Files
OmniRoute/docs/ops
Diego Rodrigues de Sa e Souza 1d5128f55f ci(release): preview-artifact build-once workflow, LTS cut dry-run script and GO/NO-GO checklist (rail 3.8.58–.59) (#16210)
* ci(release): preview-artifact build-once workflow with artifact identity (#8084)

Manual (workflow_dispatch pr_number) or `preview-artifact` label lane that
builds the PR head once (npm ci + build:release), validates that build with
check:pack-artifact and check:pack-boot (fake secrets, ephemeral DATA_DIR),
proves the uploaded tarball is byte-identical to a re-pack of the validated
tree, records its identity (head/base SHA, lockfile hash, platform/arch/node
ABI, bundler, build policy) and uploads tarball + identity + SHA256SUMS.
A separate GitHub-hosted `attest` job, the only one holding id-token/
attestations write, verifies the checksums and attests the tarball. Fork PRs
are refused on both trigger paths; nothing publishes.

Refs #8084

* feat(release): dry-run script for the 3.9.0 LTS cut (rail 3.8.58)

`npm run release:dry-run-lts-cut` prints and validates the cut without
executing anything: preconditions (source resolves, previous tag v3.8.59,
package.json = 3.9.0, release-freeze open, release branch green per the
"Release branch not green" issue, stable/v3 + develop absent), stable/v3 from
the tip, develop with the 4.0.0 bump (package.json x3, lockfile, openapi),
which dormant-workflow triggers/`if:` turn true (and which stay gated by a
repository variable or are pinned to the canonical repo), expected dist-tags
(latest -> 3.9.0, next/nightly empty; uses scripts/release/dist-tag.mjs when
present, a minimal fallback otherwise) and the rollback.

`--execute --remote <fork>` runs the two branch steps (or `--rollback`) only
against a fork: origin and any remote whose URL is the canonical repository
are refused, every step is confirmed on the terminal, the develop bump commit
is built with git plumbing (no working tree touched) and every command is an
execFile argv array. Rehearsal parameters (`--target-version`,
`--previous-tag`, `--advisory`) let 3.8.58 run it on the current tip.

* feat(release): print the stable/v3 and develop SHAs after an LTS-cut rehearsal

* docs(release): 3.9.0 LTS cut runbook and GO/NO-GO checklist (rail 3.8.58-.59)

RELEASE_CHECKLIST gains "3.9.0 LTS cut (rehearsed in 3.8.58)": the read-only
dry-run (`npm run release:dry-run-lts-cut`), the fork-only rehearsal and its
rollback, the PR preview-artifact lane (download, checksum, attestation
verify, install the same bytes) and the ordered real cut.

New docs/ops/LTS_GO_NO_GO.md: rail gates G0/G1/G2/G11/G12/G13/G14 and the
10-item closing battery, each with the command that proves it and an empty
evidence column for the owner, ending in the GO / NO-GO line. Linked from the
ROADMAP (Phase 3), the docs index and the ops sidebar.

* fix(release): model G11 as the Mergify queue and report base-green as unknown for missing branches

G11 is the Mergify `release` queue (GitHub's native merge queue was rejected
in v3.8.49: no branch wildcards on a personal-account repo), not a ruleset
`merge_queue` rule. The dry-run now reads `.mergify.yml` at the cut source
(queue_rules[release] targeting release/*, entered via the `queue` label, with
a checks_timeout) and checks the `queue` label exists; the release/* ruleset
is checked separately for deletion + non_fast_forward only.

base-green no longer reads "green" when the release branch does not exist on
origin: no open base-red issue is not evidence for a branch the observer
cannot see, so it reports `unknown ... (not verifiable)` and fails closed.
2026-10-10 10:09:19 -03:00
..
2026-07-29 15:18:55 -03:00
2026-06-29 08:40:06 -03:00
2026-07-29 15:18:55 -03:00