Skip to content

Latest commit

 

History

History
184 lines (149 loc) · 12.1 KB

release-checklist.md

File metadata and controls

184 lines (149 loc) · 12.1 KB

Release checklist

1.0.0 is published on npm under latest. Release checklist. Verified registry bytes and clean installations; beta remains 0.1.0-beta.2. Source d43e235 passed 10 × 188 = 1,880 sequential checks and 188 matching four-worker checks, with zero retries/skips and the unchanged 100 ms bound. All eleven raw reports were independently validated: 1,771 unique journey IDs, maximum passing observation gap 47.7 ms. Raw-report hashes and validation. Historical 0.1.0/alpha/beta results below retain their original identities.

Historical non-prerelease 0.1.0

PR #23 is merged. The exact qualified archive was published on GitHub and npm under latest before 1.0.0; beta remains 0.1.0-beta.2. Registry and clean-consumer verification is complete. The stable policy retains ten complete final rounds, a separate four-worker comparison, zero retries/skips and the 100 ms observation limit. The new inputs completed their own qualification; historical beta.2 records below do not certify 0.1.0. Source hashes, benchmark, guard and the release archive gate passed.

Publish a normal GitHub release only after qualification. npm publication is a separate step: preserve the exact archive and checksum, and keep npm publication pending until registry readback and a clean install pass. A stable npm publication uses latest, leaving beta unchanged. See the compatibility and support policy.

Historical alpha/beta qualification

Beta.2 qualification. Source fd8f58a completed ten rounds of 180 tests and the separate 180-test four-worker comparison. All eleven raw reports matched, with 1,771 distinct journey IDs and zero retries. Beta.2 release notes · Independent validation.

For beta.2 onward, require ten complete final rounds on one frozen candidate, then the matching four-worker comparison. Validate the raw reports, keep the 100 ms gap bound and zero retries, and verify the exact archive separately. Development rounds do not count toward the ten.

The history below records the 0.1.0-alpha.6 handoff. For the first beta candidate, see the beta scope and release checks.

Frozen candidate (31 October 2026): 1187fde. The candidate/0.1.0-alpha.6 tag remains at this commit. Candidate CI passed three separate macOS 15 ARM allocations and Linux: 172 tests per job, one worker, zero retries/skips, and both real package consumers in every job. ARM maximum passing-window gaps were 87.7 / 65.6 / 50.7 ms; Linux was 23.9 ms. The 100 ms limit is unchanged. Verified CI record. The benchmark was measured at f9c492b: 108/108, with input hashes matching this candidate. Earlier failures and measurements keep their source identities. Short checks do not guarantee shared-runner latency. Final source-bound reliability completed on 31 October 2026 UTC (1 November in Kyiv): 100 × 172 sequential executions + 172 four-worker checks, zero retries/skips. Independent validation. The release archive gate, owner sign-off and publication authorization remain separate.

Final reliability verified: The frozen 1187fde candidate completed 100 × 172 = 17,200 sequential test executions and 172 matching four-worker checks, with a 100 ms limit and zero retries/skips. All 101 raw reports were independently validated. Development still defaults to four rounds. See the protocol and retained results. Ordinary CI checks the real package and clean consumers without claiming final acceptance.

Candidate: 0.1.0-alpha.6, ruleset 0.4.0, schema 1. The earlier alpha numbers were source checkpoints, not published packages. This section preserves the alpha handoff before the later beta.1 publication.

Frozen candidate snapshot (1187fde)

The handoff guide and readiness record identify the preserved main-CI archive and the fresh benchmark. Linux passed 172/172 checks and three separate macOS ARM machines passed 516/516, with real package consumers in every job. The benchmark passed 108/108; both overhead calculations meet the 20% target for every workload. The interactive candidate review records keyboard and report checks at 1187fde; it does not imply owner sign-off. The final source-bound reliability series has now passed independently.

This is a frozen candidate with verified final reliability. releaseReady remains false; the preserved development archive still has finalAcceptance: false. Do not repeat sufficient development rounds for changes to documentation or the release-evidence checker. New archive bytes still require their own package-consumer verification and checksum. The preserved Linux archive has 115 files / 395,737 bytes, SHA-256 42f45c5eac7958d55b3b23b40ea0c484b85c56fdcd01f6b22a716e39b9d26967. These updated instructions postdate that archive; keep the candidate tag fixed. A package rebuilt with these documents must not inherit the old archive checksum.

Inspect remaining evidence without running tests

From the repository root:

node scripts/check-release-readiness.mjs
node scripts/check-release-readiness.mjs --json

This read-only command checks the saved acceptance policy, final repeatability, three benchmark records and CI-guard evidence against the current inputs. It lists all blocked records in one run. It does not build, run browser tests, start a final series, delete existing artifacts or publish anything. Exit 0 means the saved evidence passed validation; exit 1 means evidence is pending or invalid; exit 2 means the command could not inspect the checkout or its options.

