carlos / platform Public

Migrated from github.com/carlosframework/platform issue #319, opened 2026-08-27. GitHub issues are closed for this repo; see AGENTS.md, "Work is tracked on amadan, not GitHub".

Still outstanding: The dev deployment's legacy layout and hand-made units make it a poor failover rehearsal target.

---

Hit during the rung-1 rehearsal (2026-08-27, teamtito/carlos-infrastructure#7).

• The dev bucket (carlos-dev-640012953448) was still on the pre-accounts key layout (apps/helloworld/…, no apps/bab/). A fresh box's adopt pass finds ptr == nil for every route and stages nothing, silently (only the canary-key "channel has no release yet" advisory appears). sophie serves only because she staged before the layout change. I ran carlos accounts migrate -app {helloworld,console,website} -from legacy -to bab on the standby (copies, legacy keys left in place); the console migrate-app half was not run, and sophie's own registry rows still carry account="".
• helloworld and the dev console are 2026-08-02 hand-written systemd units (helloworld.service, console.service), not carlos deploy instances. The platform re-stages their releases on a restored box but nothing starts them.

Ask: redeploy dev's apps as ordinary carlos deploy apps (console per #295's direction) and drop the hand units, so the next rehearsal measures the platform's own re-materialisation rather than dev's history. Also bootstrap-box.sh regenerates none of CARLOS_SECRETS_KEY/CARLOS_TUNNEL_*/CARLOS_REGION/CARLOS_ACME_DELEGATION_ZONE — worth an SSM-backed host.env source so a restore is not hostage to an off-box copy.