avoid_: asking Valia to rename her files to fit hosting conventions.
+
 ## Relationships
 
 - A **Service** is defined by exactly one **Stack** — **flat** or wrapping a **Stack-local module** — which sources zero or more shared **Factory modules** and resolves to one or more K8s workloads.
@@ -248,6 +262,7 @@ A PoW reverse-proxy issuing a 30-day JWT cookie, used in front of public content
 - A **Service**'s image reaches the cluster via **Woodpecker deploy** (push-driven, on commit) or **Keel** (poll-driven, on a new registry tag); **Diun** only notifies. Operator-managed StatefulSets are rolled by neither.
 - An owned **Service**'s image is built by GitHub Actions from the **Canonical repo**'s **GitHub mirror** and hosted on ghcr.io (ADR-0002); the **Forgejo registry** keeps only a frozen last-known-good tag per **Service**.
 - Tier-1 **State tier** state and ~12 app databases share one **CNPG** `pg-cluster`, reached through **PgBouncer**; their credentials rotate via the `vault-database` store.
+- A **Valia site** mirrors exactly one **Content folder** and serves exactly one **Entry file** at `/`; the folder is hers, the subdomain name is Viktor's, the hosting is off-infra.
 
 ## Example dialogue
 

docs/adr/0018-valia-sites-off-infra-pages-in-cluster-sync.md

@@ -0,0 +1,47 @@
+# Valia sites are served off-infra (Cloudflare Pages), synced in-cluster
+
+Valia (Viktor's mother) authors small one-page static sites in Google Drive folders she
+shares, and keeps asking for them to be hosted — two exist already (`stem95su`, `bridge`)
+and more are expected. We decided all **Valia sites** are served **off-infra on Cloudflare
+Pages** under `<english-name>.viktorbarzin.me`, kept fresh by **one shared in-cluster
+CronJob** (`stacks/valia-sites/`) that mirrors each **Content folder** every 10 minutes
+(rclone, drive.readonly) and re-deploys only on change (wrangler direct upload). The
+existing in-cluster `stem95su` serving stack (nginx + NFS + ingress + per-site sync)
+migrates onto this and is retired.
+
+Why off-infra serving: these are her sites, shown to teachers/parents — they must survive
+homelab outages (cf. the 2026-06-27 egress incident that took every proxied in-cluster
+site down). With Pages, a homelab outage degrades to "content frozen until we're back",
+never "site down". Serving costs no cluster resources and no per-site nginx/PVC/ingress/
+Anubis. Why the syncer stays in-cluster anyway: secrets stay in Vault (no per-site GHA
+secret sprawl), and the stem95su guard patterns (hard-fail on Drive auth errors, never
+wipe a live site on an empty/partial folder, capped deletes) carry over wholesale. The
+deliberate asymmetry — off-infra serving, on-infra syncing — is the point, not an
+accident.
+
+## Considered options
+
+- **In-cluster everywhere** (generalise stem95su into a factory module): one roof, no
+  Cloudflare Pages dependency — but her sites share the homelab's fate and each site
+  spends cluster resources to serve static files a free CDN serves better.
+- **Pages for new sites only**: less work now, two patterns and two runbooks forever.
+- **GHA-scheduled sync** (fully off-infra pipeline): no cluster dependency at all, but
+  Drive + Cloudflare credentials would live as GitHub secrets per repo, outside Vault.
+
+## Consequences
+
+- Registration is one entry in the `sites` map (name, Content folder, optional Entry
+  file); CI applies Pages project, custom domain, public CNAME, and internal-DNS config
+  together. Names are English, picked by Viktor (most → bridge set the precedent).
+- The internal split-horizon zone learns Valia sites from a ConfigMap the
+  `technitium-ingress-dns-sync` script consumes — declaratively, including **removal**
+  (the previous static-CNAME approach was add-only; a retired site left a stale record).
+- Deploy-on-change is mandatory, not an optimisation: Pages caps monthly deployments on
+  the free tier, and a 10-minute cadence would burn ~4,300/month if unchanged runs
+  deployed.
+- Failure visibility is **failed-Job-only** by explicit choice (no stale-sync alert, no
+  per-site uptime monitors, no notifications to Valia) — Viktor fields "it didn't
+  update" reports, consistent with the alert-noise-reduction posture. Revisit if a
+  silent stall actually bites.
+- If the homelab is down, content updates pause; the sites keep serving last-deployed
+  content. Accepted degradation.