An evidence result alone is not release approval: the exact archive still needs the real --release consumer gate, owner review and publication authorization. Raw-report validation is complete and recorded separately. The release gate uses the same evidence validator. Historical results and development package records cannot satisfy final acceptance. For this snapshot, all six saved-evidence checks pass and the command exits 0. The completed final series is recorded, not repeated by this documentation update.

Development checks

Use the pinned environment in compatibility, a clean checkout and the committed lockfiles:

pnpm install --frozen-lockfile --ignore-scripts
pnpm --dir examples/react-portal install --frozen-lockfile --ignore-scripts
pnpm exec playwright install chromium
pnpm check
pnpm verify:reliability # Four rounds plus the separate four-worker comparison

On Linux, add --with-deps to the Chromium install when system libraries are missing. On macOS, caffeinate -is pnpm check prevents sleep during browser runs. Dependency installation and the React consumer lockfile preparation need a registry connection. For each consumer, the package check prepares a lockfile and registry metadata without installing node_modules, then uses a separate empty directory for the frozen offline installation. This preserves native TypeScript and esbuild platform dependencies. Both tarball consumers install offline from the populated store. No account, API key or external service is needed to run their browser tests.

Gate Evidence / command
Public API, types, limits and errors agree Dialog, tabs, form errors, rules/codes, limits; strict TypeScript checks
Correct and defective fixtures 172 browser regression tests; all three scenario families
Failures cannot silently pass CI 52 subprocess guard cases including dependency failure, skip, retry and interrupted runs
Release evidence fails closed node --test scripts/tests/*.test.mjs; stale or incomplete hashes, short series, retries, invalid benchmarks and contradictory guard exits are rejected
Source-bound reliability 1187fde: 100 × 172 sequential checks + 172 four-worker checks, zero retries/skips; proof and protocol
Equivalent native measurements Recorded f9c492b: 108/108 tests, three workloads, 30 measured pairs each; method, samples and runner
Vanilla tarball consumer ESM/CJS imports and declarations; 6 browser tests and 3 guarded journeys
React portal tarball consumer StrictMode, real portal/remount, 3 browser tests, 1 guarded journey, separate failing process
Reporter compatibility HTML and JUnit from passing and failing React runs; original failed attachment and not_run dependency links retained
Package inventory License, notices, schema and exports; no test results, node_modules or local configuration
Archive identity artifacts/package-verification.json and artifacts/SHA256SUMS from the exact tested .tgz
Hosted platform checks Candidate 1187fde: Linux 172 and ARM 516 passed with package consumers; the commit ultimately selected for release needs matching checks
Human review Inspect HTML failure report and try the example with a keyboard; passing automation is not a conformance assessment

pnpm check runs build, types, unit, guard, one browser suite and actual package consumers. pnpm verify:reliability is the separate four-round development check. Neither command claims final acceptance or requires a five-hour series.

Completed reliability and remaining archive gate

The explicitly requested final series for 1187fde completed on 31 October 2026 UTC (1 November in Kyiv): 100 full rounds, the four-worker comparison and independent validation of all 101 raw reports. The 100 ms bound and zero retries are unchanged. The benchmark input hashes still match; no new benchmark or full series is needed for these documentation changes.

Next, identify the release snapshot and run node scripts/check-package.mjs --release. It packs a new archive containing the updated documents and tests its real consumers. Preserve its new checksum and release-level verification record. The existing Linux archive remains development-verified; do not reuse its checksum or upgrade its record. This gate and publication authorization are still pending.

Review the output

CI uploads the candidate .tgz, verification JSON, checksum file and React reports under the platform evidence artifact. All public files remain English. React and its bundler are example-only dependencies, not installed library dependencies.

Read package-verification.json before using the archive. A failed package check removes stale success/checksum records; the presence of an archive alone is not proof of a successful gate. The record binds the archive bytes, package version, platform, verification level and consumer results. Require finalAcceptance: true for the final release gate. Checksums are integrity checks, not signatures.

Publication remains a separate step

  • Keep the reviewed candidate/0.1.0-alpha.6 tag at 1187fde. Any later release snapshot needs its own identity; do not move the existing tag. Verify that the selected release inputs and archive have matching evidence.
  • Check registry name availability, ownership and release authentication at that time.
  • Use the exact tested tarball; rebuilding after verification changes its identity.
  • Publish as an npm prerelease with the alpha dist-tag; do not update latest.
  • Attach that archive, checksum and verification record to the GitHub prerelease.
  • Verify a clean registry installation and the dist-tag, then update the status in README/changelog. Announce tested scope and limitations, not stable compatibility.

No credential, publish token, npm publication or GitHub release is created by the preparation commands. Draft notes are in first-alpha notes.