0.7.2 ship + release handover (2026-07-05)

End-of-session state after fixing the i18n CI blocker, deploying 0.7.2, and backfilling the v0.7.1 / v0.7.2 GitHub Releases. Worktree: ~/.treehouse/orrery-311982/1/orrery, branch pwa-i18n-bundling-precache (pushes to origin/main; Orrery uses direct-to-main, no PRs).

Status: everything shipped + green

• origin/main HEAD = c8180380b. Working tree clean.
• Full pipeline green on c8180380b: CI · CodeQL · docker-e2e · Deploy preview · Deploy docs · Auto-publish GH Release — all success.
• GH Pages: live — https://chipi.github.io/orrery/
• VPS prod: deployed (run #28735992312, healthcheck passed) — live at https://prod-podcast.tail6d0ed4.ts.net:8443/ (tailnet-only).

Releases (both live, green, via release.yml)

• v0.7.2 — tag → commit c8180380b; marked Latest. Published by the standard tag-push path. https://github.com/chipi/orrery/releases/tag/v0.7.2
• v0.7.1 — tag → commit 4b82dd113. Published by workflow_dispatch (retroactive path). https://github.com/chipi/orrery/releases/tag/v0.7.1

Release-process fixes made this session

1. CHANGELOG.md — [Unreleased] was actually 0.7.0's content (never renamed on release). Renamed it to ## [0.7.0], added curated ## [0.7.1] + ## [0.7.2] sections. release.yml extracts the release body from these.
2. release.yml — added a workflow_dispatch path (tag + optional target) that reads the section from the CHANGELOG on the default branch, so a retroactive tag on a historical commit publishes through the workflow. Re-publish any release with gh workflow run release.yml -f tag=vX.Y.Z [-f target=<full-40char-sha>]. (Captured in memory feedback_release_workflow.)

Gotchas learned: --target needs the full 40-char SHA (short SHA → HTTP 422); backfilling an older release with --latest steals the Latest marker (reset with gh release edit <newest> --latest).

What the i18n fix was (context)

The 0.7.2 mobile layout hid the translated site-footer on touch (links moved into the nav drawer). That footer was the only reliable visible locale-script text on canvas routes (/explore,/fleet,/fly), so i18n-all-routes.spec.ts's innerText >10 script-char check fell to 5-9 on mobile-chromium. Fix (d7fa45337): count locale-script chars across visible text AND accessible names (aria-label / title / placeholder / alt). Captured in memory feedback_e2e_pitfalls.

Open follow-up (only one, non-blocking)

• Mobile dev-test over Tailscale isn't wired for orrery. This worktree's vite dev port is 5273 (default; no VITE_DEV_PORT in .env.local). The laptop's active tailscale serve (:443 →:8080) belongs to podcast-scraper, not orrery. To test orrery on phone: npm run dev -- --host (0.0.0.0:5273) then tailscale serve --bg --https=8443 5273 → https://markos-macbook-pro-1.tail6d0ed4.ts.net:8443/.
  • UNVERIFIED: vite.config.ts sets no server.allowedHosts; vite 5.x may block the tailnet hostname. If the phone gets "host not allowed", add the FQDN to server.allowedHosts. Not tested.

Optional (not done, out of scope)

• v0.7.0 has one pre-existing failed release.yml run (2026-07-03, before the CHANGELOG had a ## [0.7.0] section). Its release is fine. Now that the section exists, gh workflow run release.yml -f tag=v0.7.0 would refresh its body + go green — but it'd replace the curated title/body. Left untouched.