Skip to content

release: independently verify all 14 published MCP Server notes #540

Open

Description

@yeongseon

Parent: #529

Goal

Provide an independent final 14/14 GitHub Release publication/content audit after the historical batch rewrites and verified human-credit updates. This is validation, not a duplicate editor of release sections.

Baseline

  • Record the live GitHub Releases snapshot as of 2026-11-08: 14 published, zero drafts in the returned set, from v0.1.0 through v0.8.0. Confirm full pagination and report if releases are added later (for example the currently unmerged v0.8.1 candidate PR chore(main): release 0.8.1 #526). Do not pretend new versions were part of the baseline.
  • Preserve each Release ID/tag, original body digest, publication timestamp, target, draft/prerelease/latest indication, and asset inventory.
  • Establish exact approved version-by-version canonical Markdown from this repository at a pinned reviewed main SHA, and the expected body digest.

Final verification

  • Verify each version-specific issue completion, independent review evidence, correct original tag ancestry, release support/security claims and contributor audit results. Keep any blocked publication explicitly open.
  • Fetch all 14 public bodies fresh. Compare each body byte-for-byte to approved source and record tag, expected SHA-256, fetched SHA-256, match result, URL, timestamp and reviewer evidence.
  • Confirm unchanged release identity, dates, draft/prerelease/latest flags, assets and the absence of inadvertent tag or PyPI mutations; distinguish new normal releases from historical body edits.
  • Validate readable Markdown and exact published-version comparisons, no duplicate change bullets, no unverified bots-as-human bylines or false first contributions; preserve substantive MCP permissions/read-only defaults and migration limitations.
  • Post a final 14-row results table and explicit unresolved exceptions to parent docs: curate all 14 historical MCP Server release descriptions #529. Source-only merge is not publication; all 14 public release pages must be actually verified.

Untagged and skipped version gate

Dependencies

Baseline/read-only audit can proceed in parallel. Final acceptance follows independently reviewed version-batch publication and contributor reconciliation. Avoid adding permanently scheduled release-repair CI solely for a one-off historical rewrite.

Activity

  1. yeongseon commented on Nov 18, 2026

    @yeongseon
    ContributorAuthor

    Fresh read-only baseline (2026-11-18)

    Read all 14 published GitHub Releases against main CHANGELOG.md with the repository's scripts/extract_release_notes.py body/compare-link rules. 9/14 remote bodies matched canonical source; 5 were different: v0.1.0, v0.2.1, v0.3.0, v0.3.1, and v0.4.0. The v0.3.0 public body is missing a reader summary; v0.5.0 is 52,109 characters and deserves its dedicated #536 review. This is only source equality, not a final proof of factual accuracy, upgrade completeness or attribution.

    PR #568 proposes correcting all five early source sections. After that PR's independent review and merge, re-run the exact 14-row canonical comparison against updated source, get separately authorized/public-body-only publication for mismatched tags, and re-read all 14 public Releases including unchanged metadata and asset digests. Do not close #540/#529 until there is both publication and human-quality evidence; neither source PR nor CI is a public Release write.

  2. yeongseon commented on Nov 19, 2026

    @yeongseon
    ContributorAuthor

    2026-11-19 read-only 14/14 baseline (not final acceptance)

    Opened PR #569 with a pinned-main audit ledger: 14 published, 0 drafts; 9/14 exact source/public body matches, 5/14 different (v0.1.0, v0.2.1, v0.3.0, v0.3.1, v0.4.0). The file records the actual public-body SHA-256 for every release, original Release IDs, publication times, lengths and existing asset IDs/sizes (10 SBOM assets, 4 Releases without assets). main baseline is a99cda67da71 and proposed source #568 is eb444e825201, still unmerged.

    This verifies body/source equality only. Human contributions, actual tagged behavior, prose adequacy, PyPI artifact provenance, and final approved-body publication remain open separate gates. PR #569 is a docs-only verification artifact, not a body PATCH/authorization or completion of #540. #568 remains pending independent review; its five corrected source sections must be published separately with backup, drift protection and exact readback.

  3. yeongseon commented on Nov 20, 2026

    @yeongseon
    ContributorAuthor

    2026-11-20 fresh release-note audit and publication path

    The organization currently has 18 repositories, 101 published software Releases across 9 repositories, and Showcase 13 published presentation Releases + 2 drafts (fresh GitHub REST listing). I compared every one of the 101 public software bodies with its corresponding current main canonical source (using the repository's existing version extractor for CHANGELOG.md repositories). Outcomes: 85 byte-exact, 11 CRLF-only (SQLAlchemy), 5 substantive differences, all on MCP v0.1.0, v0.2.1, v0.3.0, v0.3.1 and v0.4.0. cubrid-client is now 6/6 exact and historical issue #298 is closed; no longer report its prior two-body drift as outstanding.

    For MCP, the canonical CHANGELOG still differs on five early bodies because the independently reviewed correction in #568 has not merged. The 14-release audit in #569 has passed CI, but also lacks independent approval. Both PRs remain open and have requested @paikend's review. The tagged v0.1.0 server has ten @mcp.tool registrations; this confirms the correctness problem in the historical six-tool canonical source, which #568 fixes. No source-only merge is presented as public synchronization.

    I opened draft PR #581 as a separate manual, disabled-by-default historical body-only publisher proposal, after confirming the existing sync_historical_release_footers.py is only suited to footer edits. The draft adds a per-version dry-run diff/digest, pinned source SHA and authorisation checks, original JSON backup, repeated public-body/metadata drift checks, body-only PATCH and readback tests. It is not activated or merged. Explicit blocker: GitHub Release PATCH does not support atomic CAS, so an independently approved exclusive editing window and credential/security review are needed; a CLI flag alone is not a lock. Publication of the five releases is still blocked.

    Content equality, editorial quality, tagged code / real registry / human contributor provenance remain separate acceptance gates. Examples requiring editorial attention include MCP v0.5.0 (52,109-character published body), pycubrid v1.9.0 and SQLAlchemy v1.11.5; length is a triage signal, not itself grounds to rewrite. No existing GitHub Release, tags, dates, titles, assets, registries or versions were edited. Keep #540 and its publication dependencies open until approved source and actual post-PATCH public readbacks meet the full acceptance criteria.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions