Skip to content

RELEASE-CHECKLIST.md

Latest commit

 

History

History
68 lines (54 loc) · 3.32 KB

RELEASE-CHECKLIST.md

File metadata and controls

68 lines (54 loc) · 3.32 KB
Raw

v0.3 "Executor Release" — release checklist

Status update 2026-07-18: v0.3.0 was tagged and released AHEAD of the two human gates, by deliberate owner decision — pre-launch, zero traffic, and live end-to-end testing of the full pipeline (repo → tag → site download) is easier against the real release. The release notes state plainly which verification is machine-proven and which is pending. The two gates below are unchanged in substance and remain open as post-release validation — they now block the "safe to recommend to real families" claim and v0.3.1, not the tag. Findings fold into v0.3.1.

Gate 1 — the Windows dry run (THE release gate)

Someone who is not Jamie, on a real Windows 10/11 machine, recovers the example register from the printed instructions alone.

Ready-to-run kit:

  1. Prepare on any machine with the repo: cp examples/estate.example.yaml estate.yaml && scripts/setup.sh — hand-copy the three shares onto paper as instructed (this is part of the drill), and copy estate.yaml.age onto a USB stick. Delete the local estate.yaml* afterwards.
  2. Print docs/WINDOWS-RECOVERY.md and the filled guide from scripts/make-guide.sh (pages one and two).
  3. Hand the tester: the USB stick, two of the three paper shares, the two printed documents. Nothing else — no verbal help. Watch silently and take notes on every hesitation, retype, and wrong turn.
  4. Success = they are reading estate.yaml in Notepad unaided.
  5. Fold every stumble back into docs/WINDOWS-RECOVERY.md / templates/EXECUTOR-INSTRUCTIONS.md, commit, and if the changes were substantive, run the dry run again with a fresh tester.
  • Windows dry run passed by a non-author on a real machine
  • Observed failures folded back into the docs

Gate 2 — the physical fire drill

With the real register (not the example): run scripts/test-recovery.sh using two genuinely held, physically printed shares — at least one fetched back from its actual holder. The script records the date and tester in recovery-tests.log; re-run scripts/make-guide.sh and re-print so the guide carries the "Last successful recovery test" line.

  • Fire drill passed with real printed shares
  • Date + tester recorded and visible on the re-printed guide

Tagged (2026-07-18)

v0.3.0 is live: https://github.com/TheWayWithin/executor-file/releases/tag/v0.3.0 — tarball + .sha256 attached, release notes honest about the pending gates. When both gates pass: fold findings into the docs, release v0.3.1 with "Windows 10/11 (manual dry run — date)" added to platforms tested, bump the tarball filename in README's "Get the tool" section, and update the site.

Already verified (by CI and scripted tests on main)

  • Threshold defect confirmed empirically; -t/-n removed; regression tests refuse them loudly
  • All P0 safety/honesty fixes merged; grep sweep in the test suite keeps the corrected claims gone
  • render.sh produces the triage report from both example registers; dependency lines verified
  • Share rotation executed end-to-end in the suite; old shares proven dead against the rotated file
  • Schema v3 validates on both tiers; CI drift test green (and proven against planted defects); v2→v3 path documented and tested
  • SECURITY.md, CONTRIBUTING.md, issue templates in place