Update CLAUDE.md: correct production server, mail host, deploy checklist

Server IP was stale (pre-migration host, 69.62.106.94) - production has
been on h1 (163.172.67.77) for a while. Documents the MAIL_HOST incident,
the first-time git linking done today, and that PHP-FPM + queue worker
restarts are needed on every deploy, not just composer/config changes.

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

CLAUDE.md

@@ -10,9 +10,31 @@ Libra Law Firm is a bilingual (Arabic/English) web platform for managing client
 
 ## Production Environment
 
-- **Server:** Rocky Linux 9.6 on 69.62.106.94
-- **PHP:** 8.4 (use `php84` command, not `php`)
-- **Database:** MariaDB/MySQL
+- **Server:** h1 (`h1.elenoslabs.com`, `163.172.67.77`) — Rocky Linux 9, ElenosLabs-owned. **Not**
+  `69.62.106.94` — that stale IP is left over from before the migration to h1 and does not serve
+  this app; a past `.env` even had `MAIL_HOST` pointed at it by mistake (see *Mail* below). Full
+  server details: `wiki/prod_servers/elenoslabs/h1server.md` (Jarvis workspace), § libra.adv.ps.
+- **App user:** `libra` — app root `/home/libra/public_html/`, logs `/home/libra/logs/`
+- **PHP:** 8.4 (use `php84` command — there is no bare `php` binary on this box; even Composer
+  needs `php84 /usr/local/bin/composer ...`)
+- **Database:** MariaDB, Redis (sessions/cache/queue)
+- **Queue worker:** `libra-queue.service` (`queue:work redis`) — a **long-running process**, see
+  the queue note below.
+- **Git:** production was file-copied at initial migration and only properly linked to git for
+  the first time on 2026-09-28 (`git checkout -f -B main origin/main` after backing up the prior
+  state) — before that date, `git pull` on production did nothing (empty local branch, no
+  commits). It is now a normal git-tracked checkout; a plain `git pull` works going forward.
+
+### Mail
+
+`MAIL_HOST` must be `mx1.elenoslabs.com` (h1's own Postfix/Dovecot stack, MX for `adv.ps`) — **not**
+`mail.adv.ps`, which resolves to the stale `69.62.106.94` host and times out. This caused a real
+incident (2026-09-28): the workshop-registration form's emails silently never arrived because of
+this wrong value, and a separate real bug in `App\Services\AdminNotificationService` (a Carbon
+`TypeError` on a string `throttle_minutes` config value — fixed, see git history around
+2026-09-28) masked the actual SMTP error in the logs the whole time it was broken. If mail seems
+to silently fail again, check `MAIL_HOST` first, then `storage/logs/laravel.log` for a
+`TransportException`.
 
 ### Production Commands
 All artisan commands on production require `php84` prefix:
@@ -24,16 +46,25 @@ php84 artisan optimize:clear
 
 ### Deployment Checklist
 ```bash
-# On production server
-composer install --no-dev --optimize-autoloader
-php84 artisan migrate --force
+# On production server (as the libra user, in /home/libra/public_html)
+git pull
+php84 /usr/local/bin/composer install --no-dev --optimize-autoloader   # only if composer.json/lock changed
+npm install && npm run build                                           # only if frontend/Tailwind classes changed
 php84 artisan optimize:clear
 php84 artisan config:cache
 php84 artisan route:cache
 php84 artisan view:cache
 
-# IMPORTANT: Always restart PHP-FPM after deployment
+# IMPORTANT: always restart PHP-FPM after ANY deploy, even a content/lang-only change.
+# 2026-09-28 incident: opcache.validate_timestamps is On with revalidate_freq=2, which should
+# have picked up a changed lang/*.php file within 2 seconds on its own — it didn't, and only a
+# full php84-php-fpm restart made the change visible. Root cause not fully diagnosed; until it
+# is, treat the restart as mandatory on every deploy, not just composer/config changes.
 sudo systemctl restart php84-php-fpm
+
+# IMPORTANT: always restart the queue worker too — it's a long-running process that holds old
+# code and old .env/config values in memory until restarted, independent of the php-fpm restart.
+php84 artisan queue:restart
 ```
 
 **Note:** The server has both PHP 8.3 and 8.4. Always use `php84` for artisan commands. Failing to restart PHP-FPM can cause cached config issues.