docs: questions v4 — pairing mechanism fork + gitea allowlist ask [#626][#734]

questions-v3.md

@@ -1,42 +0,0 @@
-# Questions — v3 (current round)
-
-Round v2 was closed out and removed 2026-09-02 by the parallel HA session.
-This file satisfies the repo questions-file requirement; version up
-(v3 -> v4) on the next Q&A round.
-
-## Active question channel (founder-directed)
-
-`~/PTKR.md` — 12 open items: GLPI API key, AI-staff identities,
-Vault-vs-Bitwarden for agent secrets, prometheus/grafana envs, VoIP carrier
-decision, PMG MX cut-over window, SITES access, October hardware window
-dates, Friday 2026-09-04 onsite coordination, #681 Cloudron 500s, nitrokey
-purchase, Cloudron packaging session scope.
-
-## Repo-local questions
-
-### Q1 — HA remote access: proceed with the Cloudron pairing now? (decision needed before Friday)
-
-Adopted architecture (your 2026-09-01 ruling on #626): Cloudron HA = family
-frontend, pfv-bms = property brain, connected via HA Remote
-assistants/entities. Facts verified 2026-09-02 evening:
-
-- Both instances at 2026.8.3 (above the 2025.7 pairing floor), Cloudron HA
-  RUNNING, api token in ~/.creds/cloudron-ha.env, Tailscale path Cloudron ->
-  pfv-bms proven (Kuma already monitors LAN hosts from there).
-- Risk on the table: tsys-cloudron Postgres wobbles the same day (#681 500s).
-  If Cloudron degrades, FAMILY ACCESS degrades — pfv-bms itself is unaffected
-  either way.
-
-Options:
-
-- (A) Proceed with pairing now (recommended) — additive and reversible; pfv-bms
-      stays unexposed (Tailscale-only admin path); family accounts live on
-      Cloudron only. Remaining steps: mint pfv-bms long-lived token (yours, or
-      delegate to agent on approval), wire Cloudron-side integration, pair,
-      add family accounts.
-- (B) Hold until Cloudron stabilizes — nothing lost except timeline; Friday
-      family-access demo unlikely.
-- (C) Nabu Casa interim (~$6.5/mo) — fastest single-instance remote UI, but
-      does not serve the multi-property global-instance goal you ruled on.
-
-Answer inline (A/B/C + any conditions):

questions-v4.md

@@ -0,0 +1,55 @@
+# Questions — v4 (current round)
+
+v3 closed 2026-09-02: Q1 answered in chat ("cloudron postgres solved —
+wire up the remote access, hook first"). The deploy hook LANDED (#734,
+E2E proven); the pairing hit a wrong premise and needs a fork ruling
+below. Version up (v4 -> v5) on the next Q&A round.
+
+## Active question channel (founder-directed)
+
+`~/PTKR.md` — 12 open items: GLPI API key, AI-staff identities,
+Vault-vs-Bitwarden for agent secrets, prometheus/grafana envs, VoIP carrier
+decision, PMG MX cut-over window, SITES access, October hardware window
+dates, Friday 2026-09-04 onsite coordination, #681 Cloudron 500s, nitrokey
+purchase, Cloudron packaging session scope.
+
+## Repo-local questions
+
+### Q1 — HA instance pairing mechanism (blocks #626 family access; would like before Friday)
+
+The 2025.7+ "built-in pairing" premise was WRONG — no such core
+integration exists as of 2026.8.3 (verified on both instances + HA core +
+docs; full research in #626 note). Options:
+
+- **(A)** HACS custom `remote_homeassistant` on the Cloudron HA — full
+  entity mirroring + control, matches the ruled architecture. Needs a
+  founder file/HACS install on the Cloudron app (~30 min of your time;
+  no agent API for Cloudron app files). Caveat: community reports of
+  recent breakage (remote entities unavailable) — I'd vet the GitHub
+  issues before committing.
+- **(B)** One-way API bridge — pfv-bms pushes ~20 key family states
+  (temps, UPS, garage/artroom) to the Cloudron HA REST API + I build
+  family dashboards there via websocket. I can build it entirely
+  tonight; family gets read-only visibility; no control.
+- **(C)** MQTT broker on the VPS, both instances bridge — bidirectional,
+  but a new always-on service + YAML on both sides.
+
+Answer (A/B/C) + does family need to TOGGLE things at the property, or
+just see status?
+
+### Q2 — gitea webhook allowlist (1-line + restart; unlocks instant deploys)
+
+gitea silently drops webhook deliveries to CGNAT-range targets (SSRF
+`ALLOWED_HOST_LIST` default — no delivery attempt, no log line; restart
+doesn't help; that's why hook #10 never fired). Until it opens, the
+5-minute sha-watch net deploys everything reload-only — already live.
+
+To open the instant path: Cloudron file manager → gitea app.ini →
+
+```ini
+[webhook]
+ALLOWED_HOST_LIST = external,private
+```
+
+then restart gitea. Say the word when done — I'll verify hook #10 fires
+end-to-end (it carries the full restart-mapping logic).